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"
68 #include "memory-util.h"
71 #include "namespace.h"
72 #include "parse-util.h"
73 #include "path-util.h"
74 #include "process-util.h"
75 #include "rlimit-util.h"
78 #include "seccomp-util.h"
80 #include "securebits-util.h"
81 #include "selinux-util.h"
82 #include "signal-util.h"
83 #include "smack-util.h"
84 #include "socket-util.h"
86 #include "stat-util.h"
87 #include "string-table.h"
88 #include "string-util.h"
90 #include "syslog-util.h"
91 #include "terminal-util.h"
92 #include "umask-util.h"
94 #include "user-util.h"
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
);
1701 path_simplify(x
+ 5, true);
1702 our_env
[n_env
++] = x
;
1706 x
= strappend("LOGNAME=", username
);
1709 our_env
[n_env
++] = x
;
1711 x
= strappend("USER=", username
);
1714 our_env
[n_env
++] = x
;
1718 x
= strappend("SHELL=", shell
);
1722 path_simplify(x
+ 6, true);
1723 our_env
[n_env
++] = x
;
1726 if (!sd_id128_is_null(u
->invocation_id
)) {
1727 if (asprintf(&x
, "INVOCATION_ID=" SD_ID128_FORMAT_STR
, SD_ID128_FORMAT_VAL(u
->invocation_id
)) < 0)
1730 our_env
[n_env
++] = x
;
1733 if (exec_context_needs_term(c
)) {
1734 const char *tty_path
, *term
= NULL
;
1736 tty_path
= exec_context_tty_path(c
);
1738 /* If we are forked off PID 1 and we are supposed to operate on /dev/console, then let's try to inherit
1739 * the $TERM set for PID 1. This is useful for containers so that the $TERM the container manager
1740 * passes to PID 1 ends up all the way in the console login shown. */
1742 if (path_equal(tty_path
, "/dev/console") && getppid() == 1)
1743 term
= getenv("TERM");
1745 term
= default_term_for_tty(tty_path
);
1747 x
= strappend("TERM=", term
);
1750 our_env
[n_env
++] = x
;
1753 if (journal_stream_dev
!= 0 && journal_stream_ino
!= 0) {
1754 if (asprintf(&x
, "JOURNAL_STREAM=" DEV_FMT
":" INO_FMT
, journal_stream_dev
, journal_stream_ino
) < 0)
1757 our_env
[n_env
++] = x
;
1760 for (t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
1761 _cleanup_free_
char *pre
= NULL
, *joined
= NULL
;
1767 if (strv_isempty(c
->directories
[t
].paths
))
1770 n
= exec_directory_env_name_to_string(t
);
1774 pre
= strjoin(p
->prefix
[t
], "/");
1778 joined
= strv_join_prefix(c
->directories
[t
].paths
, ":", pre
);
1782 x
= strjoin(n
, "=", joined
);
1786 our_env
[n_env
++] = x
;
1789 our_env
[n_env
++] = NULL
;
1790 assert(n_env
<= 14 + _EXEC_DIRECTORY_TYPE_MAX
);
1792 *ret
= TAKE_PTR(our_env
);
1797 static int build_pass_environment(const ExecContext
*c
, char ***ret
) {
1798 _cleanup_strv_free_
char **pass_env
= NULL
;
1799 size_t n_env
= 0, n_bufsize
= 0;
1802 STRV_FOREACH(i
, c
->pass_environment
) {
1803 _cleanup_free_
char *x
= NULL
;
1809 x
= strjoin(*i
, "=", v
);
1813 if (!GREEDY_REALLOC(pass_env
, n_bufsize
, n_env
+ 2))
1816 pass_env
[n_env
++] = TAKE_PTR(x
);
1817 pass_env
[n_env
] = NULL
;
1820 *ret
= TAKE_PTR(pass_env
);
1825 static bool exec_needs_mount_namespace(
1826 const ExecContext
*context
,
1827 const ExecParameters
*params
,
1828 const ExecRuntime
*runtime
) {
1833 if (context
->root_image
)
1836 if (!strv_isempty(context
->read_write_paths
) ||
1837 !strv_isempty(context
->read_only_paths
) ||
1838 !strv_isempty(context
->inaccessible_paths
))
1841 if (context
->n_bind_mounts
> 0)
1844 if (context
->n_temporary_filesystems
> 0)
1847 if (!IN_SET(context
->mount_flags
, 0, MS_SHARED
))
1850 if (context
->private_tmp
&& runtime
&& (runtime
->tmp_dir
|| runtime
->var_tmp_dir
))
1853 if (context
->private_devices
||
1854 context
->private_mounts
||
1855 context
->protect_system
!= PROTECT_SYSTEM_NO
||
1856 context
->protect_home
!= PROTECT_HOME_NO
||
1857 context
->protect_kernel_tunables
||
1858 context
->protect_kernel_modules
||
1859 context
->protect_control_groups
)
1862 if (context
->root_directory
) {
1863 ExecDirectoryType t
;
1865 if (context
->mount_apivfs
)
1868 for (t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
1869 if (!params
->prefix
[t
])
1872 if (!strv_isempty(context
->directories
[t
].paths
))
1877 if (context
->dynamic_user
&&
1878 (!strv_isempty(context
->directories
[EXEC_DIRECTORY_STATE
].paths
) ||
1879 !strv_isempty(context
->directories
[EXEC_DIRECTORY_CACHE
].paths
) ||
1880 !strv_isempty(context
->directories
[EXEC_DIRECTORY_LOGS
].paths
)))
1886 static int setup_private_users(uid_t uid
, gid_t gid
) {
1887 _cleanup_free_
char *uid_map
= NULL
, *gid_map
= NULL
;
1888 _cleanup_close_pair_
int errno_pipe
[2] = { -1, -1 };
1889 _cleanup_close_
int unshare_ready_fd
= -1;
1890 _cleanup_(sigkill_waitp
) pid_t pid
= 0;
1895 /* Set up a user namespace and map root to root, the selected UID/GID to itself, and everything else to
1896 * nobody. In order to be able to write this mapping we need CAP_SETUID in the original user namespace, which
1897 * we however lack after opening the user namespace. To work around this we fork() a temporary child process,
1898 * which waits for the parent to create the new user namespace while staying in the original namespace. The
1899 * child then writes the UID mapping, under full privileges. The parent waits for the child to finish and
1900 * continues execution normally. */
1902 if (uid
!= 0 && uid_is_valid(uid
)) {
1903 r
= asprintf(&uid_map
,
1904 "0 0 1\n" /* Map root → root */
1905 UID_FMT
" " UID_FMT
" 1\n", /* Map $UID → $UID */
1910 uid_map
= strdup("0 0 1\n"); /* The case where the above is the same */
1915 if (gid
!= 0 && gid_is_valid(gid
)) {
1916 r
= asprintf(&gid_map
,
1917 "0 0 1\n" /* Map root → root */
1918 GID_FMT
" " GID_FMT
" 1\n", /* Map $GID → $GID */
1923 gid_map
= strdup("0 0 1\n"); /* The case where the above is the same */
1928 /* Create a communication channel so that the parent can tell the child when it finished creating the user
1930 unshare_ready_fd
= eventfd(0, EFD_CLOEXEC
);
1931 if (unshare_ready_fd
< 0)
1934 /* Create a communication channel so that the child can tell the parent a proper error code in case it
1936 if (pipe2(errno_pipe
, O_CLOEXEC
) < 0)
1939 r
= safe_fork("(sd-userns)", FORK_RESET_SIGNALS
|FORK_DEATHSIG
, &pid
);
1943 _cleanup_close_
int fd
= -1;
1947 /* Child process, running in the original user namespace. Let's update the parent's UID/GID map from
1948 * here, after the parent opened its own user namespace. */
1951 errno_pipe
[0] = safe_close(errno_pipe
[0]);
1953 /* Wait until the parent unshared the user namespace */
1954 if (read(unshare_ready_fd
, &c
, sizeof(c
)) < 0) {
1959 /* Disable the setgroups() system call in the child user namespace, for good. */
1960 a
= procfs_file_alloca(ppid
, "setgroups");
1961 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
1963 if (errno
!= ENOENT
) {
1968 /* If the file is missing the kernel is too old, let's continue anyway. */
1970 if (write(fd
, "deny\n", 5) < 0) {
1975 fd
= safe_close(fd
);
1978 /* First write the GID map */
1979 a
= procfs_file_alloca(ppid
, "gid_map");
1980 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
1985 if (write(fd
, gid_map
, strlen(gid_map
)) < 0) {
1989 fd
= safe_close(fd
);
1991 /* The write the UID map */
1992 a
= procfs_file_alloca(ppid
, "uid_map");
1993 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
1998 if (write(fd
, uid_map
, strlen(uid_map
)) < 0) {
2003 _exit(EXIT_SUCCESS
);
2006 (void) write(errno_pipe
[1], &r
, sizeof(r
));
2007 _exit(EXIT_FAILURE
);
2010 errno_pipe
[1] = safe_close(errno_pipe
[1]);
2012 if (unshare(CLONE_NEWUSER
) < 0)
2015 /* Let the child know that the namespace is ready now */
2016 if (write(unshare_ready_fd
, &c
, sizeof(c
)) < 0)
2019 /* Try to read an error code from the child */
2020 n
= read(errno_pipe
[0], &r
, sizeof(r
));
2023 if (n
== sizeof(r
)) { /* an error code was sent to us */
2028 if (n
!= 0) /* on success we should have read 0 bytes */
2031 r
= wait_for_terminate_and_check("(sd-userns)", pid
, 0);
2035 if (r
!= EXIT_SUCCESS
) /* If something strange happened with the child, let's consider this fatal, too */
2041 static int setup_exec_directory(
2042 const ExecContext
*context
,
2043 const ExecParameters
*params
,
2046 ExecDirectoryType type
,
2049 static const int exit_status_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
2050 [EXEC_DIRECTORY_RUNTIME
] = EXIT_RUNTIME_DIRECTORY
,
2051 [EXEC_DIRECTORY_STATE
] = EXIT_STATE_DIRECTORY
,
2052 [EXEC_DIRECTORY_CACHE
] = EXIT_CACHE_DIRECTORY
,
2053 [EXEC_DIRECTORY_LOGS
] = EXIT_LOGS_DIRECTORY
,
2054 [EXEC_DIRECTORY_CONFIGURATION
] = EXIT_CONFIGURATION_DIRECTORY
,
2061 assert(type
>= 0 && type
< _EXEC_DIRECTORY_TYPE_MAX
);
2062 assert(exit_status
);
2064 if (!params
->prefix
[type
])
2067 if (params
->flags
& EXEC_CHOWN_DIRECTORIES
) {
2068 if (!uid_is_valid(uid
))
2070 if (!gid_is_valid(gid
))
2074 STRV_FOREACH(rt
, context
->directories
[type
].paths
) {
2075 _cleanup_free_
char *p
= NULL
, *pp
= NULL
;
2077 p
= path_join(params
->prefix
[type
], *rt
);
2083 r
= mkdir_parents_label(p
, 0755);
2087 if (context
->dynamic_user
&&
2088 (!IN_SET(type
, EXEC_DIRECTORY_RUNTIME
, EXEC_DIRECTORY_CONFIGURATION
) ||
2089 (type
== EXEC_DIRECTORY_RUNTIME
&& context
->runtime_directory_preserve_mode
!= EXEC_PRESERVE_NO
))) {
2090 _cleanup_free_
char *private_root
= NULL
;
2092 /* So, here's one extra complication when dealing with DynamicUser=1 units. In that case we
2093 * want to avoid leaving a directory around fully accessible that is owned by a dynamic user
2094 * whose UID is later on reused. To lock this down we use the same trick used by container
2095 * managers to prohibit host users to get access to files of the same UID in containers: we
2096 * place everything inside a directory that has an access mode of 0700 and is owned root:root,
2097 * so that it acts as security boundary for unprivileged host code. We then use fs namespacing
2098 * to make this directory permeable for the service itself.
2100 * Specifically: for a service which wants a special directory "foo/" we first create a
2101 * directory "private/" with access mode 0700 owned by root:root. Then we place "foo" inside of
2102 * that directory (i.e. "private/foo/"), and make "foo" a symlink to "private/foo". This way,
2103 * privileged host users can access "foo/" as usual, but unprivileged host users can't look
2104 * into it. Inside of the namespaceof the container "private/" is replaced by a more liberally
2105 * accessible tmpfs, into which the host's "private/foo/" is mounted under the same name, thus
2106 * disabling the access boundary for the service and making sure it only gets access to the
2107 * dirs it needs but no others. Tricky? Yes, absolutely, but it works!
2109 * Note that we don't do this for EXEC_DIRECTORY_CONFIGURATION as that's assumed not to be
2110 * owned by the service itself.
2111 * Also, note that we don't do this for EXEC_DIRECTORY_RUNTIME as that's often used for sharing
2112 * files or sockets with other services. */
2114 private_root
= path_join(params
->prefix
[type
], "private");
2115 if (!private_root
) {
2120 /* First set up private root if it doesn't exist yet, with access mode 0700 and owned by root:root */
2121 r
= mkdir_safe_label(private_root
, 0700, 0, 0, MKDIR_WARN_MODE
);
2125 pp
= path_join(private_root
, *rt
);
2131 /* Create all directories between the configured directory and this private root, and mark them 0755 */
2132 r
= mkdir_parents_label(pp
, 0755);
2136 if (is_dir(p
, false) > 0 &&
2137 (laccess(pp
, F_OK
) < 0 && errno
== ENOENT
)) {
2139 /* Hmm, the private directory doesn't exist yet, but the normal one exists? If so, move
2140 * it over. Most likely the service has been upgraded from one that didn't use
2141 * DynamicUser=1, to one that does. */
2143 if (rename(p
, pp
) < 0) {
2148 /* Otherwise, create the actual directory for the service */
2150 r
= mkdir_label(pp
, context
->directories
[type
].mode
);
2151 if (r
< 0 && r
!= -EEXIST
)
2155 /* And link it up from the original place */
2156 r
= symlink_idempotent(pp
, p
, true);
2161 r
= mkdir_label(p
, context
->directories
[type
].mode
);
2166 if (type
== EXEC_DIRECTORY_CONFIGURATION
) {
2169 /* Don't change the owner/access mode of the configuration directory,
2170 * as in the common case it is not written to by a service, and shall
2171 * not be writable. */
2173 if (stat(p
, &st
) < 0) {
2178 /* Still complain if the access mode doesn't match */
2179 if (((st
.st_mode
^ context
->directories
[type
].mode
) & 07777) != 0)
2180 log_warning("%s \'%s\' already exists but the mode is different. "
2181 "(File system: %o %sMode: %o)",
2182 exec_directory_type_to_string(type
), *rt
,
2183 st
.st_mode
& 07777, exec_directory_type_to_string(type
), context
->directories
[type
].mode
& 07777);
2190 /* Lock down the access mode (we use chmod_and_chown() to make this idempotent. We don't
2191 * specifiy UID/GID here, so that path_chown_recursive() can optimize things depending on the
2192 * current UID/GID ownership.) */
2193 r
= chmod_and_chown(pp
?: p
, context
->directories
[type
].mode
, UID_INVALID
, GID_INVALID
);
2197 /* Then, change the ownership of the whole tree, if necessary. When dynamic users are used we
2198 * drop the suid/sgid bits, since we really don't want SUID/SGID files for dynamic UID/GID
2199 * assignments to exist.*/
2200 r
= path_chown_recursive(pp
?: p
, uid
, gid
, context
->dynamic_user
? 01777 : 07777);
2208 *exit_status
= exit_status_table
[type
];
2213 static int setup_smack(
2214 const ExecContext
*context
,
2215 const ExecCommand
*command
) {
2222 if (context
->smack_process_label
) {
2223 r
= mac_smack_apply_pid(0, context
->smack_process_label
);
2227 #ifdef SMACK_DEFAULT_PROCESS_LABEL
2229 _cleanup_free_
char *exec_label
= NULL
;
2231 r
= mac_smack_read(command
->path
, SMACK_ATTR_EXEC
, &exec_label
);
2232 if (r
< 0 && !IN_SET(r
, -ENODATA
, -EOPNOTSUPP
))
2235 r
= mac_smack_apply_pid(0, exec_label
? : SMACK_DEFAULT_PROCESS_LABEL
);
2245 static int compile_bind_mounts(
2246 const ExecContext
*context
,
2247 const ExecParameters
*params
,
2248 BindMount
**ret_bind_mounts
,
2249 size_t *ret_n_bind_mounts
,
2250 char ***ret_empty_directories
) {
2252 _cleanup_strv_free_
char **empty_directories
= NULL
;
2253 BindMount
*bind_mounts
;
2255 ExecDirectoryType t
;
2260 assert(ret_bind_mounts
);
2261 assert(ret_n_bind_mounts
);
2262 assert(ret_empty_directories
);
2264 n
= context
->n_bind_mounts
;
2265 for (t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
2266 if (!params
->prefix
[t
])
2269 n
+= strv_length(context
->directories
[t
].paths
);
2273 *ret_bind_mounts
= NULL
;
2274 *ret_n_bind_mounts
= 0;
2275 *ret_empty_directories
= NULL
;
2279 bind_mounts
= new(BindMount
, n
);
2283 for (i
= 0; i
< context
->n_bind_mounts
; i
++) {
2284 BindMount
*item
= context
->bind_mounts
+ i
;
2287 s
= strdup(item
->source
);
2293 d
= strdup(item
->destination
);
2300 bind_mounts
[h
++] = (BindMount
) {
2303 .read_only
= item
->read_only
,
2304 .recursive
= item
->recursive
,
2305 .ignore_enoent
= item
->ignore_enoent
,
2309 for (t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
2312 if (!params
->prefix
[t
])
2315 if (strv_isempty(context
->directories
[t
].paths
))
2318 if (context
->dynamic_user
&&
2319 !IN_SET(t
, EXEC_DIRECTORY_RUNTIME
, EXEC_DIRECTORY_CONFIGURATION
) &&
2320 !(context
->root_directory
|| context
->root_image
)) {
2323 /* So this is for a dynamic user, and we need to make sure the process can access its own
2324 * directory. For that we overmount the usually inaccessible "private" subdirectory with a
2325 * tmpfs that makes it accessible and is empty except for the submounts we do this for. */
2327 private_root
= strjoin(params
->prefix
[t
], "/private");
2328 if (!private_root
) {
2333 r
= strv_consume(&empty_directories
, private_root
);
2338 STRV_FOREACH(suffix
, context
->directories
[t
].paths
) {
2341 if (context
->dynamic_user
&&
2342 !IN_SET(t
, EXEC_DIRECTORY_RUNTIME
, EXEC_DIRECTORY_CONFIGURATION
))
2343 s
= strjoin(params
->prefix
[t
], "/private/", *suffix
);
2345 s
= strjoin(params
->prefix
[t
], "/", *suffix
);
2351 if (context
->dynamic_user
&&
2352 !IN_SET(t
, EXEC_DIRECTORY_RUNTIME
, EXEC_DIRECTORY_CONFIGURATION
) &&
2353 (context
->root_directory
|| context
->root_image
))
2354 /* When RootDirectory= or RootImage= are set, then the symbolic link to the private
2355 * directory is not created on the root directory. So, let's bind-mount the directory
2356 * on the 'non-private' place. */
2357 d
= strjoin(params
->prefix
[t
], "/", *suffix
);
2366 bind_mounts
[h
++] = (BindMount
) {
2370 .nosuid
= context
->dynamic_user
, /* don't allow suid/sgid when DynamicUser= is on */
2372 .ignore_enoent
= false,
2379 *ret_bind_mounts
= bind_mounts
;
2380 *ret_n_bind_mounts
= n
;
2381 *ret_empty_directories
= TAKE_PTR(empty_directories
);
2386 bind_mount_free_many(bind_mounts
, h
);
2390 static int apply_mount_namespace(
2392 const ExecCommand
*command
,
2393 const ExecContext
*context
,
2394 const ExecParameters
*params
,
2395 const ExecRuntime
*runtime
) {
2397 _cleanup_strv_free_
char **empty_directories
= NULL
;
2398 char *tmp
= NULL
, *var
= NULL
;
2399 const char *root_dir
= NULL
, *root_image
= NULL
;
2400 NamespaceInfo ns_info
;
2401 bool needs_sandboxing
;
2402 BindMount
*bind_mounts
= NULL
;
2403 size_t n_bind_mounts
= 0;
2408 /* The runtime struct only contains the parent of the private /tmp,
2409 * which is non-accessible to world users. Inside of it there's a /tmp
2410 * that is sticky, and that's the one we want to use here. */
2412 if (context
->private_tmp
&& runtime
) {
2413 if (runtime
->tmp_dir
)
2414 tmp
= strjoina(runtime
->tmp_dir
, "/tmp");
2415 if (runtime
->var_tmp_dir
)
2416 var
= strjoina(runtime
->var_tmp_dir
, "/tmp");
2419 if (params
->flags
& EXEC_APPLY_CHROOT
) {
2420 root_image
= context
->root_image
;
2423 root_dir
= context
->root_directory
;
2426 r
= compile_bind_mounts(context
, params
, &bind_mounts
, &n_bind_mounts
, &empty_directories
);
2430 needs_sandboxing
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && !(command
->flags
& EXEC_COMMAND_FULLY_PRIVILEGED
);
2431 if (needs_sandboxing
)
2432 ns_info
= (NamespaceInfo
) {
2433 .ignore_protect_paths
= false,
2434 .private_dev
= context
->private_devices
,
2435 .protect_control_groups
= context
->protect_control_groups
,
2436 .protect_kernel_tunables
= context
->protect_kernel_tunables
,
2437 .protect_kernel_modules
= context
->protect_kernel_modules
,
2438 .protect_hostname
= context
->protect_hostname
,
2439 .mount_apivfs
= context
->mount_apivfs
,
2440 .private_mounts
= context
->private_mounts
,
2442 else if (!context
->dynamic_user
&& root_dir
)
2444 * If DynamicUser=no and RootDirectory= is set then lets pass a relaxed
2445 * sandbox info, otherwise enforce it, don't ignore protected paths and
2446 * fail if we are enable to apply the sandbox inside the mount namespace.
2448 ns_info
= (NamespaceInfo
) {
2449 .ignore_protect_paths
= true,
2452 ns_info
= (NamespaceInfo
) {};
2454 if (context
->mount_flags
== MS_SHARED
)
2455 log_unit_debug(u
, "shared mount propagation hidden by other fs namespacing unit settings: ignoring");
2457 r
= setup_namespace(root_dir
, root_image
,
2458 &ns_info
, context
->read_write_paths
,
2459 needs_sandboxing
? context
->read_only_paths
: NULL
,
2460 needs_sandboxing
? context
->inaccessible_paths
: NULL
,
2464 context
->temporary_filesystems
,
2465 context
->n_temporary_filesystems
,
2468 needs_sandboxing
? context
->protect_home
: PROTECT_HOME_NO
,
2469 needs_sandboxing
? context
->protect_system
: PROTECT_SYSTEM_NO
,
2470 context
->mount_flags
,
2471 DISSECT_IMAGE_DISCARD_ON_LOOP
);
2473 bind_mount_free_many(bind_mounts
, n_bind_mounts
);
2475 /* If we couldn't set up the namespace this is probably due to a missing capability. setup_namespace() reports
2476 * that with a special, recognizable error ENOANO. In this case, silently proceeed, but only if exclusively
2477 * sandboxing options were used, i.e. nothing such as RootDirectory= or BindMount= that would result in a
2478 * completely different execution environment. */
2480 if (n_bind_mounts
== 0 &&
2481 context
->n_temporary_filesystems
== 0 &&
2482 !root_dir
&& !root_image
&&
2483 !context
->dynamic_user
) {
2484 log_unit_debug(u
, "Failed to set up namespace, assuming containerized execution and ignoring.");
2488 log_unit_debug(u
, "Failed to set up namespace, and refusing to continue since the selected namespacing options alter mount environment non-trivially.\n"
2489 "Bind mounts: %zu, temporary filesystems: %zu, root directory: %s, root image: %s, dynamic user: %s",
2490 n_bind_mounts
, context
->n_temporary_filesystems
, yes_no(root_dir
), yes_no(root_image
), yes_no(context
->dynamic_user
));
2498 static int apply_working_directory(
2499 const ExecContext
*context
,
2500 const ExecParameters
*params
,
2502 const bool needs_mount_ns
,
2508 assert(exit_status
);
2510 if (context
->working_directory_home
) {
2513 *exit_status
= EXIT_CHDIR
;
2519 } else if (context
->working_directory
)
2520 wd
= context
->working_directory
;
2524 if (params
->flags
& EXEC_APPLY_CHROOT
) {
2525 if (!needs_mount_ns
&& context
->root_directory
)
2526 if (chroot(context
->root_directory
) < 0) {
2527 *exit_status
= EXIT_CHROOT
;
2533 d
= prefix_roota(context
->root_directory
, wd
);
2535 if (chdir(d
) < 0 && !context
->working_directory_missing_ok
) {
2536 *exit_status
= EXIT_CHDIR
;
2543 static int setup_keyring(
2545 const ExecContext
*context
,
2546 const ExecParameters
*p
,
2547 uid_t uid
, gid_t gid
) {
2549 key_serial_t keyring
;
2558 /* Let's set up a new per-service "session" kernel keyring for each system service. This has the benefit that
2559 * each service runs with its own keyring shared among all processes of the service, but with no hook-up beyond
2560 * that scope, and in particular no link to the per-UID keyring. If we don't do this the keyring will be
2561 * automatically created on-demand and then linked to the per-UID keyring, by the kernel. The kernel's built-in
2562 * on-demand behaviour is very appropriate for login users, but probably not so much for system services, where
2563 * UIDs are not necessarily specific to a service but reused (at least in the case of UID 0). */
2565 if (context
->keyring_mode
== EXEC_KEYRING_INHERIT
)
2568 /* Acquiring a reference to the user keyring is nasty. We briefly change identity in order to get things set up
2569 * properly by the kernel. If we don't do that then we can't create it atomically, and that sucks for parallel
2570 * execution. This mimics what pam_keyinit does, too. Setting up session keyring, to be owned by the right user
2571 * & group is just as nasty as acquiring a reference to the user keyring. */
2573 saved_uid
= getuid();
2574 saved_gid
= getgid();
2576 if (gid_is_valid(gid
) && gid
!= saved_gid
) {
2577 if (setregid(gid
, -1) < 0)
2578 return log_unit_error_errno(u
, errno
, "Failed to change GID for user keyring: %m");
2581 if (uid_is_valid(uid
) && uid
!= saved_uid
) {
2582 if (setreuid(uid
, -1) < 0) {
2583 r
= log_unit_error_errno(u
, errno
, "Failed to change UID for user keyring: %m");
2588 keyring
= keyctl(KEYCTL_JOIN_SESSION_KEYRING
, 0, 0, 0, 0);
2589 if (keyring
== -1) {
2590 if (errno
== ENOSYS
)
2591 log_unit_debug_errno(u
, errno
, "Kernel keyring not supported, ignoring.");
2592 else if (IN_SET(errno
, EACCES
, EPERM
))
2593 log_unit_debug_errno(u
, errno
, "Kernel keyring access prohibited, ignoring.");
2594 else if (errno
== EDQUOT
)
2595 log_unit_debug_errno(u
, errno
, "Out of kernel keyrings to allocate, ignoring.");
2597 r
= log_unit_error_errno(u
, errno
, "Setting up kernel keyring failed: %m");
2602 /* When requested link the user keyring into the session keyring. */
2603 if (context
->keyring_mode
== EXEC_KEYRING_SHARED
) {
2605 if (keyctl(KEYCTL_LINK
,
2606 KEY_SPEC_USER_KEYRING
,
2607 KEY_SPEC_SESSION_KEYRING
, 0, 0) < 0) {
2608 r
= log_unit_error_errno(u
, errno
, "Failed to link user keyring into session keyring: %m");
2613 /* Restore uid/gid back */
2614 if (uid_is_valid(uid
) && uid
!= saved_uid
) {
2615 if (setreuid(saved_uid
, -1) < 0) {
2616 r
= log_unit_error_errno(u
, errno
, "Failed to change UID back for user keyring: %m");
2621 if (gid_is_valid(gid
) && gid
!= saved_gid
) {
2622 if (setregid(saved_gid
, -1) < 0)
2623 return log_unit_error_errno(u
, errno
, "Failed to change GID back for user keyring: %m");
2626 /* Populate they keyring with the invocation ID by default, as original saved_uid. */
2627 if (!sd_id128_is_null(u
->invocation_id
)) {
2630 key
= add_key("user", "invocation_id", &u
->invocation_id
, sizeof(u
->invocation_id
), KEY_SPEC_SESSION_KEYRING
);
2632 log_unit_debug_errno(u
, errno
, "Failed to add invocation ID to keyring, ignoring: %m");
2634 if (keyctl(KEYCTL_SETPERM
, key
,
2635 KEY_POS_VIEW
|KEY_POS_READ
|KEY_POS_SEARCH
|
2636 KEY_USR_VIEW
|KEY_USR_READ
|KEY_USR_SEARCH
, 0, 0) < 0)
2637 r
= log_unit_error_errno(u
, errno
, "Failed to restrict invocation ID permission: %m");
2642 /* Revert back uid & gid for the the last time, and exit */
2643 /* no extra logging, as only the first already reported error matters */
2644 if (getuid() != saved_uid
)
2645 (void) setreuid(saved_uid
, -1);
2647 if (getgid() != saved_gid
)
2648 (void) setregid(saved_gid
, -1);
2653 static void append_socket_pair(int *array
, size_t *n
, const int pair
[static 2]) {
2661 array
[(*n
)++] = pair
[0];
2663 array
[(*n
)++] = pair
[1];
2666 static int close_remaining_fds(
2667 const ExecParameters
*params
,
2668 const ExecRuntime
*runtime
,
2669 const DynamicCreds
*dcreds
,
2673 int *fds
, size_t n_fds
) {
2675 size_t n_dont_close
= 0;
2676 int dont_close
[n_fds
+ 12];
2680 if (params
->stdin_fd
>= 0)
2681 dont_close
[n_dont_close
++] = params
->stdin_fd
;
2682 if (params
->stdout_fd
>= 0)
2683 dont_close
[n_dont_close
++] = params
->stdout_fd
;
2684 if (params
->stderr_fd
>= 0)
2685 dont_close
[n_dont_close
++] = params
->stderr_fd
;
2688 dont_close
[n_dont_close
++] = socket_fd
;
2690 dont_close
[n_dont_close
++] = exec_fd
;
2692 memcpy(dont_close
+ n_dont_close
, fds
, sizeof(int) * n_fds
);
2693 n_dont_close
+= n_fds
;
2697 append_socket_pair(dont_close
, &n_dont_close
, runtime
->netns_storage_socket
);
2701 append_socket_pair(dont_close
, &n_dont_close
, dcreds
->user
->storage_socket
);
2703 append_socket_pair(dont_close
, &n_dont_close
, dcreds
->group
->storage_socket
);
2706 if (user_lookup_fd
>= 0)
2707 dont_close
[n_dont_close
++] = user_lookup_fd
;
2709 return close_all_fds(dont_close
, n_dont_close
);
2712 static int send_user_lookup(
2720 /* Send the resolved UID/GID to PID 1 after we learnt it. We send a single datagram, containing the UID/GID
2721 * data as well as the unit name. Note that we suppress sending this if no user/group to resolve was
2724 if (user_lookup_fd
< 0)
2727 if (!uid_is_valid(uid
) && !gid_is_valid(gid
))
2730 if (writev(user_lookup_fd
,
2732 IOVEC_INIT(&uid
, sizeof(uid
)),
2733 IOVEC_INIT(&gid
, sizeof(gid
)),
2734 IOVEC_INIT_STRING(unit
->id
) }, 3) < 0)
2740 static int acquire_home(const ExecContext
*c
, uid_t uid
, const char** home
, char **buf
) {
2747 /* If WorkingDirectory=~ is set, try to acquire a usable home directory. */
2752 if (!c
->working_directory_home
)
2755 r
= get_home_dir(buf
);
2763 static int compile_suggested_paths(const ExecContext
*c
, const ExecParameters
*p
, char ***ret
) {
2764 _cleanup_strv_free_
char ** list
= NULL
;
2765 ExecDirectoryType t
;
2772 assert(c
->dynamic_user
);
2774 /* Compile a list of paths that it might make sense to read the owning UID from to use as initial candidate for
2775 * dynamic UID allocation, in order to save us from doing costly recursive chown()s of the special
2778 for (t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
2781 if (t
== EXEC_DIRECTORY_CONFIGURATION
)
2787 STRV_FOREACH(i
, c
->directories
[t
].paths
) {
2790 if (t
== EXEC_DIRECTORY_RUNTIME
)
2791 e
= strjoin(p
->prefix
[t
], "/", *i
);
2793 e
= strjoin(p
->prefix
[t
], "/private/", *i
);
2797 r
= strv_consume(&list
, e
);
2803 *ret
= TAKE_PTR(list
);
2808 static char *exec_command_line(char **argv
);
2810 static int exec_parameters_get_cgroup_path(const ExecParameters
*params
, char **ret
) {
2811 bool using_subcgroup
;
2817 if (!params
->cgroup_path
)
2820 /* If we are called for a unit where cgroup delegation is on, and the payload created its own populated
2821 * subcgroup (which we expect it to do, after all it asked for delegation), then we cannot place the control
2822 * processes started after the main unit's process in the unit's main cgroup because it is now an inner one,
2823 * and inner cgroups may not contain processes. Hence, if delegation is on, and this is a control process,
2824 * let's use ".control" as subcgroup instead. Note that we do so only for ExecStartPost=, ExecReload=,
2825 * ExecStop=, ExecStopPost=, i.e. for the commands where the main process is already forked. For ExecStartPre=
2826 * this is not necessary, the cgroup is still empty. We distinguish these cases with the EXEC_CONTROL_CGROUP
2827 * flag, which is only passed for the former statements, not for the latter. */
2829 using_subcgroup
= FLAGS_SET(params
->flags
, EXEC_CONTROL_CGROUP
|EXEC_CGROUP_DELEGATE
|EXEC_IS_CONTROL
);
2830 if (using_subcgroup
)
2831 p
= strjoin(params
->cgroup_path
, "/.control");
2833 p
= strdup(params
->cgroup_path
);
2838 return using_subcgroup
;
2841 static int exec_child(
2843 const ExecCommand
*command
,
2844 const ExecContext
*context
,
2845 const ExecParameters
*params
,
2846 ExecRuntime
*runtime
,
2847 DynamicCreds
*dcreds
,
2851 size_t n_socket_fds
,
2852 size_t n_storage_fds
,
2857 _cleanup_strv_free_
char **our_env
= NULL
, **pass_env
= NULL
, **accum_env
= NULL
, **replaced_argv
= NULL
;
2858 int *fds_with_exec_fd
, n_fds_with_exec_fd
, r
, ngids
= 0, exec_fd
= -1;
2859 _cleanup_free_ gid_t
*supplementary_gids
= NULL
;
2860 const char *username
= NULL
, *groupname
= NULL
;
2861 _cleanup_free_
char *home_buffer
= NULL
;
2862 const char *home
= NULL
, *shell
= NULL
;
2863 char **final_argv
= NULL
;
2864 dev_t journal_stream_dev
= 0;
2865 ino_t journal_stream_ino
= 0;
2866 bool needs_sandboxing
, /* Do we need to set up full sandboxing? (i.e. all namespacing, all MAC stuff, caps, yadda yadda */
2867 needs_setuid
, /* Do we need to do the actual setresuid()/setresgid() calls? */
2868 needs_mount_namespace
, /* Do we need to set up a mount namespace for this kernel? */
2869 needs_ambient_hack
; /* Do we need to apply the ambient capabilities hack? */
2871 _cleanup_free_
char *mac_selinux_context_net
= NULL
;
2872 bool use_selinux
= false;
2875 bool use_smack
= false;
2878 bool use_apparmor
= false;
2880 uid_t uid
= UID_INVALID
;
2881 gid_t gid
= GID_INVALID
;
2883 ExecDirectoryType dt
;
2890 assert(exit_status
);
2892 rename_process_from_path(command
->path
);
2894 /* We reset exactly these signals, since they are the
2895 * only ones we set to SIG_IGN in the main daemon. All
2896 * others we leave untouched because we set them to
2897 * SIG_DFL or a valid handler initially, both of which
2898 * will be demoted to SIG_DFL. */
2899 (void) default_signals(SIGNALS_CRASH_HANDLER
,
2900 SIGNALS_IGNORE
, -1);
2902 if (context
->ignore_sigpipe
)
2903 (void) ignore_signals(SIGPIPE
, -1);
2905 r
= reset_signal_mask();
2907 *exit_status
= EXIT_SIGNAL_MASK
;
2908 return log_unit_error_errno(unit
, r
, "Failed to set process signal mask: %m");
2911 if (params
->idle_pipe
)
2912 do_idle_pipe_dance(params
->idle_pipe
);
2914 /* Close fds we don't need very early to make sure we don't block init reexecution because it cannot bind its
2915 * sockets. Among the fds we close are the logging fds, and we want to keep them closed, so that we don't have
2916 * any fds open we don't really want open during the transition. In order to make logging work, we switch the
2917 * log subsystem into open_when_needed mode, so that it reopens the logs on every single log call. */
2920 log_set_open_when_needed(true);
2922 /* In case anything used libc syslog(), close this here, too */
2925 n_fds
= n_socket_fds
+ n_storage_fds
;
2926 r
= close_remaining_fds(params
, runtime
, dcreds
, user_lookup_fd
, socket_fd
, params
->exec_fd
, fds
, n_fds
);
2928 *exit_status
= EXIT_FDS
;
2929 return log_unit_error_errno(unit
, r
, "Failed to close unwanted file descriptors: %m");
2932 if (!context
->same_pgrp
)
2934 *exit_status
= EXIT_SETSID
;
2935 return log_unit_error_errno(unit
, errno
, "Failed to create new process session: %m");
2938 exec_context_tty_reset(context
, params
);
2940 if (unit_shall_confirm_spawn(unit
)) {
2941 const char *vc
= params
->confirm_spawn
;
2942 _cleanup_free_
char *cmdline
= NULL
;
2944 cmdline
= exec_command_line(command
->argv
);
2946 *exit_status
= EXIT_MEMORY
;
2950 r
= ask_for_confirmation(vc
, unit
, cmdline
);
2951 if (r
!= CONFIRM_EXECUTE
) {
2952 if (r
== CONFIRM_PRETEND_SUCCESS
) {
2953 *exit_status
= EXIT_SUCCESS
;
2956 *exit_status
= EXIT_CONFIRM
;
2957 log_unit_error(unit
, "Execution cancelled by the user");
2962 /* We are about to invoke NSS and PAM modules. Let's tell them what we are doing here, maybe they care. This is
2963 * used by nss-resolve to disable itself when we are about to start systemd-resolved, to avoid deadlocks. Note
2964 * that these env vars do not survive the execve(), which means they really only apply to the PAM and NSS
2965 * invocations themselves. Also note that while we'll only invoke NSS modules involved in user management they
2966 * might internally call into other NSS modules that are involved in hostname resolution, we never know. */
2967 if (setenv("SYSTEMD_ACTIVATION_UNIT", unit
->id
, true) != 0 ||
2968 setenv("SYSTEMD_ACTIVATION_SCOPE", MANAGER_IS_SYSTEM(unit
->manager
) ? "system" : "user", true) != 0) {
2969 *exit_status
= EXIT_MEMORY
;
2970 return log_unit_error_errno(unit
, errno
, "Failed to update environment: %m");
2973 if (context
->dynamic_user
&& dcreds
) {
2974 _cleanup_strv_free_
char **suggested_paths
= NULL
;
2976 /* On top of that, make sure we bypass our own NSS module nss-systemd comprehensively for any NSS
2977 * checks, if DynamicUser=1 is used, as we shouldn't create a feedback loop with ourselves here.*/
2978 if (putenv((char*) "SYSTEMD_NSS_DYNAMIC_BYPASS=1") != 0) {
2979 *exit_status
= EXIT_USER
;
2980 return log_unit_error_errno(unit
, errno
, "Failed to update environment: %m");
2983 r
= compile_suggested_paths(context
, params
, &suggested_paths
);
2985 *exit_status
= EXIT_MEMORY
;
2989 r
= dynamic_creds_realize(dcreds
, suggested_paths
, &uid
, &gid
);
2991 *exit_status
= EXIT_USER
;
2993 log_unit_error(unit
, "Failed to update dynamic user credentials: User or group with specified name already exists.");
2996 return log_unit_error_errno(unit
, r
, "Failed to update dynamic user credentials: %m");
2999 if (!uid_is_valid(uid
)) {
3000 *exit_status
= EXIT_USER
;
3001 log_unit_error(unit
, "UID validation failed for \""UID_FMT
"\"", uid
);
3005 if (!gid_is_valid(gid
)) {
3006 *exit_status
= EXIT_USER
;
3007 log_unit_error(unit
, "GID validation failed for \""GID_FMT
"\"", gid
);
3012 username
= dcreds
->user
->name
;
3015 r
= get_fixed_user(context
, &username
, &uid
, &gid
, &home
, &shell
);
3017 *exit_status
= EXIT_USER
;
3018 return log_unit_error_errno(unit
, r
, "Failed to determine user credentials: %m");
3021 r
= get_fixed_group(context
, &groupname
, &gid
);
3023 *exit_status
= EXIT_GROUP
;
3024 return log_unit_error_errno(unit
, r
, "Failed to determine group credentials: %m");
3028 /* Initialize user supplementary groups and get SupplementaryGroups= ones */
3029 r
= get_supplementary_groups(context
, username
, groupname
, gid
,
3030 &supplementary_gids
, &ngids
);
3032 *exit_status
= EXIT_GROUP
;
3033 return log_unit_error_errno(unit
, r
, "Failed to determine supplementary groups: %m");
3036 r
= send_user_lookup(unit
, user_lookup_fd
, uid
, gid
);
3038 *exit_status
= EXIT_USER
;
3039 return log_unit_error_errno(unit
, r
, "Failed to send user credentials to PID1: %m");
3042 user_lookup_fd
= safe_close(user_lookup_fd
);
3044 r
= acquire_home(context
, uid
, &home
, &home_buffer
);
3046 *exit_status
= EXIT_CHDIR
;
3047 return log_unit_error_errno(unit
, r
, "Failed to determine $HOME for user: %m");
3050 /* If a socket is connected to STDIN/STDOUT/STDERR, we
3051 * must sure to drop O_NONBLOCK */
3053 (void) fd_nonblock(socket_fd
, false);
3055 /* Journald will try to look-up our cgroup in order to populate _SYSTEMD_CGROUP and _SYSTEMD_UNIT fields.
3056 * Hence we need to migrate to the target cgroup from init.scope before connecting to journald */
3057 if (params
->cgroup_path
) {
3058 _cleanup_free_
char *p
= NULL
;
3060 r
= exec_parameters_get_cgroup_path(params
, &p
);
3062 *exit_status
= EXIT_CGROUP
;
3063 return log_unit_error_errno(unit
, r
, "Failed to acquire cgroup path: %m");
3066 r
= cg_attach_everywhere(params
->cgroup_supported
, p
, 0, NULL
, NULL
);
3068 *exit_status
= EXIT_CGROUP
;
3069 return log_unit_error_errno(unit
, r
, "Failed to attach to cgroup %s: %m", p
);
3073 if (context
->network_namespace_path
&& runtime
&& runtime
->netns_storage_socket
[0] >= 0) {
3074 r
= open_netns_path(runtime
->netns_storage_socket
, context
->network_namespace_path
);
3076 *exit_status
= EXIT_NETWORK
;
3077 return log_unit_error_errno(unit
, r
, "Failed to open network namespace path %s: %m", context
->network_namespace_path
);
3081 r
= setup_input(context
, params
, socket_fd
, named_iofds
);
3083 *exit_status
= EXIT_STDIN
;
3084 return log_unit_error_errno(unit
, r
, "Failed to set up standard input: %m");
3087 r
= setup_output(unit
, context
, params
, STDOUT_FILENO
, socket_fd
, named_iofds
, basename(command
->path
), uid
, gid
, &journal_stream_dev
, &journal_stream_ino
);
3089 *exit_status
= EXIT_STDOUT
;
3090 return log_unit_error_errno(unit
, r
, "Failed to set up standard output: %m");
3093 r
= setup_output(unit
, context
, params
, STDERR_FILENO
, socket_fd
, named_iofds
, basename(command
->path
), uid
, gid
, &journal_stream_dev
, &journal_stream_ino
);
3095 *exit_status
= EXIT_STDERR
;
3096 return log_unit_error_errno(unit
, r
, "Failed to set up standard error output: %m");
3099 if (context
->oom_score_adjust_set
) {
3100 /* When we can't make this change due to EPERM, then let's silently skip over it. User namespaces
3101 * prohibit write access to this file, and we shouldn't trip up over that. */
3102 r
= set_oom_score_adjust(context
->oom_score_adjust
);
3103 if (IN_SET(r
, -EPERM
, -EACCES
))
3104 log_unit_debug_errno(unit
, r
, "Failed to adjust OOM setting, assuming containerized execution, ignoring: %m");
3106 *exit_status
= EXIT_OOM_ADJUST
;
3107 return log_unit_error_errno(unit
, r
, "Failed to adjust OOM setting: %m");
3111 if (context
->nice_set
)
3112 if (setpriority(PRIO_PROCESS
, 0, context
->nice
) < 0) {
3113 *exit_status
= EXIT_NICE
;
3114 return log_unit_error_errno(unit
, errno
, "Failed to set up process scheduling priority (nice level): %m");
3117 if (context
->cpu_sched_set
) {
3118 struct sched_param param
= {
3119 .sched_priority
= context
->cpu_sched_priority
,
3122 r
= sched_setscheduler(0,
3123 context
->cpu_sched_policy
|
3124 (context
->cpu_sched_reset_on_fork
?
3125 SCHED_RESET_ON_FORK
: 0),
3128 *exit_status
= EXIT_SETSCHEDULER
;
3129 return log_unit_error_errno(unit
, errno
, "Failed to set up CPU scheduling: %m");
3133 if (context
->cpuset
)
3134 if (sched_setaffinity(0, CPU_ALLOC_SIZE(context
->cpuset_ncpus
), context
->cpuset
) < 0) {
3135 *exit_status
= EXIT_CPUAFFINITY
;
3136 return log_unit_error_errno(unit
, errno
, "Failed to set up CPU affinity: %m");
3139 if (context
->ioprio_set
)
3140 if (ioprio_set(IOPRIO_WHO_PROCESS
, 0, context
->ioprio
) < 0) {
3141 *exit_status
= EXIT_IOPRIO
;
3142 return log_unit_error_errno(unit
, errno
, "Failed to set up IO scheduling priority: %m");
3145 if (context
->timer_slack_nsec
!= NSEC_INFINITY
)
3146 if (prctl(PR_SET_TIMERSLACK
, context
->timer_slack_nsec
) < 0) {
3147 *exit_status
= EXIT_TIMERSLACK
;
3148 return log_unit_error_errno(unit
, errno
, "Failed to set up timer slack: %m");
3151 if (context
->personality
!= PERSONALITY_INVALID
) {
3152 r
= safe_personality(context
->personality
);
3154 *exit_status
= EXIT_PERSONALITY
;
3155 return log_unit_error_errno(unit
, r
, "Failed to set up execution domain (personality): %m");
3159 if (context
->utmp_id
)
3160 utmp_put_init_process(context
->utmp_id
, getpid_cached(), getsid(0),
3162 context
->utmp_mode
== EXEC_UTMP_INIT
? INIT_PROCESS
:
3163 context
->utmp_mode
== EXEC_UTMP_LOGIN
? LOGIN_PROCESS
:
3167 if (uid_is_valid(uid
)) {
3168 r
= chown_terminal(STDIN_FILENO
, uid
);
3170 *exit_status
= EXIT_STDIN
;
3171 return log_unit_error_errno(unit
, r
, "Failed to change ownership of terminal: %m");
3175 /* If delegation is enabled we'll pass ownership of the cgroup to the user of the new process. On cgroup v1
3176 * this is only about systemd's own hierarchy, i.e. not the controller hierarchies, simply because that's not
3177 * safe. On cgroup v2 there's only one hierarchy anyway, and delegation is safe there, hence in that case only
3178 * touch a single hierarchy too. */
3179 if (params
->cgroup_path
&& context
->user
&& (params
->flags
& EXEC_CGROUP_DELEGATE
)) {
3180 r
= cg_set_access(SYSTEMD_CGROUP_CONTROLLER
, params
->cgroup_path
, uid
, gid
);
3182 *exit_status
= EXIT_CGROUP
;
3183 return log_unit_error_errno(unit
, r
, "Failed to adjust control group access: %m");
3187 for (dt
= 0; dt
< _EXEC_DIRECTORY_TYPE_MAX
; dt
++) {
3188 r
= setup_exec_directory(context
, params
, uid
, gid
, dt
, exit_status
);
3190 return log_unit_error_errno(unit
, r
, "Failed to set up special execution directory in %s: %m", params
->prefix
[dt
]);
3193 r
= build_environment(
3205 *exit_status
= EXIT_MEMORY
;
3209 r
= build_pass_environment(context
, &pass_env
);
3211 *exit_status
= EXIT_MEMORY
;
3215 accum_env
= strv_env_merge(5,
3216 params
->environment
,
3219 context
->environment
,
3223 *exit_status
= EXIT_MEMORY
;
3226 accum_env
= strv_env_clean(accum_env
);
3228 (void) umask(context
->umask
);
3230 r
= setup_keyring(unit
, context
, params
, uid
, gid
);
3232 *exit_status
= EXIT_KEYRING
;
3233 return log_unit_error_errno(unit
, r
, "Failed to set up kernel keyring: %m");
3236 /* We need sandboxing if the caller asked us to apply it and the command isn't explicitly excepted from it */
3237 needs_sandboxing
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && !(command
->flags
& EXEC_COMMAND_FULLY_PRIVILEGED
);
3239 /* 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 */
3240 needs_ambient_hack
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && (command
->flags
& EXEC_COMMAND_AMBIENT_MAGIC
) && !ambient_capabilities_supported();
3242 /* 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 */
3243 if (needs_ambient_hack
)
3244 needs_setuid
= false;
3246 needs_setuid
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && !(command
->flags
& (EXEC_COMMAND_FULLY_PRIVILEGED
|EXEC_COMMAND_NO_SETUID
));
3248 if (needs_sandboxing
) {
3249 /* MAC enablement checks need to be done before a new mount ns is created, as they rely on /sys being
3250 * present. The actual MAC context application will happen later, as late as possible, to avoid
3251 * impacting our own code paths. */
3254 use_selinux
= mac_selinux_use();
3257 use_smack
= mac_smack_use();
3260 use_apparmor
= mac_apparmor_use();
3264 if (needs_sandboxing
) {
3267 /* Let's set the resource limits before we call into PAM, so that pam_limits wins over what
3268 * is set here. (See below.) */
3270 r
= setrlimit_closest_all((const struct rlimit
* const *) context
->rlimit
, &which_failed
);
3272 *exit_status
= EXIT_LIMITS
;
3273 return log_unit_error_errno(unit
, r
, "Failed to adjust resource limit RLIMIT_%s: %m", rlimit_to_string(which_failed
));
3279 /* Let's call into PAM after we set up our own idea of resource limits to that pam_limits
3280 * wins here. (See above.) */
3282 if (context
->pam_name
&& username
) {
3283 r
= setup_pam(context
->pam_name
, username
, uid
, gid
, context
->tty_path
, &accum_env
, fds
, n_fds
);
3285 *exit_status
= EXIT_PAM
;
3286 return log_unit_error_errno(unit
, r
, "Failed to set up PAM session: %m");
3291 if ((context
->private_network
|| context
->network_namespace_path
) && runtime
&& runtime
->netns_storage_socket
[0] >= 0) {
3293 if (ns_type_supported(NAMESPACE_NET
)) {
3294 r
= setup_netns(runtime
->netns_storage_socket
);
3296 *exit_status
= EXIT_NETWORK
;
3297 return log_unit_error_errno(unit
, r
, "Failed to set up network namespacing: %m");
3299 } else if (context
->network_namespace_path
) {
3300 *exit_status
= EXIT_NETWORK
;
3301 return log_unit_error_errno(unit
, SYNTHETIC_ERRNO(EOPNOTSUPP
), "NetworkNamespacePath= is not supported, refusing.");
3303 log_unit_warning(unit
, "PrivateNetwork=yes is configured, but the kernel does not support network namespaces, ignoring.");
3306 needs_mount_namespace
= exec_needs_mount_namespace(context
, params
, runtime
);
3307 if (needs_mount_namespace
) {
3308 r
= apply_mount_namespace(unit
, command
, context
, params
, runtime
);
3310 *exit_status
= EXIT_NAMESPACE
;
3311 return log_unit_error_errno(unit
, r
, "Failed to set up mount namespacing: %m");
3315 if (context
->protect_hostname
) {
3316 if (ns_type_supported(NAMESPACE_UTS
)) {
3317 if (unshare(CLONE_NEWUTS
) < 0) {
3318 *exit_status
= EXIT_NAMESPACE
;
3319 return log_unit_error_errno(unit
, errno
, "Failed to set up UTS namespacing: %m");
3322 log_unit_warning(unit
, "ProtectHostname=yes is configured, but the kernel does not support UTS namespaces, ignoring namespace setup.");
3324 r
= seccomp_protect_hostname();
3326 *exit_status
= EXIT_SECCOMP
;
3327 return log_unit_error_errno(unit
, r
, "Failed to apply hostname restrictions: %m");
3332 /* Drop groups as early as possbile */
3334 r
= enforce_groups(gid
, supplementary_gids
, ngids
);
3336 *exit_status
= EXIT_GROUP
;
3337 return log_unit_error_errno(unit
, r
, "Changing group credentials failed: %m");
3341 if (needs_sandboxing
) {
3343 if (use_selinux
&& params
->selinux_context_net
&& socket_fd
>= 0) {
3344 r
= mac_selinux_get_child_mls_label(socket_fd
, command
->path
, context
->selinux_context
, &mac_selinux_context_net
);
3346 *exit_status
= EXIT_SELINUX_CONTEXT
;
3347 return log_unit_error_errno(unit
, r
, "Failed to determine SELinux context: %m");
3352 if (context
->private_users
) {
3353 r
= setup_private_users(uid
, gid
);
3355 *exit_status
= EXIT_USER
;
3356 return log_unit_error_errno(unit
, r
, "Failed to set up user namespacing: %m");
3361 /* We repeat the fd closing here, to make sure that nothing is leaked from the PAM modules. Note that we are
3362 * more aggressive this time since socket_fd and the netns fds we don't need anymore. We do keep the exec_fd
3363 * however if we have it as we want to keep it open until the final execve(). */
3365 if (params
->exec_fd
>= 0) {
3366 exec_fd
= params
->exec_fd
;
3368 if (exec_fd
< 3 + (int) n_fds
) {
3371 /* Let's move the exec fd far up, so that it's outside of the fd range we want to pass to the
3372 * process we are about to execute. */
3374 moved_fd
= fcntl(exec_fd
, F_DUPFD_CLOEXEC
, 3 + (int) n_fds
);
3376 *exit_status
= EXIT_FDS
;
3377 return log_unit_error_errno(unit
, errno
, "Couldn't move exec fd up: %m");
3380 safe_close(exec_fd
);
3383 /* This fd should be FD_CLOEXEC already, but let's make sure. */
3384 r
= fd_cloexec(exec_fd
, true);
3386 *exit_status
= EXIT_FDS
;
3387 return log_unit_error_errno(unit
, r
, "Failed to make exec fd FD_CLOEXEC: %m");
3391 fds_with_exec_fd
= newa(int, n_fds
+ 1);
3392 memcpy_safe(fds_with_exec_fd
, fds
, n_fds
* sizeof(int));
3393 fds_with_exec_fd
[n_fds
] = exec_fd
;
3394 n_fds_with_exec_fd
= n_fds
+ 1;
3396 fds_with_exec_fd
= fds
;
3397 n_fds_with_exec_fd
= n_fds
;
3400 r
= close_all_fds(fds_with_exec_fd
, n_fds_with_exec_fd
);
3402 r
= shift_fds(fds
, n_fds
);
3404 r
= flags_fds(fds
, n_socket_fds
, n_storage_fds
, context
->non_blocking
);
3406 *exit_status
= EXIT_FDS
;
3407 return log_unit_error_errno(unit
, r
, "Failed to adjust passed file descriptors: %m");
3410 /* At this point, the fds we want to pass to the program are all ready and set up, with O_CLOEXEC turned off
3411 * and at the right fd numbers. The are no other fds open, with one exception: the exec_fd if it is defined,
3412 * and it has O_CLOEXEC set, after all we want it to be closed by the execve(), so that our parent knows we
3415 secure_bits
= context
->secure_bits
;
3417 if (needs_sandboxing
) {
3420 /* Set the RTPRIO resource limit to 0, but only if nothing else was explicitly
3421 * requested. (Note this is placed after the general resource limit initialization, see
3422 * above, in order to take precedence.) */
3423 if (context
->restrict_realtime
&& !context
->rlimit
[RLIMIT_RTPRIO
]) {
3424 if (setrlimit(RLIMIT_RTPRIO
, &RLIMIT_MAKE_CONST(0)) < 0) {
3425 *exit_status
= EXIT_LIMITS
;
3426 return log_unit_error_errno(unit
, errno
, "Failed to adjust RLIMIT_RTPRIO resource limit: %m");
3431 /* LSM Smack needs the capability CAP_MAC_ADMIN to change the current execution security context of the
3432 * process. This is the latest place before dropping capabilities. Other MAC context are set later. */
3434 r
= setup_smack(context
, command
);
3436 *exit_status
= EXIT_SMACK_PROCESS_LABEL
;
3437 return log_unit_error_errno(unit
, r
, "Failed to set SMACK process label: %m");
3442 bset
= context
->capability_bounding_set
;
3443 /* If the ambient caps hack is enabled (which means the kernel can't do them, and the user asked for
3444 * our magic fallback), then let's add some extra caps, so that the service can drop privs of its own,
3445 * instead of us doing that */
3446 if (needs_ambient_hack
)
3447 bset
|= (UINT64_C(1) << CAP_SETPCAP
) |
3448 (UINT64_C(1) << CAP_SETUID
) |
3449 (UINT64_C(1) << CAP_SETGID
);
3451 if (!cap_test_all(bset
)) {
3452 r
= capability_bounding_set_drop(bset
, false);
3454 *exit_status
= EXIT_CAPABILITIES
;
3455 return log_unit_error_errno(unit
, r
, "Failed to drop capabilities: %m");
3459 /* This is done before enforce_user, but ambient set
3460 * does not survive over setresuid() if keep_caps is not set. */
3461 if (!needs_ambient_hack
&&
3462 context
->capability_ambient_set
!= 0) {
3463 r
= capability_ambient_set_apply(context
->capability_ambient_set
, true);
3465 *exit_status
= EXIT_CAPABILITIES
;
3466 return log_unit_error_errno(unit
, r
, "Failed to apply ambient capabilities (before UID change): %m");
3472 if (uid_is_valid(uid
)) {
3473 r
= enforce_user(context
, uid
);
3475 *exit_status
= EXIT_USER
;
3476 return log_unit_error_errno(unit
, r
, "Failed to change UID to " UID_FMT
": %m", uid
);
3479 if (!needs_ambient_hack
&&
3480 context
->capability_ambient_set
!= 0) {
3482 /* Fix the ambient capabilities after user change. */
3483 r
= capability_ambient_set_apply(context
->capability_ambient_set
, false);
3485 *exit_status
= EXIT_CAPABILITIES
;
3486 return log_unit_error_errno(unit
, r
, "Failed to apply ambient capabilities (after UID change): %m");
3489 /* If we were asked to change user and ambient capabilities
3490 * were requested, we had to add keep-caps to the securebits
3491 * so that we would maintain the inherited capability set
3492 * through the setresuid(). Make sure that the bit is added
3493 * also to the context secure_bits so that we don't try to
3494 * drop the bit away next. */
3496 secure_bits
|= 1<<SECURE_KEEP_CAPS
;
3501 /* Apply working directory here, because the working directory might be on NFS and only the user running
3502 * this service might have the correct privilege to change to the working directory */
3503 r
= apply_working_directory(context
, params
, home
, needs_mount_namespace
, exit_status
);
3505 return log_unit_error_errno(unit
, r
, "Changing to the requested working directory failed: %m");
3507 if (needs_sandboxing
) {
3508 /* Apply other MAC contexts late, but before seccomp syscall filtering, as those should really be last to
3509 * influence our own codepaths as little as possible. Moreover, applying MAC contexts usually requires
3510 * syscalls that are subject to seccomp filtering, hence should probably be applied before the syscalls
3511 * are restricted. */
3515 char *exec_context
= mac_selinux_context_net
?: context
->selinux_context
;
3518 r
= setexeccon(exec_context
);
3520 *exit_status
= EXIT_SELINUX_CONTEXT
;
3521 return log_unit_error_errno(unit
, r
, "Failed to change SELinux context to %s: %m", exec_context
);
3528 if (use_apparmor
&& context
->apparmor_profile
) {
3529 r
= aa_change_onexec(context
->apparmor_profile
);
3530 if (r
< 0 && !context
->apparmor_profile_ignore
) {
3531 *exit_status
= EXIT_APPARMOR_PROFILE
;
3532 return log_unit_error_errno(unit
, errno
, "Failed to prepare AppArmor profile change to %s: %m", context
->apparmor_profile
);
3537 /* PR_GET_SECUREBITS is not privileged, while PR_SET_SECUREBITS is. So to suppress potential EPERMs
3538 * we'll try not to call PR_SET_SECUREBITS unless necessary. */
3539 if (prctl(PR_GET_SECUREBITS
) != secure_bits
)
3540 if (prctl(PR_SET_SECUREBITS
, secure_bits
) < 0) {
3541 *exit_status
= EXIT_SECUREBITS
;
3542 return log_unit_error_errno(unit
, errno
, "Failed to set process secure bits: %m");
3545 if (context_has_no_new_privileges(context
))
3546 if (prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0) < 0) {
3547 *exit_status
= EXIT_NO_NEW_PRIVILEGES
;
3548 return log_unit_error_errno(unit
, errno
, "Failed to disable new privileges: %m");
3552 r
= apply_address_families(unit
, context
);
3554 *exit_status
= EXIT_ADDRESS_FAMILIES
;
3555 return log_unit_error_errno(unit
, r
, "Failed to restrict address families: %m");
3558 r
= apply_memory_deny_write_execute(unit
, context
);
3560 *exit_status
= EXIT_SECCOMP
;
3561 return log_unit_error_errno(unit
, r
, "Failed to disable writing to executable memory: %m");
3564 r
= apply_restrict_realtime(unit
, context
);
3566 *exit_status
= EXIT_SECCOMP
;
3567 return log_unit_error_errno(unit
, r
, "Failed to apply realtime restrictions: %m");
3570 r
= apply_restrict_namespaces(unit
, context
);
3572 *exit_status
= EXIT_SECCOMP
;
3573 return log_unit_error_errno(unit
, r
, "Failed to apply namespace restrictions: %m");
3576 r
= apply_protect_sysctl(unit
, context
);
3578 *exit_status
= EXIT_SECCOMP
;
3579 return log_unit_error_errno(unit
, r
, "Failed to apply sysctl restrictions: %m");
3582 r
= apply_protect_kernel_modules(unit
, context
);
3584 *exit_status
= EXIT_SECCOMP
;
3585 return log_unit_error_errno(unit
, r
, "Failed to apply module loading restrictions: %m");
3588 r
= apply_private_devices(unit
, context
);
3590 *exit_status
= EXIT_SECCOMP
;
3591 return log_unit_error_errno(unit
, r
, "Failed to set up private devices: %m");
3594 r
= apply_syscall_archs(unit
, context
);
3596 *exit_status
= EXIT_SECCOMP
;
3597 return log_unit_error_errno(unit
, r
, "Failed to apply syscall architecture restrictions: %m");
3600 r
= apply_lock_personality(unit
, context
);
3602 *exit_status
= EXIT_SECCOMP
;
3603 return log_unit_error_errno(unit
, r
, "Failed to lock personalities: %m");
3606 /* This really should remain the last step before the execve(), to make sure our own code is unaffected
3607 * by the filter as little as possible. */
3608 r
= apply_syscall_filter(unit
, context
, needs_ambient_hack
);
3610 *exit_status
= EXIT_SECCOMP
;
3611 return log_unit_error_errno(unit
, r
, "Failed to apply system call filters: %m");
3616 if (!strv_isempty(context
->unset_environment
)) {
3619 ee
= strv_env_delete(accum_env
, 1, context
->unset_environment
);
3621 *exit_status
= EXIT_MEMORY
;
3625 strv_free_and_replace(accum_env
, ee
);
3628 if (!FLAGS_SET(command
->flags
, EXEC_COMMAND_NO_ENV_EXPAND
)) {
3629 replaced_argv
= replace_env_argv(command
->argv
, accum_env
);
3630 if (!replaced_argv
) {
3631 *exit_status
= EXIT_MEMORY
;
3634 final_argv
= replaced_argv
;
3636 final_argv
= command
->argv
;
3638 if (DEBUG_LOGGING
) {
3639 _cleanup_free_
char *line
;
3641 line
= exec_command_line(final_argv
);
3643 log_struct(LOG_DEBUG
,
3644 "EXECUTABLE=%s", command
->path
,
3645 LOG_UNIT_MESSAGE(unit
, "Executing: %s", line
),
3647 LOG_UNIT_INVOCATION_ID(unit
));
3653 /* We have finished with all our initializations. Let's now let the manager know that. From this point
3654 * on, if the manager sees POLLHUP on the exec_fd, then execve() was successful. */
3656 if (write(exec_fd
, &hot
, sizeof(hot
)) < 0) {
3657 *exit_status
= EXIT_EXEC
;
3658 return log_unit_error_errno(unit
, errno
, "Failed to enable exec_fd: %m");
3662 execve(command
->path
, final_argv
, accum_env
);
3668 /* The execve() failed. This means the exec_fd is still open. Which means we need to tell the manager
3669 * that POLLHUP on it no longer means execve() succeeded. */
3671 if (write(exec_fd
, &hot
, sizeof(hot
)) < 0) {
3672 *exit_status
= EXIT_EXEC
;
3673 return log_unit_error_errno(unit
, errno
, "Failed to disable exec_fd: %m");
3677 if (r
== -ENOENT
&& (command
->flags
& EXEC_COMMAND_IGNORE_FAILURE
)) {
3678 log_struct_errno(LOG_INFO
, r
,
3679 "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR
,
3681 LOG_UNIT_INVOCATION_ID(unit
),
3682 LOG_UNIT_MESSAGE(unit
, "Executable %s missing, skipping: %m",
3684 "EXECUTABLE=%s", command
->path
);
3688 *exit_status
= EXIT_EXEC
;
3689 return log_unit_error_errno(unit
, r
, "Failed to execute command: %m");
3692 static int exec_context_load_environment(const Unit
*unit
, const ExecContext
*c
, char ***l
);
3693 static int exec_context_named_iofds(const ExecContext
*c
, const ExecParameters
*p
, int named_iofds
[3]);
3695 int exec_spawn(Unit
*unit
,
3696 ExecCommand
*command
,
3697 const ExecContext
*context
,
3698 const ExecParameters
*params
,
3699 ExecRuntime
*runtime
,
3700 DynamicCreds
*dcreds
,
3703 int socket_fd
, r
, named_iofds
[3] = { -1, -1, -1 }, *fds
= NULL
;
3704 _cleanup_free_
char *subcgroup_path
= NULL
;
3705 _cleanup_strv_free_
char **files_env
= NULL
;
3706 size_t n_storage_fds
= 0, n_socket_fds
= 0;
3707 _cleanup_free_
char *line
= NULL
;
3715 assert(params
->fds
|| (params
->n_socket_fds
+ params
->n_storage_fds
<= 0));
3717 if (context
->std_input
== EXEC_INPUT_SOCKET
||
3718 context
->std_output
== EXEC_OUTPUT_SOCKET
||
3719 context
->std_error
== EXEC_OUTPUT_SOCKET
) {
3721 if (params
->n_socket_fds
> 1) {
3722 log_unit_error(unit
, "Got more than one socket.");
3726 if (params
->n_socket_fds
== 0) {
3727 log_unit_error(unit
, "Got no socket.");
3731 socket_fd
= params
->fds
[0];
3735 n_socket_fds
= params
->n_socket_fds
;
3736 n_storage_fds
= params
->n_storage_fds
;
3739 r
= exec_context_named_iofds(context
, params
, named_iofds
);
3741 return log_unit_error_errno(unit
, r
, "Failed to load a named file descriptor: %m");
3743 r
= exec_context_load_environment(unit
, context
, &files_env
);
3745 return log_unit_error_errno(unit
, r
, "Failed to load environment files: %m");
3747 line
= exec_command_line(command
->argv
);
3751 log_struct(LOG_DEBUG
,
3752 LOG_UNIT_MESSAGE(unit
, "About to execute: %s", line
),
3753 "EXECUTABLE=%s", command
->path
,
3755 LOG_UNIT_INVOCATION_ID(unit
));
3757 if (params
->cgroup_path
) {
3758 r
= exec_parameters_get_cgroup_path(params
, &subcgroup_path
);
3760 return log_unit_error_errno(unit
, r
, "Failed to acquire subcgroup path: %m");
3761 if (r
> 0) { /* We are using a child cgroup */
3762 r
= cg_create(SYSTEMD_CGROUP_CONTROLLER
, subcgroup_path
);
3764 return log_unit_error_errno(unit
, r
, "Failed to create control group '%s': %m", subcgroup_path
);
3770 return log_unit_error_errno(unit
, errno
, "Failed to fork: %m");
3773 int exit_status
= EXIT_SUCCESS
;
3775 r
= exec_child(unit
,
3787 unit
->manager
->user_lookup_fds
[1],
3791 log_struct_errno(LOG_ERR
, r
,
3792 "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR
,
3794 LOG_UNIT_INVOCATION_ID(unit
),
3795 LOG_UNIT_MESSAGE(unit
, "Failed at step %s spawning %s: %m",
3796 exit_status_to_string(exit_status
, EXIT_STATUS_SYSTEMD
),
3798 "EXECUTABLE=%s", command
->path
);
3803 log_unit_debug(unit
, "Forked %s as "PID_FMT
, command
->path
, pid
);
3805 /* We add the new process to the cgroup both in the child (so that we can be sure that no user code is ever
3806 * executed outside of the cgroup) and in the parent (so that we can be sure that when we kill the cgroup the
3807 * process will be killed too). */
3809 (void) cg_attach(SYSTEMD_CGROUP_CONTROLLER
, subcgroup_path
, pid
);
3811 exec_status_start(&command
->exec_status
, pid
);
3817 void exec_context_init(ExecContext
*c
) {
3818 ExecDirectoryType i
;
3823 c
->ioprio
= IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE
, 0);
3824 c
->cpu_sched_policy
= SCHED_OTHER
;
3825 c
->syslog_priority
= LOG_DAEMON
|LOG_INFO
;
3826 c
->syslog_level_prefix
= true;
3827 c
->ignore_sigpipe
= true;
3828 c
->timer_slack_nsec
= NSEC_INFINITY
;
3829 c
->personality
= PERSONALITY_INVALID
;
3830 for (i
= 0; i
< _EXEC_DIRECTORY_TYPE_MAX
; i
++)
3831 c
->directories
[i
].mode
= 0755;
3832 c
->capability_bounding_set
= CAP_ALL
;
3833 assert_cc(NAMESPACE_FLAGS_INITIAL
!= NAMESPACE_FLAGS_ALL
);
3834 c
->restrict_namespaces
= NAMESPACE_FLAGS_INITIAL
;
3835 c
->log_level_max
= -1;
3838 void exec_context_done(ExecContext
*c
) {
3839 ExecDirectoryType i
;
3844 c
->environment
= strv_free(c
->environment
);
3845 c
->environment_files
= strv_free(c
->environment_files
);
3846 c
->pass_environment
= strv_free(c
->pass_environment
);
3847 c
->unset_environment
= strv_free(c
->unset_environment
);
3849 rlimit_free_all(c
->rlimit
);
3851 for (l
= 0; l
< 3; l
++) {
3852 c
->stdio_fdname
[l
] = mfree(c
->stdio_fdname
[l
]);
3853 c
->stdio_file
[l
] = mfree(c
->stdio_file
[l
]);
3856 c
->working_directory
= mfree(c
->working_directory
);
3857 c
->root_directory
= mfree(c
->root_directory
);
3858 c
->root_image
= mfree(c
->root_image
);
3859 c
->tty_path
= mfree(c
->tty_path
);
3860 c
->syslog_identifier
= mfree(c
->syslog_identifier
);
3861 c
->user
= mfree(c
->user
);
3862 c
->group
= mfree(c
->group
);
3864 c
->supplementary_groups
= strv_free(c
->supplementary_groups
);
3866 c
->pam_name
= mfree(c
->pam_name
);
3868 c
->read_only_paths
= strv_free(c
->read_only_paths
);
3869 c
->read_write_paths
= strv_free(c
->read_write_paths
);
3870 c
->inaccessible_paths
= strv_free(c
->inaccessible_paths
);
3872 bind_mount_free_many(c
->bind_mounts
, c
->n_bind_mounts
);
3873 c
->bind_mounts
= NULL
;
3874 c
->n_bind_mounts
= 0;
3875 temporary_filesystem_free_many(c
->temporary_filesystems
, c
->n_temporary_filesystems
);
3876 c
->temporary_filesystems
= NULL
;
3877 c
->n_temporary_filesystems
= 0;
3879 c
->cpuset
= cpu_set_mfree(c
->cpuset
);
3881 c
->utmp_id
= mfree(c
->utmp_id
);
3882 c
->selinux_context
= mfree(c
->selinux_context
);
3883 c
->apparmor_profile
= mfree(c
->apparmor_profile
);
3884 c
->smack_process_label
= mfree(c
->smack_process_label
);
3886 c
->syscall_filter
= hashmap_free(c
->syscall_filter
);
3887 c
->syscall_archs
= set_free(c
->syscall_archs
);
3888 c
->address_families
= set_free(c
->address_families
);
3890 for (i
= 0; i
< _EXEC_DIRECTORY_TYPE_MAX
; i
++)
3891 c
->directories
[i
].paths
= strv_free(c
->directories
[i
].paths
);
3893 c
->log_level_max
= -1;
3895 exec_context_free_log_extra_fields(c
);
3897 c
->log_rate_limit_interval_usec
= 0;
3898 c
->log_rate_limit_burst
= 0;
3900 c
->stdin_data
= mfree(c
->stdin_data
);
3901 c
->stdin_data_size
= 0;
3903 c
->network_namespace_path
= mfree(c
->network_namespace_path
);
3906 int exec_context_destroy_runtime_directory(const ExecContext
*c
, const char *runtime_prefix
) {
3911 if (!runtime_prefix
)
3914 STRV_FOREACH(i
, c
->directories
[EXEC_DIRECTORY_RUNTIME
].paths
) {
3915 _cleanup_free_
char *p
;
3917 p
= path_join(runtime_prefix
, *i
);
3921 /* We execute this synchronously, since we need to be sure this is gone when we start the
3923 (void) rm_rf(p
, REMOVE_ROOT
);
3929 static void exec_command_done(ExecCommand
*c
) {
3932 c
->path
= mfree(c
->path
);
3933 c
->argv
= strv_free(c
->argv
);
3936 void exec_command_done_array(ExecCommand
*c
, size_t n
) {
3939 for (i
= 0; i
< n
; i
++)
3940 exec_command_done(c
+i
);
3943 ExecCommand
* exec_command_free_list(ExecCommand
*c
) {
3947 LIST_REMOVE(command
, c
, i
);
3948 exec_command_done(i
);
3955 void exec_command_free_array(ExecCommand
**c
, size_t n
) {
3958 for (i
= 0; i
< n
; i
++)
3959 c
[i
] = exec_command_free_list(c
[i
]);
3962 void exec_command_reset_status_array(ExecCommand
*c
, size_t n
) {
3965 for (i
= 0; i
< n
; i
++)
3966 exec_status_reset(&c
[i
].exec_status
);
3969 void exec_command_reset_status_list_array(ExecCommand
**c
, size_t n
) {
3972 for (i
= 0; i
< n
; i
++) {
3975 LIST_FOREACH(command
, z
, c
[i
])
3976 exec_status_reset(&z
->exec_status
);
3980 typedef struct InvalidEnvInfo
{
3985 static void invalid_env(const char *p
, void *userdata
) {
3986 InvalidEnvInfo
*info
= userdata
;
3988 log_unit_error(info
->unit
, "Ignoring invalid environment assignment '%s': %s", p
, info
->path
);
3991 const char* exec_context_fdname(const ExecContext
*c
, int fd_index
) {
3997 if (c
->std_input
!= EXEC_INPUT_NAMED_FD
)
4000 return c
->stdio_fdname
[STDIN_FILENO
] ?: "stdin";
4003 if (c
->std_output
!= EXEC_OUTPUT_NAMED_FD
)
4006 return c
->stdio_fdname
[STDOUT_FILENO
] ?: "stdout";
4009 if (c
->std_error
!= EXEC_OUTPUT_NAMED_FD
)
4012 return c
->stdio_fdname
[STDERR_FILENO
] ?: "stderr";
4019 static int exec_context_named_iofds(const ExecContext
*c
, const ExecParameters
*p
, int named_iofds
[static 3]) {
4021 const char* stdio_fdname
[3];
4027 targets
= (c
->std_input
== EXEC_INPUT_NAMED_FD
) +
4028 (c
->std_output
== EXEC_OUTPUT_NAMED_FD
) +
4029 (c
->std_error
== EXEC_OUTPUT_NAMED_FD
);
4031 for (i
= 0; i
< 3; i
++)
4032 stdio_fdname
[i
] = exec_context_fdname(c
, i
);
4034 n_fds
= p
->n_storage_fds
+ p
->n_socket_fds
;
4036 for (i
= 0; i
< n_fds
&& targets
> 0; i
++)
4037 if (named_iofds
[STDIN_FILENO
] < 0 &&
4038 c
->std_input
== EXEC_INPUT_NAMED_FD
&&
4039 stdio_fdname
[STDIN_FILENO
] &&
4040 streq(p
->fd_names
[i
], stdio_fdname
[STDIN_FILENO
])) {
4042 named_iofds
[STDIN_FILENO
] = p
->fds
[i
];
4045 } else if (named_iofds
[STDOUT_FILENO
] < 0 &&
4046 c
->std_output
== EXEC_OUTPUT_NAMED_FD
&&
4047 stdio_fdname
[STDOUT_FILENO
] &&
4048 streq(p
->fd_names
[i
], stdio_fdname
[STDOUT_FILENO
])) {
4050 named_iofds
[STDOUT_FILENO
] = p
->fds
[i
];
4053 } else if (named_iofds
[STDERR_FILENO
] < 0 &&
4054 c
->std_error
== EXEC_OUTPUT_NAMED_FD
&&
4055 stdio_fdname
[STDERR_FILENO
] &&
4056 streq(p
->fd_names
[i
], stdio_fdname
[STDERR_FILENO
])) {
4058 named_iofds
[STDERR_FILENO
] = p
->fds
[i
];
4062 return targets
== 0 ? 0 : -ENOENT
;
4065 static int exec_context_load_environment(const Unit
*unit
, const ExecContext
*c
, char ***l
) {
4066 char **i
, **r
= NULL
;
4071 STRV_FOREACH(i
, c
->environment_files
) {
4075 bool ignore
= false;
4077 _cleanup_globfree_ glob_t pglob
= {};
4086 if (!path_is_absolute(fn
)) {
4094 /* Filename supports globbing, take all matching files */
4095 k
= safe_glob(fn
, 0, &pglob
);
4104 /* When we don't match anything, -ENOENT should be returned */
4105 assert(pglob
.gl_pathc
> 0);
4107 for (n
= 0; n
< pglob
.gl_pathc
; n
++) {
4108 k
= load_env_file(NULL
, pglob
.gl_pathv
[n
], &p
);
4116 /* Log invalid environment variables with filename */
4118 InvalidEnvInfo info
= {
4120 .path
= pglob
.gl_pathv
[n
]
4123 p
= strv_env_clean_with_callback(p
, invalid_env
, &info
);
4131 m
= strv_env_merge(2, r
, p
);
4147 static bool tty_may_match_dev_console(const char *tty
) {
4148 _cleanup_free_
char *resolved
= NULL
;
4153 tty
= skip_dev_prefix(tty
);
4155 /* trivial identity? */
4156 if (streq(tty
, "console"))
4159 if (resolve_dev_console(&resolved
) < 0)
4160 return true; /* if we could not resolve, assume it may */
4162 /* "tty0" means the active VC, so it may be the same sometimes */
4163 return path_equal(resolved
, tty
) || (streq(resolved
, "tty0") && tty_is_vc(tty
));
4166 static bool exec_context_may_touch_tty(const ExecContext
*ec
) {
4169 return ec
->tty_reset
||
4171 ec
->tty_vt_disallocate
||
4172 is_terminal_input(ec
->std_input
) ||
4173 is_terminal_output(ec
->std_output
) ||
4174 is_terminal_output(ec
->std_error
);
4177 bool exec_context_may_touch_console(const ExecContext
*ec
) {
4179 return exec_context_may_touch_tty(ec
) &&
4180 tty_may_match_dev_console(exec_context_tty_path(ec
));
4183 static void strv_fprintf(FILE *f
, char **l
) {
4189 fprintf(f
, " %s", *g
);
4192 void exec_context_dump(const ExecContext
*c
, FILE* f
, const char *prefix
) {
4193 ExecDirectoryType dt
;
4201 prefix
= strempty(prefix
);
4205 "%sWorkingDirectory: %s\n"
4206 "%sRootDirectory: %s\n"
4207 "%sNonBlocking: %s\n"
4208 "%sPrivateTmp: %s\n"
4209 "%sPrivateDevices: %s\n"
4210 "%sProtectKernelTunables: %s\n"
4211 "%sProtectKernelModules: %s\n"
4212 "%sProtectControlGroups: %s\n"
4213 "%sPrivateNetwork: %s\n"
4214 "%sPrivateUsers: %s\n"
4215 "%sProtectHome: %s\n"
4216 "%sProtectSystem: %s\n"
4217 "%sMountAPIVFS: %s\n"
4218 "%sIgnoreSIGPIPE: %s\n"
4219 "%sMemoryDenyWriteExecute: %s\n"
4220 "%sRestrictRealtime: %s\n"
4221 "%sKeyringMode: %s\n"
4222 "%sProtectHostname: %s\n",
4224 prefix
, c
->working_directory
? c
->working_directory
: "/",
4225 prefix
, c
->root_directory
? c
->root_directory
: "/",
4226 prefix
, yes_no(c
->non_blocking
),
4227 prefix
, yes_no(c
->private_tmp
),
4228 prefix
, yes_no(c
->private_devices
),
4229 prefix
, yes_no(c
->protect_kernel_tunables
),
4230 prefix
, yes_no(c
->protect_kernel_modules
),
4231 prefix
, yes_no(c
->protect_control_groups
),
4232 prefix
, yes_no(c
->private_network
),
4233 prefix
, yes_no(c
->private_users
),
4234 prefix
, protect_home_to_string(c
->protect_home
),
4235 prefix
, protect_system_to_string(c
->protect_system
),
4236 prefix
, yes_no(c
->mount_apivfs
),
4237 prefix
, yes_no(c
->ignore_sigpipe
),
4238 prefix
, yes_no(c
->memory_deny_write_execute
),
4239 prefix
, yes_no(c
->restrict_realtime
),
4240 prefix
, exec_keyring_mode_to_string(c
->keyring_mode
),
4241 prefix
, yes_no(c
->protect_hostname
));
4244 fprintf(f
, "%sRootImage: %s\n", prefix
, c
->root_image
);
4246 STRV_FOREACH(e
, c
->environment
)
4247 fprintf(f
, "%sEnvironment: %s\n", prefix
, *e
);
4249 STRV_FOREACH(e
, c
->environment_files
)
4250 fprintf(f
, "%sEnvironmentFile: %s\n", prefix
, *e
);
4252 STRV_FOREACH(e
, c
->pass_environment
)
4253 fprintf(f
, "%sPassEnvironment: %s\n", prefix
, *e
);
4255 STRV_FOREACH(e
, c
->unset_environment
)
4256 fprintf(f
, "%sUnsetEnvironment: %s\n", prefix
, *e
);
4258 fprintf(f
, "%sRuntimeDirectoryPreserve: %s\n", prefix
, exec_preserve_mode_to_string(c
->runtime_directory_preserve_mode
));
4260 for (dt
= 0; dt
< _EXEC_DIRECTORY_TYPE_MAX
; dt
++) {
4261 fprintf(f
, "%s%sMode: %04o\n", prefix
, exec_directory_type_to_string(dt
), c
->directories
[dt
].mode
);
4263 STRV_FOREACH(d
, c
->directories
[dt
].paths
)
4264 fprintf(f
, "%s%s: %s\n", prefix
, exec_directory_type_to_string(dt
), *d
);
4272 if (c
->oom_score_adjust_set
)
4274 "%sOOMScoreAdjust: %i\n",
4275 prefix
, c
->oom_score_adjust
);
4277 for (i
= 0; i
< RLIM_NLIMITS
; i
++)
4279 fprintf(f
, "%sLimit%s: " RLIM_FMT
"\n",
4280 prefix
, rlimit_to_string(i
), c
->rlimit
[i
]->rlim_max
);
4281 fprintf(f
, "%sLimit%sSoft: " RLIM_FMT
"\n",
4282 prefix
, rlimit_to_string(i
), c
->rlimit
[i
]->rlim_cur
);
4285 if (c
->ioprio_set
) {
4286 _cleanup_free_
char *class_str
= NULL
;
4288 r
= ioprio_class_to_string_alloc(IOPRIO_PRIO_CLASS(c
->ioprio
), &class_str
);
4290 fprintf(f
, "%sIOSchedulingClass: %s\n", prefix
, class_str
);
4292 fprintf(f
, "%sIOPriority: %lu\n", prefix
, IOPRIO_PRIO_DATA(c
->ioprio
));
4295 if (c
->cpu_sched_set
) {
4296 _cleanup_free_
char *policy_str
= NULL
;
4298 r
= sched_policy_to_string_alloc(c
->cpu_sched_policy
, &policy_str
);
4300 fprintf(f
, "%sCPUSchedulingPolicy: %s\n", prefix
, policy_str
);
4303 "%sCPUSchedulingPriority: %i\n"
4304 "%sCPUSchedulingResetOnFork: %s\n",
4305 prefix
, c
->cpu_sched_priority
,
4306 prefix
, yes_no(c
->cpu_sched_reset_on_fork
));
4310 fprintf(f
, "%sCPUAffinity:", prefix
);
4311 for (i
= 0; i
< c
->cpuset_ncpus
; i
++)
4312 if (CPU_ISSET_S(i
, CPU_ALLOC_SIZE(c
->cpuset_ncpus
), c
->cpuset
))
4313 fprintf(f
, " %u", i
);
4317 if (c
->timer_slack_nsec
!= NSEC_INFINITY
)
4318 fprintf(f
, "%sTimerSlackNSec: "NSEC_FMT
"\n", prefix
, c
->timer_slack_nsec
);
4321 "%sStandardInput: %s\n"
4322 "%sStandardOutput: %s\n"
4323 "%sStandardError: %s\n",
4324 prefix
, exec_input_to_string(c
->std_input
),
4325 prefix
, exec_output_to_string(c
->std_output
),
4326 prefix
, exec_output_to_string(c
->std_error
));
4328 if (c
->std_input
== EXEC_INPUT_NAMED_FD
)
4329 fprintf(f
, "%sStandardInputFileDescriptorName: %s\n", prefix
, c
->stdio_fdname
[STDIN_FILENO
]);
4330 if (c
->std_output
== EXEC_OUTPUT_NAMED_FD
)
4331 fprintf(f
, "%sStandardOutputFileDescriptorName: %s\n", prefix
, c
->stdio_fdname
[STDOUT_FILENO
]);
4332 if (c
->std_error
== EXEC_OUTPUT_NAMED_FD
)
4333 fprintf(f
, "%sStandardErrorFileDescriptorName: %s\n", prefix
, c
->stdio_fdname
[STDERR_FILENO
]);
4335 if (c
->std_input
== EXEC_INPUT_FILE
)
4336 fprintf(f
, "%sStandardInputFile: %s\n", prefix
, c
->stdio_file
[STDIN_FILENO
]);
4337 if (c
->std_output
== EXEC_OUTPUT_FILE
)
4338 fprintf(f
, "%sStandardOutputFile: %s\n", prefix
, c
->stdio_file
[STDOUT_FILENO
]);
4339 if (c
->std_output
== EXEC_OUTPUT_FILE_APPEND
)
4340 fprintf(f
, "%sStandardOutputFileToAppend: %s\n", prefix
, c
->stdio_file
[STDOUT_FILENO
]);
4341 if (c
->std_error
== EXEC_OUTPUT_FILE
)
4342 fprintf(f
, "%sStandardErrorFile: %s\n", prefix
, c
->stdio_file
[STDERR_FILENO
]);
4343 if (c
->std_error
== EXEC_OUTPUT_FILE_APPEND
)
4344 fprintf(f
, "%sStandardErrorFileToAppend: %s\n", prefix
, c
->stdio_file
[STDERR_FILENO
]);
4350 "%sTTYVHangup: %s\n"
4351 "%sTTYVTDisallocate: %s\n",
4352 prefix
, c
->tty_path
,
4353 prefix
, yes_no(c
->tty_reset
),
4354 prefix
, yes_no(c
->tty_vhangup
),
4355 prefix
, yes_no(c
->tty_vt_disallocate
));
4357 if (IN_SET(c
->std_output
,
4360 EXEC_OUTPUT_JOURNAL
,
4361 EXEC_OUTPUT_SYSLOG_AND_CONSOLE
,
4362 EXEC_OUTPUT_KMSG_AND_CONSOLE
,
4363 EXEC_OUTPUT_JOURNAL_AND_CONSOLE
) ||
4364 IN_SET(c
->std_error
,
4367 EXEC_OUTPUT_JOURNAL
,
4368 EXEC_OUTPUT_SYSLOG_AND_CONSOLE
,
4369 EXEC_OUTPUT_KMSG_AND_CONSOLE
,
4370 EXEC_OUTPUT_JOURNAL_AND_CONSOLE
)) {
4372 _cleanup_free_
char *fac_str
= NULL
, *lvl_str
= NULL
;
4374 r
= log_facility_unshifted_to_string_alloc(c
->syslog_priority
>> 3, &fac_str
);
4376 fprintf(f
, "%sSyslogFacility: %s\n", prefix
, fac_str
);
4378 r
= log_level_to_string_alloc(LOG_PRI(c
->syslog_priority
), &lvl_str
);
4380 fprintf(f
, "%sSyslogLevel: %s\n", prefix
, lvl_str
);
4383 if (c
->log_level_max
>= 0) {
4384 _cleanup_free_
char *t
= NULL
;
4386 (void) log_level_to_string_alloc(c
->log_level_max
, &t
);
4388 fprintf(f
, "%sLogLevelMax: %s\n", prefix
, strna(t
));
4391 if (c
->log_rate_limit_interval_usec
> 0) {
4392 char buf_timespan
[FORMAT_TIMESPAN_MAX
];
4395 "%sLogRateLimitIntervalSec: %s\n",
4396 prefix
, format_timespan(buf_timespan
, sizeof(buf_timespan
), c
->log_rate_limit_interval_usec
, USEC_PER_SEC
));
4399 if (c
->log_rate_limit_burst
> 0)
4400 fprintf(f
, "%sLogRateLimitBurst: %u\n", prefix
, c
->log_rate_limit_burst
);
4402 if (c
->n_log_extra_fields
> 0) {
4405 for (j
= 0; j
< c
->n_log_extra_fields
; j
++) {
4406 fprintf(f
, "%sLogExtraFields: ", prefix
);
4407 fwrite(c
->log_extra_fields
[j
].iov_base
,
4408 1, c
->log_extra_fields
[j
].iov_len
,
4414 if (c
->secure_bits
) {
4415 _cleanup_free_
char *str
= NULL
;
4417 r
= secure_bits_to_string_alloc(c
->secure_bits
, &str
);
4419 fprintf(f
, "%sSecure Bits: %s\n", prefix
, str
);
4422 if (c
->capability_bounding_set
!= CAP_ALL
) {
4423 _cleanup_free_
char *str
= NULL
;
4425 r
= capability_set_to_string_alloc(c
->capability_bounding_set
, &str
);
4427 fprintf(f
, "%sCapabilityBoundingSet: %s\n", prefix
, str
);
4430 if (c
->capability_ambient_set
!= 0) {
4431 _cleanup_free_
char *str
= NULL
;
4433 r
= capability_set_to_string_alloc(c
->capability_ambient_set
, &str
);
4435 fprintf(f
, "%sAmbientCapabilities: %s\n", prefix
, str
);
4439 fprintf(f
, "%sUser: %s\n", prefix
, c
->user
);
4441 fprintf(f
, "%sGroup: %s\n", prefix
, c
->group
);
4443 fprintf(f
, "%sDynamicUser: %s\n", prefix
, yes_no(c
->dynamic_user
));
4445 if (!strv_isempty(c
->supplementary_groups
)) {
4446 fprintf(f
, "%sSupplementaryGroups:", prefix
);
4447 strv_fprintf(f
, c
->supplementary_groups
);
4452 fprintf(f
, "%sPAMName: %s\n", prefix
, c
->pam_name
);
4454 if (!strv_isempty(c
->read_write_paths
)) {
4455 fprintf(f
, "%sReadWritePaths:", prefix
);
4456 strv_fprintf(f
, c
->read_write_paths
);
4460 if (!strv_isempty(c
->read_only_paths
)) {
4461 fprintf(f
, "%sReadOnlyPaths:", prefix
);
4462 strv_fprintf(f
, c
->read_only_paths
);
4466 if (!strv_isempty(c
->inaccessible_paths
)) {
4467 fprintf(f
, "%sInaccessiblePaths:", prefix
);
4468 strv_fprintf(f
, c
->inaccessible_paths
);
4472 if (c
->n_bind_mounts
> 0)
4473 for (i
= 0; i
< c
->n_bind_mounts
; i
++)
4474 fprintf(f
, "%s%s: %s%s:%s:%s\n", prefix
,
4475 c
->bind_mounts
[i
].read_only
? "BindReadOnlyPaths" : "BindPaths",
4476 c
->bind_mounts
[i
].ignore_enoent
? "-": "",
4477 c
->bind_mounts
[i
].source
,
4478 c
->bind_mounts
[i
].destination
,
4479 c
->bind_mounts
[i
].recursive
? "rbind" : "norbind");
4481 if (c
->n_temporary_filesystems
> 0)
4482 for (i
= 0; i
< c
->n_temporary_filesystems
; i
++) {
4483 TemporaryFileSystem
*t
= c
->temporary_filesystems
+ i
;
4485 fprintf(f
, "%sTemporaryFileSystem: %s%s%s\n", prefix
,
4487 isempty(t
->options
) ? "" : ":",
4488 strempty(t
->options
));
4493 "%sUtmpIdentifier: %s\n",
4494 prefix
, c
->utmp_id
);
4496 if (c
->selinux_context
)
4498 "%sSELinuxContext: %s%s\n",
4499 prefix
, c
->selinux_context_ignore
? "-" : "", c
->selinux_context
);
4501 if (c
->apparmor_profile
)
4503 "%sAppArmorProfile: %s%s\n",
4504 prefix
, c
->apparmor_profile_ignore
? "-" : "", c
->apparmor_profile
);
4506 if (c
->smack_process_label
)
4508 "%sSmackProcessLabel: %s%s\n",
4509 prefix
, c
->smack_process_label_ignore
? "-" : "", c
->smack_process_label
);
4511 if (c
->personality
!= PERSONALITY_INVALID
)
4513 "%sPersonality: %s\n",
4514 prefix
, strna(personality_to_string(c
->personality
)));
4517 "%sLockPersonality: %s\n",
4518 prefix
, yes_no(c
->lock_personality
));
4520 if (c
->syscall_filter
) {
4528 "%sSystemCallFilter: ",
4531 if (!c
->syscall_whitelist
)
4535 HASHMAP_FOREACH_KEY(val
, id
, c
->syscall_filter
, j
) {
4536 _cleanup_free_
char *name
= NULL
;
4537 const char *errno_name
= NULL
;
4538 int num
= PTR_TO_INT(val
);
4545 name
= seccomp_syscall_resolve_num_arch(SCMP_ARCH_NATIVE
, PTR_TO_INT(id
) - 1);
4546 fputs(strna(name
), f
);
4549 errno_name
= errno_to_name(num
);
4551 fprintf(f
, ":%s", errno_name
);
4553 fprintf(f
, ":%d", num
);
4561 if (c
->syscall_archs
) {
4568 "%sSystemCallArchitectures:",
4572 SET_FOREACH(id
, c
->syscall_archs
, j
)
4573 fprintf(f
, " %s", strna(seccomp_arch_to_string(PTR_TO_UINT32(id
) - 1)));
4578 if (exec_context_restrict_namespaces_set(c
)) {
4579 _cleanup_free_
char *s
= NULL
;
4581 r
= namespace_flags_to_string(c
->restrict_namespaces
, &s
);
4583 fprintf(f
, "%sRestrictNamespaces: %s\n",
4587 if (c
->network_namespace_path
)
4589 "%sNetworkNamespacePath: %s\n",
4590 prefix
, c
->network_namespace_path
);
4592 if (c
->syscall_errno
> 0) {
4593 const char *errno_name
;
4595 fprintf(f
, "%sSystemCallErrorNumber: ", prefix
);
4597 errno_name
= errno_to_name(c
->syscall_errno
);
4599 fprintf(f
, "%s\n", errno_name
);
4601 fprintf(f
, "%d\n", c
->syscall_errno
);
4605 bool exec_context_maintains_privileges(const ExecContext
*c
) {
4608 /* Returns true if the process forked off would run under
4609 * an unchanged UID or as root. */
4614 if (streq(c
->user
, "root") || streq(c
->user
, "0"))
4620 int exec_context_get_effective_ioprio(const ExecContext
*c
) {
4628 p
= ioprio_get(IOPRIO_WHO_PROCESS
, 0);
4630 return IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE
, 4);
4635 void exec_context_free_log_extra_fields(ExecContext
*c
) {
4640 for (l
= 0; l
< c
->n_log_extra_fields
; l
++)
4641 free(c
->log_extra_fields
[l
].iov_base
);
4642 c
->log_extra_fields
= mfree(c
->log_extra_fields
);
4643 c
->n_log_extra_fields
= 0;
4646 void exec_context_revert_tty(ExecContext
*c
) {
4651 /* First, reset the TTY (possibly kicking everybody else from the TTY) */
4652 exec_context_tty_reset(c
, NULL
);
4654 /* And then undo what chown_terminal() did earlier. Note that we only do this if we have a path
4655 * configured. If the TTY was passed to us as file descriptor we assume the TTY is opened and managed
4656 * by whoever passed it to us and thus knows better when and how to chmod()/chown() it back. */
4658 if (exec_context_may_touch_tty(c
)) {
4661 path
= exec_context_tty_path(c
);
4663 r
= chmod_and_chown(path
, TTY_MODE
, 0, TTY_GID
);
4664 if (r
< 0 && r
!= -ENOENT
)
4665 log_warning_errno(r
, "Failed to reset TTY ownership/access mode of %s, ignoring: %m", path
);
4670 void exec_status_start(ExecStatus
*s
, pid_t pid
) {
4677 dual_timestamp_get(&s
->start_timestamp
);
4680 void exec_status_exit(ExecStatus
*s
, const ExecContext
*context
, pid_t pid
, int code
, int status
) {
4683 if (s
->pid
!= pid
) {
4689 dual_timestamp_get(&s
->exit_timestamp
);
4694 if (context
&& context
->utmp_id
)
4695 (void) utmp_put_dead_process(context
->utmp_id
, pid
, code
, status
);
4698 void exec_status_reset(ExecStatus
*s
) {
4701 *s
= (ExecStatus
) {};
4704 void exec_status_dump(const ExecStatus
*s
, FILE *f
, const char *prefix
) {
4705 char buf
[FORMAT_TIMESTAMP_MAX
];
4713 prefix
= strempty(prefix
);
4716 "%sPID: "PID_FMT
"\n",
4719 if (dual_timestamp_is_set(&s
->start_timestamp
))
4721 "%sStart Timestamp: %s\n",
4722 prefix
, format_timestamp(buf
, sizeof(buf
), s
->start_timestamp
.realtime
));
4724 if (dual_timestamp_is_set(&s
->exit_timestamp
))
4726 "%sExit Timestamp: %s\n"
4728 "%sExit Status: %i\n",
4729 prefix
, format_timestamp(buf
, sizeof(buf
), s
->exit_timestamp
.realtime
),
4730 prefix
, sigchld_code_to_string(s
->code
),
4734 static char *exec_command_line(char **argv
) {
4742 STRV_FOREACH(a
, argv
)
4750 STRV_FOREACH(a
, argv
) {
4757 if (strpbrk(*a
, WHITESPACE
)) {
4768 /* FIXME: this doesn't really handle arguments that have
4769 * spaces and ticks in them */
4774 static void exec_command_dump(ExecCommand
*c
, FILE *f
, const char *prefix
) {
4775 _cleanup_free_
char *cmd
= NULL
;
4776 const char *prefix2
;
4781 prefix
= strempty(prefix
);
4782 prefix2
= strjoina(prefix
, "\t");
4784 cmd
= exec_command_line(c
->argv
);
4786 "%sCommand Line: %s\n",
4787 prefix
, cmd
? cmd
: strerror(ENOMEM
));
4789 exec_status_dump(&c
->exec_status
, f
, prefix2
);
4792 void exec_command_dump_list(ExecCommand
*c
, FILE *f
, const char *prefix
) {
4795 prefix
= strempty(prefix
);
4797 LIST_FOREACH(command
, c
, c
)
4798 exec_command_dump(c
, f
, prefix
);
4801 void exec_command_append_list(ExecCommand
**l
, ExecCommand
*e
) {
4808 /* It's kind of important, that we keep the order here */
4809 LIST_FIND_TAIL(command
, *l
, end
);
4810 LIST_INSERT_AFTER(command
, *l
, end
, e
);
4815 int exec_command_set(ExecCommand
*c
, const char *path
, ...) {
4823 l
= strv_new_ap(path
, ap
);
4835 free_and_replace(c
->path
, p
);
4837 return strv_free_and_replace(c
->argv
, l
);
4840 int exec_command_append(ExecCommand
*c
, const char *path
, ...) {
4841 _cleanup_strv_free_
char **l
= NULL
;
4849 l
= strv_new_ap(path
, ap
);
4855 r
= strv_extend_strv(&c
->argv
, l
, false);
4862 static void *remove_tmpdir_thread(void *p
) {
4863 _cleanup_free_
char *path
= p
;
4865 (void) rm_rf(path
, REMOVE_ROOT
|REMOVE_PHYSICAL
);
4869 static ExecRuntime
* exec_runtime_free(ExecRuntime
*rt
, bool destroy
) {
4876 (void) hashmap_remove(rt
->manager
->exec_runtime_by_id
, rt
->id
);
4878 /* When destroy is true, then rm_rf tmp_dir and var_tmp_dir. */
4879 if (destroy
&& rt
->tmp_dir
) {
4880 log_debug("Spawning thread to nuke %s", rt
->tmp_dir
);
4882 r
= asynchronous_job(remove_tmpdir_thread
, rt
->tmp_dir
);
4884 log_warning_errno(r
, "Failed to nuke %s: %m", rt
->tmp_dir
);
4891 if (destroy
&& rt
->var_tmp_dir
) {
4892 log_debug("Spawning thread to nuke %s", rt
->var_tmp_dir
);
4894 r
= asynchronous_job(remove_tmpdir_thread
, rt
->var_tmp_dir
);
4896 log_warning_errno(r
, "Failed to nuke %s: %m", rt
->var_tmp_dir
);
4897 free(rt
->var_tmp_dir
);
4900 rt
->var_tmp_dir
= NULL
;
4903 rt
->id
= mfree(rt
->id
);
4904 rt
->tmp_dir
= mfree(rt
->tmp_dir
);
4905 rt
->var_tmp_dir
= mfree(rt
->var_tmp_dir
);
4906 safe_close_pair(rt
->netns_storage_socket
);
4910 static void exec_runtime_freep(ExecRuntime
**rt
) {
4911 (void) exec_runtime_free(*rt
, false);
4914 static int exec_runtime_allocate(ExecRuntime
**ret
) {
4919 n
= new(ExecRuntime
, 1);
4923 *n
= (ExecRuntime
) {
4924 .netns_storage_socket
= { -1, -1 },
4931 static int exec_runtime_add(
4934 const char *tmp_dir
,
4935 const char *var_tmp_dir
,
4936 const int netns_storage_socket
[2],
4937 ExecRuntime
**ret
) {
4939 _cleanup_(exec_runtime_freep
) ExecRuntime
*rt
= NULL
;
4945 r
= hashmap_ensure_allocated(&m
->exec_runtime_by_id
, &string_hash_ops
);
4949 r
= exec_runtime_allocate(&rt
);
4953 rt
->id
= strdup(id
);
4958 rt
->tmp_dir
= strdup(tmp_dir
);
4962 /* When tmp_dir is set, then we require var_tmp_dir is also set. */
4963 assert(var_tmp_dir
);
4964 rt
->var_tmp_dir
= strdup(var_tmp_dir
);
4965 if (!rt
->var_tmp_dir
)
4969 if (netns_storage_socket
) {
4970 rt
->netns_storage_socket
[0] = netns_storage_socket
[0];
4971 rt
->netns_storage_socket
[1] = netns_storage_socket
[1];
4974 r
= hashmap_put(m
->exec_runtime_by_id
, rt
->id
, rt
);
4983 /* do not remove created ExecRuntime object when the operation succeeds. */
4988 static int exec_runtime_make(Manager
*m
, const ExecContext
*c
, const char *id
, ExecRuntime
**ret
) {
4989 _cleanup_free_
char *tmp_dir
= NULL
, *var_tmp_dir
= NULL
;
4990 _cleanup_close_pair_
int netns_storage_socket
[2] = { -1, -1 };
4997 /* It is not necessary to create ExecRuntime object. */
4998 if (!c
->private_network
&& !c
->private_tmp
&& !c
->network_namespace_path
)
5001 if (c
->private_tmp
) {
5002 r
= setup_tmp_dirs(id
, &tmp_dir
, &var_tmp_dir
);
5007 if (c
->private_network
|| c
->network_namespace_path
) {
5008 if (socketpair(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
, 0, netns_storage_socket
) < 0)
5012 r
= exec_runtime_add(m
, id
, tmp_dir
, var_tmp_dir
, netns_storage_socket
, ret
);
5017 netns_storage_socket
[0] = netns_storage_socket
[1] = -1;
5021 int exec_runtime_acquire(Manager
*m
, const ExecContext
*c
, const char *id
, bool create
, ExecRuntime
**ret
) {
5029 rt
= hashmap_get(m
->exec_runtime_by_id
, id
);
5031 /* We already have a ExecRuntime object, let's increase the ref count and reuse it */
5037 /* If not found, then create a new object. */
5038 r
= exec_runtime_make(m
, c
, id
, &rt
);
5040 /* When r == 0, it is not necessary to create ExecRuntime object. */
5044 /* increment reference counter. */
5050 ExecRuntime
*exec_runtime_unref(ExecRuntime
*rt
, bool destroy
) {
5054 assert(rt
->n_ref
> 0);
5060 return exec_runtime_free(rt
, destroy
);
5063 int exec_runtime_serialize(const Manager
*m
, FILE *f
, FDSet
*fds
) {
5071 HASHMAP_FOREACH(rt
, m
->exec_runtime_by_id
, i
) {
5072 fprintf(f
, "exec-runtime=%s", rt
->id
);
5075 fprintf(f
, " tmp-dir=%s", rt
->tmp_dir
);
5077 if (rt
->var_tmp_dir
)
5078 fprintf(f
, " var-tmp-dir=%s", rt
->var_tmp_dir
);
5080 if (rt
->netns_storage_socket
[0] >= 0) {
5083 copy
= fdset_put_dup(fds
, rt
->netns_storage_socket
[0]);
5087 fprintf(f
, " netns-socket-0=%i", copy
);
5090 if (rt
->netns_storage_socket
[1] >= 0) {
5093 copy
= fdset_put_dup(fds
, rt
->netns_storage_socket
[1]);
5097 fprintf(f
, " netns-socket-1=%i", copy
);
5106 int exec_runtime_deserialize_compat(Unit
*u
, const char *key
, const char *value
, FDSet
*fds
) {
5107 _cleanup_(exec_runtime_freep
) ExecRuntime
*rt_create
= NULL
;
5111 /* This is for the migration from old (v237 or earlier) deserialization text.
5112 * Due to the bug #7790, this may not work with the units that use JoinsNamespaceOf=.
5113 * Even if the ExecRuntime object originally created by the other unit, we cannot judge
5114 * so or not from the serialized text, then we always creates a new object owned by this. */
5120 /* Manager manages ExecRuntime objects by the unit id.
5121 * So, we omit the serialized text when the unit does not have id (yet?)... */
5122 if (isempty(u
->id
)) {
5123 log_unit_debug(u
, "Invocation ID not found. Dropping runtime parameter.");
5127 r
= hashmap_ensure_allocated(&u
->manager
->exec_runtime_by_id
, &string_hash_ops
);
5129 log_unit_debug_errno(u
, r
, "Failed to allocate storage for runtime parameter: %m");
5133 rt
= hashmap_get(u
->manager
->exec_runtime_by_id
, u
->id
);
5135 r
= exec_runtime_allocate(&rt_create
);
5139 rt_create
->id
= strdup(u
->id
);
5146 if (streq(key
, "tmp-dir")) {
5149 copy
= strdup(value
);
5153 free_and_replace(rt
->tmp_dir
, copy
);
5155 } else if (streq(key
, "var-tmp-dir")) {
5158 copy
= strdup(value
);
5162 free_and_replace(rt
->var_tmp_dir
, copy
);
5164 } else if (streq(key
, "netns-socket-0")) {
5167 if (safe_atoi(value
, &fd
) < 0 || !fdset_contains(fds
, fd
)) {
5168 log_unit_debug(u
, "Failed to parse netns socket value: %s", value
);
5172 safe_close(rt
->netns_storage_socket
[0]);
5173 rt
->netns_storage_socket
[0] = fdset_remove(fds
, fd
);
5175 } else if (streq(key
, "netns-socket-1")) {
5178 if (safe_atoi(value
, &fd
) < 0 || !fdset_contains(fds
, fd
)) {
5179 log_unit_debug(u
, "Failed to parse netns socket value: %s", value
);
5183 safe_close(rt
->netns_storage_socket
[1]);
5184 rt
->netns_storage_socket
[1] = fdset_remove(fds
, fd
);
5188 /* If the object is newly created, then put it to the hashmap which manages ExecRuntime objects. */
5190 r
= hashmap_put(u
->manager
->exec_runtime_by_id
, rt_create
->id
, rt_create
);
5192 log_unit_debug_errno(u
, r
, "Failed to put runtime parameter to manager's storage: %m");
5196 rt_create
->manager
= u
->manager
;
5205 void exec_runtime_deserialize_one(Manager
*m
, const char *value
, FDSet
*fds
) {
5206 char *id
= NULL
, *tmp_dir
= NULL
, *var_tmp_dir
= NULL
;
5207 int r
, fd0
= -1, fd1
= -1;
5208 const char *p
, *v
= value
;
5215 n
= strcspn(v
, " ");
5216 id
= strndupa(v
, n
);
5221 v
= startswith(p
, "tmp-dir=");
5223 n
= strcspn(v
, " ");
5224 tmp_dir
= strndupa(v
, n
);
5230 v
= startswith(p
, "var-tmp-dir=");
5232 n
= strcspn(v
, " ");
5233 var_tmp_dir
= strndupa(v
, n
);
5239 v
= startswith(p
, "netns-socket-0=");
5243 n
= strcspn(v
, " ");
5244 buf
= strndupa(v
, n
);
5245 if (safe_atoi(buf
, &fd0
) < 0 || !fdset_contains(fds
, fd0
)) {
5246 log_debug("Unable to process exec-runtime netns fd specification.");
5249 fd0
= fdset_remove(fds
, fd0
);
5255 v
= startswith(p
, "netns-socket-1=");
5259 n
= strcspn(v
, " ");
5260 buf
= strndupa(v
, n
);
5261 if (safe_atoi(buf
, &fd1
) < 0 || !fdset_contains(fds
, fd1
)) {
5262 log_debug("Unable to process exec-runtime netns fd specification.");
5265 fd1
= fdset_remove(fds
, fd1
);
5270 r
= exec_runtime_add(m
, id
, tmp_dir
, var_tmp_dir
, (int[]) { fd0
, fd1
}, NULL
);
5272 log_debug_errno(r
, "Failed to add exec-runtime: %m");
5275 void exec_runtime_vacuum(Manager
*m
) {
5281 /* Free unreferenced ExecRuntime objects. This is used after manager deserialization process. */
5283 HASHMAP_FOREACH(rt
, m
->exec_runtime_by_id
, i
) {
5287 (void) exec_runtime_free(rt
, false);
5291 void exec_params_clear(ExecParameters
*p
) {
5295 strv_free(p
->environment
);
5298 static const char* const exec_input_table
[_EXEC_INPUT_MAX
] = {
5299 [EXEC_INPUT_NULL
] = "null",
5300 [EXEC_INPUT_TTY
] = "tty",
5301 [EXEC_INPUT_TTY_FORCE
] = "tty-force",
5302 [EXEC_INPUT_TTY_FAIL
] = "tty-fail",
5303 [EXEC_INPUT_SOCKET
] = "socket",
5304 [EXEC_INPUT_NAMED_FD
] = "fd",
5305 [EXEC_INPUT_DATA
] = "data",
5306 [EXEC_INPUT_FILE
] = "file",
5309 DEFINE_STRING_TABLE_LOOKUP(exec_input
, ExecInput
);
5311 static const char* const exec_output_table
[_EXEC_OUTPUT_MAX
] = {
5312 [EXEC_OUTPUT_INHERIT
] = "inherit",
5313 [EXEC_OUTPUT_NULL
] = "null",
5314 [EXEC_OUTPUT_TTY
] = "tty",
5315 [EXEC_OUTPUT_SYSLOG
] = "syslog",
5316 [EXEC_OUTPUT_SYSLOG_AND_CONSOLE
] = "syslog+console",
5317 [EXEC_OUTPUT_KMSG
] = "kmsg",
5318 [EXEC_OUTPUT_KMSG_AND_CONSOLE
] = "kmsg+console",
5319 [EXEC_OUTPUT_JOURNAL
] = "journal",
5320 [EXEC_OUTPUT_JOURNAL_AND_CONSOLE
] = "journal+console",
5321 [EXEC_OUTPUT_SOCKET
] = "socket",
5322 [EXEC_OUTPUT_NAMED_FD
] = "fd",
5323 [EXEC_OUTPUT_FILE
] = "file",
5324 [EXEC_OUTPUT_FILE_APPEND
] = "append",
5327 DEFINE_STRING_TABLE_LOOKUP(exec_output
, ExecOutput
);
5329 static const char* const exec_utmp_mode_table
[_EXEC_UTMP_MODE_MAX
] = {
5330 [EXEC_UTMP_INIT
] = "init",
5331 [EXEC_UTMP_LOGIN
] = "login",
5332 [EXEC_UTMP_USER
] = "user",
5335 DEFINE_STRING_TABLE_LOOKUP(exec_utmp_mode
, ExecUtmpMode
);
5337 static const char* const exec_preserve_mode_table
[_EXEC_PRESERVE_MODE_MAX
] = {
5338 [EXEC_PRESERVE_NO
] = "no",
5339 [EXEC_PRESERVE_YES
] = "yes",
5340 [EXEC_PRESERVE_RESTART
] = "restart",
5343 DEFINE_STRING_TABLE_LOOKUP_WITH_BOOLEAN(exec_preserve_mode
, ExecPreserveMode
, EXEC_PRESERVE_YES
);
5345 static const char* const exec_directory_type_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
5346 [EXEC_DIRECTORY_RUNTIME
] = "RuntimeDirectory",
5347 [EXEC_DIRECTORY_STATE
] = "StateDirectory",
5348 [EXEC_DIRECTORY_CACHE
] = "CacheDirectory",
5349 [EXEC_DIRECTORY_LOGS
] = "LogsDirectory",
5350 [EXEC_DIRECTORY_CONFIGURATION
] = "ConfigurationDirectory",
5353 DEFINE_STRING_TABLE_LOOKUP(exec_directory_type
, ExecDirectoryType
);
5355 static const char* const exec_directory_env_name_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
5356 [EXEC_DIRECTORY_RUNTIME
] = "RUNTIME_DIRECTORY",
5357 [EXEC_DIRECTORY_STATE
] = "STATE_DIRECTORY",
5358 [EXEC_DIRECTORY_CACHE
] = "CACHE_DIRECTORY",
5359 [EXEC_DIRECTORY_LOGS
] = "LOGS_DIRECTORY",
5360 [EXEC_DIRECTORY_CONFIGURATION
] = "CONFIGURATION_DIRECTORY",
5363 DEFINE_PRIVATE_STRING_TABLE_LOOKUP_TO_STRING(exec_directory_env_name
, ExecDirectoryType
);
5365 static const char* const exec_keyring_mode_table
[_EXEC_KEYRING_MODE_MAX
] = {
5366 [EXEC_KEYRING_INHERIT
] = "inherit",
5367 [EXEC_KEYRING_PRIVATE
] = "private",
5368 [EXEC_KEYRING_SHARED
] = "shared",
5371 DEFINE_STRING_TABLE_LOOKUP(exec_keyring_mode
, ExecKeyringMode
);