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
) {
2371 .ignore_enoent
= false,
2378 *ret_bind_mounts
= bind_mounts
;
2379 *ret_n_bind_mounts
= n
;
2380 *ret_empty_directories
= TAKE_PTR(empty_directories
);
2385 bind_mount_free_many(bind_mounts
, h
);
2389 static int apply_mount_namespace(
2391 const ExecCommand
*command
,
2392 const ExecContext
*context
,
2393 const ExecParameters
*params
,
2394 const ExecRuntime
*runtime
) {
2396 _cleanup_strv_free_
char **empty_directories
= NULL
;
2397 char *tmp
= NULL
, *var
= NULL
;
2398 const char *root_dir
= NULL
, *root_image
= NULL
;
2399 NamespaceInfo ns_info
;
2400 bool needs_sandboxing
;
2401 BindMount
*bind_mounts
= NULL
;
2402 size_t n_bind_mounts
= 0;
2407 /* The runtime struct only contains the parent of the private /tmp,
2408 * which is non-accessible to world users. Inside of it there's a /tmp
2409 * that is sticky, and that's the one we want to use here. */
2411 if (context
->private_tmp
&& runtime
) {
2412 if (runtime
->tmp_dir
)
2413 tmp
= strjoina(runtime
->tmp_dir
, "/tmp");
2414 if (runtime
->var_tmp_dir
)
2415 var
= strjoina(runtime
->var_tmp_dir
, "/tmp");
2418 if (params
->flags
& EXEC_APPLY_CHROOT
) {
2419 root_image
= context
->root_image
;
2422 root_dir
= context
->root_directory
;
2425 r
= compile_bind_mounts(context
, params
, &bind_mounts
, &n_bind_mounts
, &empty_directories
);
2429 needs_sandboxing
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && !(command
->flags
& EXEC_COMMAND_FULLY_PRIVILEGED
);
2430 if (needs_sandboxing
)
2431 ns_info
= (NamespaceInfo
) {
2432 .ignore_protect_paths
= false,
2433 .private_dev
= context
->private_devices
,
2434 .protect_control_groups
= context
->protect_control_groups
,
2435 .protect_kernel_tunables
= context
->protect_kernel_tunables
,
2436 .protect_kernel_modules
= context
->protect_kernel_modules
,
2437 .protect_hostname
= context
->protect_hostname
,
2438 .mount_apivfs
= context
->mount_apivfs
,
2439 .private_mounts
= context
->private_mounts
,
2441 else if (!context
->dynamic_user
&& root_dir
)
2443 * If DynamicUser=no and RootDirectory= is set then lets pass a relaxed
2444 * sandbox info, otherwise enforce it, don't ignore protected paths and
2445 * fail if we are enable to apply the sandbox inside the mount namespace.
2447 ns_info
= (NamespaceInfo
) {
2448 .ignore_protect_paths
= true,
2451 ns_info
= (NamespaceInfo
) {};
2453 if (context
->mount_flags
== MS_SHARED
)
2454 log_unit_debug(u
, "shared mount propagation hidden by other fs namespacing unit settings: ignoring");
2456 r
= setup_namespace(root_dir
, root_image
,
2457 &ns_info
, context
->read_write_paths
,
2458 needs_sandboxing
? context
->read_only_paths
: NULL
,
2459 needs_sandboxing
? context
->inaccessible_paths
: NULL
,
2463 context
->temporary_filesystems
,
2464 context
->n_temporary_filesystems
,
2467 needs_sandboxing
? context
->protect_home
: PROTECT_HOME_NO
,
2468 needs_sandboxing
? context
->protect_system
: PROTECT_SYSTEM_NO
,
2469 context
->mount_flags
,
2470 DISSECT_IMAGE_DISCARD_ON_LOOP
);
2472 bind_mount_free_many(bind_mounts
, n_bind_mounts
);
2474 /* If we couldn't set up the namespace this is probably due to a missing capability. setup_namespace() reports
2475 * that with a special, recognizable error ENOANO. In this case, silently proceeed, but only if exclusively
2476 * sandboxing options were used, i.e. nothing such as RootDirectory= or BindMount= that would result in a
2477 * completely different execution environment. */
2479 if (n_bind_mounts
== 0 &&
2480 context
->n_temporary_filesystems
== 0 &&
2481 !root_dir
&& !root_image
&&
2482 !context
->dynamic_user
) {
2483 log_unit_debug(u
, "Failed to set up namespace, assuming containerized execution and ignoring.");
2487 log_unit_debug(u
, "Failed to set up namespace, and refusing to continue since the selected namespacing options alter mount environment non-trivially.\n"
2488 "Bind mounts: %zu, temporary filesystems: %zu, root directory: %s, root image: %s, dynamic user: %s",
2489 n_bind_mounts
, context
->n_temporary_filesystems
, yes_no(root_dir
), yes_no(root_image
), yes_no(context
->dynamic_user
));
2497 static int apply_working_directory(
2498 const ExecContext
*context
,
2499 const ExecParameters
*params
,
2501 const bool needs_mount_ns
,
2507 assert(exit_status
);
2509 if (context
->working_directory_home
) {
2512 *exit_status
= EXIT_CHDIR
;
2518 } else if (context
->working_directory
)
2519 wd
= context
->working_directory
;
2523 if (params
->flags
& EXEC_APPLY_CHROOT
) {
2524 if (!needs_mount_ns
&& context
->root_directory
)
2525 if (chroot(context
->root_directory
) < 0) {
2526 *exit_status
= EXIT_CHROOT
;
2532 d
= prefix_roota(context
->root_directory
, wd
);
2534 if (chdir(d
) < 0 && !context
->working_directory_missing_ok
) {
2535 *exit_status
= EXIT_CHDIR
;
2542 static int setup_keyring(
2544 const ExecContext
*context
,
2545 const ExecParameters
*p
,
2546 uid_t uid
, gid_t gid
) {
2548 key_serial_t keyring
;
2557 /* Let's set up a new per-service "session" kernel keyring for each system service. This has the benefit that
2558 * each service runs with its own keyring shared among all processes of the service, but with no hook-up beyond
2559 * that scope, and in particular no link to the per-UID keyring. If we don't do this the keyring will be
2560 * automatically created on-demand and then linked to the per-UID keyring, by the kernel. The kernel's built-in
2561 * on-demand behaviour is very appropriate for login users, but probably not so much for system services, where
2562 * UIDs are not necessarily specific to a service but reused (at least in the case of UID 0). */
2564 if (context
->keyring_mode
== EXEC_KEYRING_INHERIT
)
2567 /* Acquiring a reference to the user keyring is nasty. We briefly change identity in order to get things set up
2568 * properly by the kernel. If we don't do that then we can't create it atomically, and that sucks for parallel
2569 * execution. This mimics what pam_keyinit does, too. Setting up session keyring, to be owned by the right user
2570 * & group is just as nasty as acquiring a reference to the user keyring. */
2572 saved_uid
= getuid();
2573 saved_gid
= getgid();
2575 if (gid_is_valid(gid
) && gid
!= saved_gid
) {
2576 if (setregid(gid
, -1) < 0)
2577 return log_unit_error_errno(u
, errno
, "Failed to change GID for user keyring: %m");
2580 if (uid_is_valid(uid
) && uid
!= saved_uid
) {
2581 if (setreuid(uid
, -1) < 0) {
2582 r
= log_unit_error_errno(u
, errno
, "Failed to change UID for user keyring: %m");
2587 keyring
= keyctl(KEYCTL_JOIN_SESSION_KEYRING
, 0, 0, 0, 0);
2588 if (keyring
== -1) {
2589 if (errno
== ENOSYS
)
2590 log_unit_debug_errno(u
, errno
, "Kernel keyring not supported, ignoring.");
2591 else if (IN_SET(errno
, EACCES
, EPERM
))
2592 log_unit_debug_errno(u
, errno
, "Kernel keyring access prohibited, ignoring.");
2593 else if (errno
== EDQUOT
)
2594 log_unit_debug_errno(u
, errno
, "Out of kernel keyrings to allocate, ignoring.");
2596 r
= log_unit_error_errno(u
, errno
, "Setting up kernel keyring failed: %m");
2601 /* When requested link the user keyring into the session keyring. */
2602 if (context
->keyring_mode
== EXEC_KEYRING_SHARED
) {
2604 if (keyctl(KEYCTL_LINK
,
2605 KEY_SPEC_USER_KEYRING
,
2606 KEY_SPEC_SESSION_KEYRING
, 0, 0) < 0) {
2607 r
= log_unit_error_errno(u
, errno
, "Failed to link user keyring into session keyring: %m");
2612 /* Restore uid/gid back */
2613 if (uid_is_valid(uid
) && uid
!= saved_uid
) {
2614 if (setreuid(saved_uid
, -1) < 0) {
2615 r
= log_unit_error_errno(u
, errno
, "Failed to change UID back for user keyring: %m");
2620 if (gid_is_valid(gid
) && gid
!= saved_gid
) {
2621 if (setregid(saved_gid
, -1) < 0)
2622 return log_unit_error_errno(u
, errno
, "Failed to change GID back for user keyring: %m");
2625 /* Populate they keyring with the invocation ID by default, as original saved_uid. */
2626 if (!sd_id128_is_null(u
->invocation_id
)) {
2629 key
= add_key("user", "invocation_id", &u
->invocation_id
, sizeof(u
->invocation_id
), KEY_SPEC_SESSION_KEYRING
);
2631 log_unit_debug_errno(u
, errno
, "Failed to add invocation ID to keyring, ignoring: %m");
2633 if (keyctl(KEYCTL_SETPERM
, key
,
2634 KEY_POS_VIEW
|KEY_POS_READ
|KEY_POS_SEARCH
|
2635 KEY_USR_VIEW
|KEY_USR_READ
|KEY_USR_SEARCH
, 0, 0) < 0)
2636 r
= log_unit_error_errno(u
, errno
, "Failed to restrict invocation ID permission: %m");
2641 /* Revert back uid & gid for the the last time, and exit */
2642 /* no extra logging, as only the first already reported error matters */
2643 if (getuid() != saved_uid
)
2644 (void) setreuid(saved_uid
, -1);
2646 if (getgid() != saved_gid
)
2647 (void) setregid(saved_gid
, -1);
2652 static void append_socket_pair(int *array
, size_t *n
, const int pair
[static 2]) {
2660 array
[(*n
)++] = pair
[0];
2662 array
[(*n
)++] = pair
[1];
2665 static int close_remaining_fds(
2666 const ExecParameters
*params
,
2667 const ExecRuntime
*runtime
,
2668 const DynamicCreds
*dcreds
,
2672 int *fds
, size_t n_fds
) {
2674 size_t n_dont_close
= 0;
2675 int dont_close
[n_fds
+ 12];
2679 if (params
->stdin_fd
>= 0)
2680 dont_close
[n_dont_close
++] = params
->stdin_fd
;
2681 if (params
->stdout_fd
>= 0)
2682 dont_close
[n_dont_close
++] = params
->stdout_fd
;
2683 if (params
->stderr_fd
>= 0)
2684 dont_close
[n_dont_close
++] = params
->stderr_fd
;
2687 dont_close
[n_dont_close
++] = socket_fd
;
2689 dont_close
[n_dont_close
++] = exec_fd
;
2691 memcpy(dont_close
+ n_dont_close
, fds
, sizeof(int) * n_fds
);
2692 n_dont_close
+= n_fds
;
2696 append_socket_pair(dont_close
, &n_dont_close
, runtime
->netns_storage_socket
);
2700 append_socket_pair(dont_close
, &n_dont_close
, dcreds
->user
->storage_socket
);
2702 append_socket_pair(dont_close
, &n_dont_close
, dcreds
->group
->storage_socket
);
2705 if (user_lookup_fd
>= 0)
2706 dont_close
[n_dont_close
++] = user_lookup_fd
;
2708 return close_all_fds(dont_close
, n_dont_close
);
2711 static int send_user_lookup(
2719 /* Send the resolved UID/GID to PID 1 after we learnt it. We send a single datagram, containing the UID/GID
2720 * data as well as the unit name. Note that we suppress sending this if no user/group to resolve was
2723 if (user_lookup_fd
< 0)
2726 if (!uid_is_valid(uid
) && !gid_is_valid(gid
))
2729 if (writev(user_lookup_fd
,
2731 IOVEC_INIT(&uid
, sizeof(uid
)),
2732 IOVEC_INIT(&gid
, sizeof(gid
)),
2733 IOVEC_INIT_STRING(unit
->id
) }, 3) < 0)
2739 static int acquire_home(const ExecContext
*c
, uid_t uid
, const char** home
, char **buf
) {
2746 /* If WorkingDirectory=~ is set, try to acquire a usable home directory. */
2751 if (!c
->working_directory_home
)
2754 r
= get_home_dir(buf
);
2762 static int compile_suggested_paths(const ExecContext
*c
, const ExecParameters
*p
, char ***ret
) {
2763 _cleanup_strv_free_
char ** list
= NULL
;
2764 ExecDirectoryType t
;
2771 assert(c
->dynamic_user
);
2773 /* Compile a list of paths that it might make sense to read the owning UID from to use as initial candidate for
2774 * dynamic UID allocation, in order to save us from doing costly recursive chown()s of the special
2777 for (t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
2780 if (t
== EXEC_DIRECTORY_CONFIGURATION
)
2786 STRV_FOREACH(i
, c
->directories
[t
].paths
) {
2789 if (t
== EXEC_DIRECTORY_RUNTIME
)
2790 e
= strjoin(p
->prefix
[t
], "/", *i
);
2792 e
= strjoin(p
->prefix
[t
], "/private/", *i
);
2796 r
= strv_consume(&list
, e
);
2802 *ret
= TAKE_PTR(list
);
2807 static char *exec_command_line(char **argv
);
2809 static int exec_parameters_get_cgroup_path(const ExecParameters
*params
, char **ret
) {
2810 bool using_subcgroup
;
2816 if (!params
->cgroup_path
)
2819 /* If we are called for a unit where cgroup delegation is on, and the payload created its own populated
2820 * subcgroup (which we expect it to do, after all it asked for delegation), then we cannot place the control
2821 * processes started after the main unit's process in the unit's main cgroup because it is now an inner one,
2822 * and inner cgroups may not contain processes. Hence, if delegation is on, and this is a control process,
2823 * let's use ".control" as subcgroup instead. Note that we do so only for ExecStartPost=, ExecReload=,
2824 * ExecStop=, ExecStopPost=, i.e. for the commands where the main process is already forked. For ExecStartPre=
2825 * this is not necessary, the cgroup is still empty. We distinguish these cases with the EXEC_CONTROL_CGROUP
2826 * flag, which is only passed for the former statements, not for the latter. */
2828 using_subcgroup
= FLAGS_SET(params
->flags
, EXEC_CONTROL_CGROUP
|EXEC_CGROUP_DELEGATE
|EXEC_IS_CONTROL
);
2829 if (using_subcgroup
)
2830 p
= strjoin(params
->cgroup_path
, "/.control");
2832 p
= strdup(params
->cgroup_path
);
2837 return using_subcgroup
;
2840 static int exec_child(
2842 const ExecCommand
*command
,
2843 const ExecContext
*context
,
2844 const ExecParameters
*params
,
2845 ExecRuntime
*runtime
,
2846 DynamicCreds
*dcreds
,
2850 size_t n_socket_fds
,
2851 size_t n_storage_fds
,
2856 _cleanup_strv_free_
char **our_env
= NULL
, **pass_env
= NULL
, **accum_env
= NULL
, **replaced_argv
= NULL
;
2857 int *fds_with_exec_fd
, n_fds_with_exec_fd
, r
, ngids
= 0, exec_fd
= -1;
2858 _cleanup_free_ gid_t
*supplementary_gids
= NULL
;
2859 const char *username
= NULL
, *groupname
= NULL
;
2860 _cleanup_free_
char *home_buffer
= NULL
;
2861 const char *home
= NULL
, *shell
= NULL
;
2862 char **final_argv
= NULL
;
2863 dev_t journal_stream_dev
= 0;
2864 ino_t journal_stream_ino
= 0;
2865 bool needs_sandboxing
, /* Do we need to set up full sandboxing? (i.e. all namespacing, all MAC stuff, caps, yadda yadda */
2866 needs_setuid
, /* Do we need to do the actual setresuid()/setresgid() calls? */
2867 needs_mount_namespace
, /* Do we need to set up a mount namespace for this kernel? */
2868 needs_ambient_hack
; /* Do we need to apply the ambient capabilities hack? */
2870 _cleanup_free_
char *mac_selinux_context_net
= NULL
;
2871 bool use_selinux
= false;
2874 bool use_smack
= false;
2877 bool use_apparmor
= false;
2879 uid_t uid
= UID_INVALID
;
2880 gid_t gid
= GID_INVALID
;
2882 ExecDirectoryType dt
;
2889 assert(exit_status
);
2891 rename_process_from_path(command
->path
);
2893 /* We reset exactly these signals, since they are the
2894 * only ones we set to SIG_IGN in the main daemon. All
2895 * others we leave untouched because we set them to
2896 * SIG_DFL or a valid handler initially, both of which
2897 * will be demoted to SIG_DFL. */
2898 (void) default_signals(SIGNALS_CRASH_HANDLER
,
2899 SIGNALS_IGNORE
, -1);
2901 if (context
->ignore_sigpipe
)
2902 (void) ignore_signals(SIGPIPE
, -1);
2904 r
= reset_signal_mask();
2906 *exit_status
= EXIT_SIGNAL_MASK
;
2907 return log_unit_error_errno(unit
, r
, "Failed to set process signal mask: %m");
2910 if (params
->idle_pipe
)
2911 do_idle_pipe_dance(params
->idle_pipe
);
2913 /* Close fds we don't need very early to make sure we don't block init reexecution because it cannot bind its
2914 * sockets. Among the fds we close are the logging fds, and we want to keep them closed, so that we don't have
2915 * any fds open we don't really want open during the transition. In order to make logging work, we switch the
2916 * log subsystem into open_when_needed mode, so that it reopens the logs on every single log call. */
2919 log_set_open_when_needed(true);
2921 /* In case anything used libc syslog(), close this here, too */
2924 n_fds
= n_socket_fds
+ n_storage_fds
;
2925 r
= close_remaining_fds(params
, runtime
, dcreds
, user_lookup_fd
, socket_fd
, params
->exec_fd
, fds
, n_fds
);
2927 *exit_status
= EXIT_FDS
;
2928 return log_unit_error_errno(unit
, r
, "Failed to close unwanted file descriptors: %m");
2931 if (!context
->same_pgrp
)
2933 *exit_status
= EXIT_SETSID
;
2934 return log_unit_error_errno(unit
, errno
, "Failed to create new process session: %m");
2937 exec_context_tty_reset(context
, params
);
2939 if (unit_shall_confirm_spawn(unit
)) {
2940 const char *vc
= params
->confirm_spawn
;
2941 _cleanup_free_
char *cmdline
= NULL
;
2943 cmdline
= exec_command_line(command
->argv
);
2945 *exit_status
= EXIT_MEMORY
;
2949 r
= ask_for_confirmation(vc
, unit
, cmdline
);
2950 if (r
!= CONFIRM_EXECUTE
) {
2951 if (r
== CONFIRM_PRETEND_SUCCESS
) {
2952 *exit_status
= EXIT_SUCCESS
;
2955 *exit_status
= EXIT_CONFIRM
;
2956 log_unit_error(unit
, "Execution cancelled by the user");
2961 /* We are about to invoke NSS and PAM modules. Let's tell them what we are doing here, maybe they care. This is
2962 * used by nss-resolve to disable itself when we are about to start systemd-resolved, to avoid deadlocks. Note
2963 * that these env vars do not survive the execve(), which means they really only apply to the PAM and NSS
2964 * invocations themselves. Also note that while we'll only invoke NSS modules involved in user management they
2965 * might internally call into other NSS modules that are involved in hostname resolution, we never know. */
2966 if (setenv("SYSTEMD_ACTIVATION_UNIT", unit
->id
, true) != 0 ||
2967 setenv("SYSTEMD_ACTIVATION_SCOPE", MANAGER_IS_SYSTEM(unit
->manager
) ? "system" : "user", true) != 0) {
2968 *exit_status
= EXIT_MEMORY
;
2969 return log_unit_error_errno(unit
, errno
, "Failed to update environment: %m");
2972 if (context
->dynamic_user
&& dcreds
) {
2973 _cleanup_strv_free_
char **suggested_paths
= NULL
;
2975 /* On top of that, make sure we bypass our own NSS module nss-systemd comprehensively for any NSS
2976 * checks, if DynamicUser=1 is used, as we shouldn't create a feedback loop with ourselves here.*/
2977 if (putenv((char*) "SYSTEMD_NSS_DYNAMIC_BYPASS=1") != 0) {
2978 *exit_status
= EXIT_USER
;
2979 return log_unit_error_errno(unit
, errno
, "Failed to update environment: %m");
2982 r
= compile_suggested_paths(context
, params
, &suggested_paths
);
2984 *exit_status
= EXIT_MEMORY
;
2988 r
= dynamic_creds_realize(dcreds
, suggested_paths
, &uid
, &gid
);
2990 *exit_status
= EXIT_USER
;
2992 log_unit_error(unit
, "Failed to update dynamic user credentials: User or group with specified name already exists.");
2995 return log_unit_error_errno(unit
, r
, "Failed to update dynamic user credentials: %m");
2998 if (!uid_is_valid(uid
)) {
2999 *exit_status
= EXIT_USER
;
3000 log_unit_error(unit
, "UID validation failed for \""UID_FMT
"\"", uid
);
3004 if (!gid_is_valid(gid
)) {
3005 *exit_status
= EXIT_USER
;
3006 log_unit_error(unit
, "GID validation failed for \""GID_FMT
"\"", gid
);
3011 username
= dcreds
->user
->name
;
3014 r
= get_fixed_user(context
, &username
, &uid
, &gid
, &home
, &shell
);
3016 *exit_status
= EXIT_USER
;
3017 return log_unit_error_errno(unit
, r
, "Failed to determine user credentials: %m");
3020 r
= get_fixed_group(context
, &groupname
, &gid
);
3022 *exit_status
= EXIT_GROUP
;
3023 return log_unit_error_errno(unit
, r
, "Failed to determine group credentials: %m");
3027 /* Initialize user supplementary groups and get SupplementaryGroups= ones */
3028 r
= get_supplementary_groups(context
, username
, groupname
, gid
,
3029 &supplementary_gids
, &ngids
);
3031 *exit_status
= EXIT_GROUP
;
3032 return log_unit_error_errno(unit
, r
, "Failed to determine supplementary groups: %m");
3035 r
= send_user_lookup(unit
, user_lookup_fd
, uid
, gid
);
3037 *exit_status
= EXIT_USER
;
3038 return log_unit_error_errno(unit
, r
, "Failed to send user credentials to PID1: %m");
3041 user_lookup_fd
= safe_close(user_lookup_fd
);
3043 r
= acquire_home(context
, uid
, &home
, &home_buffer
);
3045 *exit_status
= EXIT_CHDIR
;
3046 return log_unit_error_errno(unit
, r
, "Failed to determine $HOME for user: %m");
3049 /* If a socket is connected to STDIN/STDOUT/STDERR, we
3050 * must sure to drop O_NONBLOCK */
3052 (void) fd_nonblock(socket_fd
, false);
3054 /* Journald will try to look-up our cgroup in order to populate _SYSTEMD_CGROUP and _SYSTEMD_UNIT fields.
3055 * Hence we need to migrate to the target cgroup from init.scope before connecting to journald */
3056 if (params
->cgroup_path
) {
3057 _cleanup_free_
char *p
= NULL
;
3059 r
= exec_parameters_get_cgroup_path(params
, &p
);
3061 *exit_status
= EXIT_CGROUP
;
3062 return log_unit_error_errno(unit
, r
, "Failed to acquire cgroup path: %m");
3065 r
= cg_attach_everywhere(params
->cgroup_supported
, p
, 0, NULL
, NULL
);
3067 *exit_status
= EXIT_CGROUP
;
3068 return log_unit_error_errno(unit
, r
, "Failed to attach to cgroup %s: %m", p
);
3072 if (context
->network_namespace_path
&& runtime
&& runtime
->netns_storage_socket
[0] >= 0) {
3073 r
= open_netns_path(runtime
->netns_storage_socket
, context
->network_namespace_path
);
3075 *exit_status
= EXIT_NETWORK
;
3076 return log_unit_error_errno(unit
, r
, "Failed to open network namespace path %s: %m", context
->network_namespace_path
);
3080 r
= setup_input(context
, params
, socket_fd
, named_iofds
);
3082 *exit_status
= EXIT_STDIN
;
3083 return log_unit_error_errno(unit
, r
, "Failed to set up standard input: %m");
3086 r
= setup_output(unit
, context
, params
, STDOUT_FILENO
, socket_fd
, named_iofds
, basename(command
->path
), uid
, gid
, &journal_stream_dev
, &journal_stream_ino
);
3088 *exit_status
= EXIT_STDOUT
;
3089 return log_unit_error_errno(unit
, r
, "Failed to set up standard output: %m");
3092 r
= setup_output(unit
, context
, params
, STDERR_FILENO
, socket_fd
, named_iofds
, basename(command
->path
), uid
, gid
, &journal_stream_dev
, &journal_stream_ino
);
3094 *exit_status
= EXIT_STDERR
;
3095 return log_unit_error_errno(unit
, r
, "Failed to set up standard error output: %m");
3098 if (context
->oom_score_adjust_set
) {
3099 /* When we can't make this change due to EPERM, then let's silently skip over it. User namespaces
3100 * prohibit write access to this file, and we shouldn't trip up over that. */
3101 r
= set_oom_score_adjust(context
->oom_score_adjust
);
3102 if (IN_SET(r
, -EPERM
, -EACCES
))
3103 log_unit_debug_errno(unit
, r
, "Failed to adjust OOM setting, assuming containerized execution, ignoring: %m");
3105 *exit_status
= EXIT_OOM_ADJUST
;
3106 return log_unit_error_errno(unit
, r
, "Failed to adjust OOM setting: %m");
3110 if (context
->nice_set
)
3111 if (setpriority(PRIO_PROCESS
, 0, context
->nice
) < 0) {
3112 *exit_status
= EXIT_NICE
;
3113 return log_unit_error_errno(unit
, errno
, "Failed to set up process scheduling priority (nice level): %m");
3116 if (context
->cpu_sched_set
) {
3117 struct sched_param param
= {
3118 .sched_priority
= context
->cpu_sched_priority
,
3121 r
= sched_setscheduler(0,
3122 context
->cpu_sched_policy
|
3123 (context
->cpu_sched_reset_on_fork
?
3124 SCHED_RESET_ON_FORK
: 0),
3127 *exit_status
= EXIT_SETSCHEDULER
;
3128 return log_unit_error_errno(unit
, errno
, "Failed to set up CPU scheduling: %m");
3132 if (context
->cpuset
)
3133 if (sched_setaffinity(0, CPU_ALLOC_SIZE(context
->cpuset_ncpus
), context
->cpuset
) < 0) {
3134 *exit_status
= EXIT_CPUAFFINITY
;
3135 return log_unit_error_errno(unit
, errno
, "Failed to set up CPU affinity: %m");
3138 if (context
->ioprio_set
)
3139 if (ioprio_set(IOPRIO_WHO_PROCESS
, 0, context
->ioprio
) < 0) {
3140 *exit_status
= EXIT_IOPRIO
;
3141 return log_unit_error_errno(unit
, errno
, "Failed to set up IO scheduling priority: %m");
3144 if (context
->timer_slack_nsec
!= NSEC_INFINITY
)
3145 if (prctl(PR_SET_TIMERSLACK
, context
->timer_slack_nsec
) < 0) {
3146 *exit_status
= EXIT_TIMERSLACK
;
3147 return log_unit_error_errno(unit
, errno
, "Failed to set up timer slack: %m");
3150 if (context
->personality
!= PERSONALITY_INVALID
) {
3151 r
= safe_personality(context
->personality
);
3153 *exit_status
= EXIT_PERSONALITY
;
3154 return log_unit_error_errno(unit
, r
, "Failed to set up execution domain (personality): %m");
3158 if (context
->utmp_id
)
3159 utmp_put_init_process(context
->utmp_id
, getpid_cached(), getsid(0),
3161 context
->utmp_mode
== EXEC_UTMP_INIT
? INIT_PROCESS
:
3162 context
->utmp_mode
== EXEC_UTMP_LOGIN
? LOGIN_PROCESS
:
3166 if (uid_is_valid(uid
)) {
3167 r
= chown_terminal(STDIN_FILENO
, uid
);
3169 *exit_status
= EXIT_STDIN
;
3170 return log_unit_error_errno(unit
, r
, "Failed to change ownership of terminal: %m");
3174 /* If delegation is enabled we'll pass ownership of the cgroup to the user of the new process. On cgroup v1
3175 * this is only about systemd's own hierarchy, i.e. not the controller hierarchies, simply because that's not
3176 * safe. On cgroup v2 there's only one hierarchy anyway, and delegation is safe there, hence in that case only
3177 * touch a single hierarchy too. */
3178 if (params
->cgroup_path
&& context
->user
&& (params
->flags
& EXEC_CGROUP_DELEGATE
)) {
3179 r
= cg_set_access(SYSTEMD_CGROUP_CONTROLLER
, params
->cgroup_path
, uid
, gid
);
3181 *exit_status
= EXIT_CGROUP
;
3182 return log_unit_error_errno(unit
, r
, "Failed to adjust control group access: %m");
3186 for (dt
= 0; dt
< _EXEC_DIRECTORY_TYPE_MAX
; dt
++) {
3187 r
= setup_exec_directory(context
, params
, uid
, gid
, dt
, exit_status
);
3189 return log_unit_error_errno(unit
, r
, "Failed to set up special execution directory in %s: %m", params
->prefix
[dt
]);
3192 r
= build_environment(
3204 *exit_status
= EXIT_MEMORY
;
3208 r
= build_pass_environment(context
, &pass_env
);
3210 *exit_status
= EXIT_MEMORY
;
3214 accum_env
= strv_env_merge(5,
3215 params
->environment
,
3218 context
->environment
,
3222 *exit_status
= EXIT_MEMORY
;
3225 accum_env
= strv_env_clean(accum_env
);
3227 (void) umask(context
->umask
);
3229 r
= setup_keyring(unit
, context
, params
, uid
, gid
);
3231 *exit_status
= EXIT_KEYRING
;
3232 return log_unit_error_errno(unit
, r
, "Failed to set up kernel keyring: %m");
3235 /* We need sandboxing if the caller asked us to apply it and the command isn't explicitly excepted from it */
3236 needs_sandboxing
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && !(command
->flags
& EXEC_COMMAND_FULLY_PRIVILEGED
);
3238 /* 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 */
3239 needs_ambient_hack
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && (command
->flags
& EXEC_COMMAND_AMBIENT_MAGIC
) && !ambient_capabilities_supported();
3241 /* 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 */
3242 if (needs_ambient_hack
)
3243 needs_setuid
= false;
3245 needs_setuid
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && !(command
->flags
& (EXEC_COMMAND_FULLY_PRIVILEGED
|EXEC_COMMAND_NO_SETUID
));
3247 if (needs_sandboxing
) {
3248 /* MAC enablement checks need to be done before a new mount ns is created, as they rely on /sys being
3249 * present. The actual MAC context application will happen later, as late as possible, to avoid
3250 * impacting our own code paths. */
3253 use_selinux
= mac_selinux_use();
3256 use_smack
= mac_smack_use();
3259 use_apparmor
= mac_apparmor_use();
3263 if (needs_sandboxing
) {
3266 /* Let's set the resource limits before we call into PAM, so that pam_limits wins over what
3267 * is set here. (See below.) */
3269 r
= setrlimit_closest_all((const struct rlimit
* const *) context
->rlimit
, &which_failed
);
3271 *exit_status
= EXIT_LIMITS
;
3272 return log_unit_error_errno(unit
, r
, "Failed to adjust resource limit RLIMIT_%s: %m", rlimit_to_string(which_failed
));
3278 /* Let's call into PAM after we set up our own idea of resource limits to that pam_limits
3279 * wins here. (See above.) */
3281 if (context
->pam_name
&& username
) {
3282 r
= setup_pam(context
->pam_name
, username
, uid
, gid
, context
->tty_path
, &accum_env
, fds
, n_fds
);
3284 *exit_status
= EXIT_PAM
;
3285 return log_unit_error_errno(unit
, r
, "Failed to set up PAM session: %m");
3290 if ((context
->private_network
|| context
->network_namespace_path
) && runtime
&& runtime
->netns_storage_socket
[0] >= 0) {
3292 if (ns_type_supported(NAMESPACE_NET
)) {
3293 r
= setup_netns(runtime
->netns_storage_socket
);
3295 *exit_status
= EXIT_NETWORK
;
3296 return log_unit_error_errno(unit
, r
, "Failed to set up network namespacing: %m");
3298 } else if (context
->network_namespace_path
) {
3299 *exit_status
= EXIT_NETWORK
;
3300 return log_unit_error_errno(unit
, SYNTHETIC_ERRNO(EOPNOTSUPP
), "NetworkNamespacePath= is not supported, refusing.");
3302 log_unit_warning(unit
, "PrivateNetwork=yes is configured, but the kernel does not support network namespaces, ignoring.");
3305 needs_mount_namespace
= exec_needs_mount_namespace(context
, params
, runtime
);
3306 if (needs_mount_namespace
) {
3307 r
= apply_mount_namespace(unit
, command
, context
, params
, runtime
);
3309 *exit_status
= EXIT_NAMESPACE
;
3310 return log_unit_error_errno(unit
, r
, "Failed to set up mount namespacing: %m");
3314 if (context
->protect_hostname
) {
3315 if (ns_type_supported(NAMESPACE_UTS
)) {
3316 if (unshare(CLONE_NEWUTS
) < 0) {
3317 *exit_status
= EXIT_NAMESPACE
;
3318 return log_unit_error_errno(unit
, errno
, "Failed to set up UTS namespacing: %m");
3321 log_unit_warning(unit
, "ProtectHostname=yes is configured, but the kernel does not support UTS namespaces, ignoring namespace setup.");
3323 r
= seccomp_protect_hostname();
3325 *exit_status
= EXIT_SECCOMP
;
3326 return log_unit_error_errno(unit
, r
, "Failed to apply hostname restrictions: %m");
3331 /* Drop groups as early as possbile */
3333 r
= enforce_groups(gid
, supplementary_gids
, ngids
);
3335 *exit_status
= EXIT_GROUP
;
3336 return log_unit_error_errno(unit
, r
, "Changing group credentials failed: %m");
3340 if (needs_sandboxing
) {
3342 if (use_selinux
&& params
->selinux_context_net
&& socket_fd
>= 0) {
3343 r
= mac_selinux_get_child_mls_label(socket_fd
, command
->path
, context
->selinux_context
, &mac_selinux_context_net
);
3345 *exit_status
= EXIT_SELINUX_CONTEXT
;
3346 return log_unit_error_errno(unit
, r
, "Failed to determine SELinux context: %m");
3351 if (context
->private_users
) {
3352 r
= setup_private_users(uid
, gid
);
3354 *exit_status
= EXIT_USER
;
3355 return log_unit_error_errno(unit
, r
, "Failed to set up user namespacing: %m");
3360 /* We repeat the fd closing here, to make sure that nothing is leaked from the PAM modules. Note that we are
3361 * more aggressive this time since socket_fd and the netns fds we don't need anymore. We do keep the exec_fd
3362 * however if we have it as we want to keep it open until the final execve(). */
3364 if (params
->exec_fd
>= 0) {
3365 exec_fd
= params
->exec_fd
;
3367 if (exec_fd
< 3 + (int) n_fds
) {
3370 /* Let's move the exec fd far up, so that it's outside of the fd range we want to pass to the
3371 * process we are about to execute. */
3373 moved_fd
= fcntl(exec_fd
, F_DUPFD_CLOEXEC
, 3 + (int) n_fds
);
3375 *exit_status
= EXIT_FDS
;
3376 return log_unit_error_errno(unit
, errno
, "Couldn't move exec fd up: %m");
3379 safe_close(exec_fd
);
3382 /* This fd should be FD_CLOEXEC already, but let's make sure. */
3383 r
= fd_cloexec(exec_fd
, true);
3385 *exit_status
= EXIT_FDS
;
3386 return log_unit_error_errno(unit
, r
, "Failed to make exec fd FD_CLOEXEC: %m");
3390 fds_with_exec_fd
= newa(int, n_fds
+ 1);
3391 memcpy_safe(fds_with_exec_fd
, fds
, n_fds
* sizeof(int));
3392 fds_with_exec_fd
[n_fds
] = exec_fd
;
3393 n_fds_with_exec_fd
= n_fds
+ 1;
3395 fds_with_exec_fd
= fds
;
3396 n_fds_with_exec_fd
= n_fds
;
3399 r
= close_all_fds(fds_with_exec_fd
, n_fds_with_exec_fd
);
3401 r
= shift_fds(fds
, n_fds
);
3403 r
= flags_fds(fds
, n_socket_fds
, n_storage_fds
, context
->non_blocking
);
3405 *exit_status
= EXIT_FDS
;
3406 return log_unit_error_errno(unit
, r
, "Failed to adjust passed file descriptors: %m");
3409 /* At this point, the fds we want to pass to the program are all ready and set up, with O_CLOEXEC turned off
3410 * and at the right fd numbers. The are no other fds open, with one exception: the exec_fd if it is defined,
3411 * and it has O_CLOEXEC set, after all we want it to be closed by the execve(), so that our parent knows we
3414 secure_bits
= context
->secure_bits
;
3416 if (needs_sandboxing
) {
3419 /* Set the RTPRIO resource limit to 0, but only if nothing else was explicitly
3420 * requested. (Note this is placed after the general resource limit initialization, see
3421 * above, in order to take precedence.) */
3422 if (context
->restrict_realtime
&& !context
->rlimit
[RLIMIT_RTPRIO
]) {
3423 if (setrlimit(RLIMIT_RTPRIO
, &RLIMIT_MAKE_CONST(0)) < 0) {
3424 *exit_status
= EXIT_LIMITS
;
3425 return log_unit_error_errno(unit
, errno
, "Failed to adjust RLIMIT_RTPRIO resource limit: %m");
3430 /* LSM Smack needs the capability CAP_MAC_ADMIN to change the current execution security context of the
3431 * process. This is the latest place before dropping capabilities. Other MAC context are set later. */
3433 r
= setup_smack(context
, command
);
3435 *exit_status
= EXIT_SMACK_PROCESS_LABEL
;
3436 return log_unit_error_errno(unit
, r
, "Failed to set SMACK process label: %m");
3441 bset
= context
->capability_bounding_set
;
3442 /* If the ambient caps hack is enabled (which means the kernel can't do them, and the user asked for
3443 * our magic fallback), then let's add some extra caps, so that the service can drop privs of its own,
3444 * instead of us doing that */
3445 if (needs_ambient_hack
)
3446 bset
|= (UINT64_C(1) << CAP_SETPCAP
) |
3447 (UINT64_C(1) << CAP_SETUID
) |
3448 (UINT64_C(1) << CAP_SETGID
);
3450 if (!cap_test_all(bset
)) {
3451 r
= capability_bounding_set_drop(bset
, false);
3453 *exit_status
= EXIT_CAPABILITIES
;
3454 return log_unit_error_errno(unit
, r
, "Failed to drop capabilities: %m");
3458 /* This is done before enforce_user, but ambient set
3459 * does not survive over setresuid() if keep_caps is not set. */
3460 if (!needs_ambient_hack
&&
3461 context
->capability_ambient_set
!= 0) {
3462 r
= capability_ambient_set_apply(context
->capability_ambient_set
, true);
3464 *exit_status
= EXIT_CAPABILITIES
;
3465 return log_unit_error_errno(unit
, r
, "Failed to apply ambient capabilities (before UID change): %m");
3471 if (uid_is_valid(uid
)) {
3472 r
= enforce_user(context
, uid
);
3474 *exit_status
= EXIT_USER
;
3475 return log_unit_error_errno(unit
, r
, "Failed to change UID to " UID_FMT
": %m", uid
);
3478 if (!needs_ambient_hack
&&
3479 context
->capability_ambient_set
!= 0) {
3481 /* Fix the ambient capabilities after user change. */
3482 r
= capability_ambient_set_apply(context
->capability_ambient_set
, false);
3484 *exit_status
= EXIT_CAPABILITIES
;
3485 return log_unit_error_errno(unit
, r
, "Failed to apply ambient capabilities (after UID change): %m");
3488 /* If we were asked to change user and ambient capabilities
3489 * were requested, we had to add keep-caps to the securebits
3490 * so that we would maintain the inherited capability set
3491 * through the setresuid(). Make sure that the bit is added
3492 * also to the context secure_bits so that we don't try to
3493 * drop the bit away next. */
3495 secure_bits
|= 1<<SECURE_KEEP_CAPS
;
3500 /* Apply working directory here, because the working directory might be on NFS and only the user running
3501 * this service might have the correct privilege to change to the working directory */
3502 r
= apply_working_directory(context
, params
, home
, needs_mount_namespace
, exit_status
);
3504 return log_unit_error_errno(unit
, r
, "Changing to the requested working directory failed: %m");
3506 if (needs_sandboxing
) {
3507 /* Apply other MAC contexts late, but before seccomp syscall filtering, as those should really be last to
3508 * influence our own codepaths as little as possible. Moreover, applying MAC contexts usually requires
3509 * syscalls that are subject to seccomp filtering, hence should probably be applied before the syscalls
3510 * are restricted. */
3514 char *exec_context
= mac_selinux_context_net
?: context
->selinux_context
;
3517 r
= setexeccon(exec_context
);
3519 *exit_status
= EXIT_SELINUX_CONTEXT
;
3520 return log_unit_error_errno(unit
, r
, "Failed to change SELinux context to %s: %m", exec_context
);
3527 if (use_apparmor
&& context
->apparmor_profile
) {
3528 r
= aa_change_onexec(context
->apparmor_profile
);
3529 if (r
< 0 && !context
->apparmor_profile_ignore
) {
3530 *exit_status
= EXIT_APPARMOR_PROFILE
;
3531 return log_unit_error_errno(unit
, errno
, "Failed to prepare AppArmor profile change to %s: %m", context
->apparmor_profile
);
3536 /* PR_GET_SECUREBITS is not privileged, while PR_SET_SECUREBITS is. So to suppress potential EPERMs
3537 * we'll try not to call PR_SET_SECUREBITS unless necessary. */
3538 if (prctl(PR_GET_SECUREBITS
) != secure_bits
)
3539 if (prctl(PR_SET_SECUREBITS
, secure_bits
) < 0) {
3540 *exit_status
= EXIT_SECUREBITS
;
3541 return log_unit_error_errno(unit
, errno
, "Failed to set process secure bits: %m");
3544 if (context_has_no_new_privileges(context
))
3545 if (prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0) < 0) {
3546 *exit_status
= EXIT_NO_NEW_PRIVILEGES
;
3547 return log_unit_error_errno(unit
, errno
, "Failed to disable new privileges: %m");
3551 r
= apply_address_families(unit
, context
);
3553 *exit_status
= EXIT_ADDRESS_FAMILIES
;
3554 return log_unit_error_errno(unit
, r
, "Failed to restrict address families: %m");
3557 r
= apply_memory_deny_write_execute(unit
, context
);
3559 *exit_status
= EXIT_SECCOMP
;
3560 return log_unit_error_errno(unit
, r
, "Failed to disable writing to executable memory: %m");
3563 r
= apply_restrict_realtime(unit
, context
);
3565 *exit_status
= EXIT_SECCOMP
;
3566 return log_unit_error_errno(unit
, r
, "Failed to apply realtime restrictions: %m");
3569 r
= apply_restrict_namespaces(unit
, context
);
3571 *exit_status
= EXIT_SECCOMP
;
3572 return log_unit_error_errno(unit
, r
, "Failed to apply namespace restrictions: %m");
3575 r
= apply_protect_sysctl(unit
, context
);
3577 *exit_status
= EXIT_SECCOMP
;
3578 return log_unit_error_errno(unit
, r
, "Failed to apply sysctl restrictions: %m");
3581 r
= apply_protect_kernel_modules(unit
, context
);
3583 *exit_status
= EXIT_SECCOMP
;
3584 return log_unit_error_errno(unit
, r
, "Failed to apply module loading restrictions: %m");
3587 r
= apply_private_devices(unit
, context
);
3589 *exit_status
= EXIT_SECCOMP
;
3590 return log_unit_error_errno(unit
, r
, "Failed to set up private devices: %m");
3593 r
= apply_syscall_archs(unit
, context
);
3595 *exit_status
= EXIT_SECCOMP
;
3596 return log_unit_error_errno(unit
, r
, "Failed to apply syscall architecture restrictions: %m");
3599 r
= apply_lock_personality(unit
, context
);
3601 *exit_status
= EXIT_SECCOMP
;
3602 return log_unit_error_errno(unit
, r
, "Failed to lock personalities: %m");
3605 /* This really should remain the last step before the execve(), to make sure our own code is unaffected
3606 * by the filter as little as possible. */
3607 r
= apply_syscall_filter(unit
, context
, needs_ambient_hack
);
3609 *exit_status
= EXIT_SECCOMP
;
3610 return log_unit_error_errno(unit
, r
, "Failed to apply system call filters: %m");
3615 if (!strv_isempty(context
->unset_environment
)) {
3618 ee
= strv_env_delete(accum_env
, 1, context
->unset_environment
);
3620 *exit_status
= EXIT_MEMORY
;
3624 strv_free_and_replace(accum_env
, ee
);
3627 if (!FLAGS_SET(command
->flags
, EXEC_COMMAND_NO_ENV_EXPAND
)) {
3628 replaced_argv
= replace_env_argv(command
->argv
, accum_env
);
3629 if (!replaced_argv
) {
3630 *exit_status
= EXIT_MEMORY
;
3633 final_argv
= replaced_argv
;
3635 final_argv
= command
->argv
;
3637 if (DEBUG_LOGGING
) {
3638 _cleanup_free_
char *line
;
3640 line
= exec_command_line(final_argv
);
3642 log_struct(LOG_DEBUG
,
3643 "EXECUTABLE=%s", command
->path
,
3644 LOG_UNIT_MESSAGE(unit
, "Executing: %s", line
),
3646 LOG_UNIT_INVOCATION_ID(unit
));
3652 /* We have finished with all our initializations. Let's now let the manager know that. From this point
3653 * on, if the manager sees POLLHUP on the exec_fd, then execve() was successful. */
3655 if (write(exec_fd
, &hot
, sizeof(hot
)) < 0) {
3656 *exit_status
= EXIT_EXEC
;
3657 return log_unit_error_errno(unit
, errno
, "Failed to enable exec_fd: %m");
3661 execve(command
->path
, final_argv
, accum_env
);
3667 /* The execve() failed. This means the exec_fd is still open. Which means we need to tell the manager
3668 * that POLLHUP on it no longer means execve() succeeded. */
3670 if (write(exec_fd
, &hot
, sizeof(hot
)) < 0) {
3671 *exit_status
= EXIT_EXEC
;
3672 return log_unit_error_errno(unit
, errno
, "Failed to disable exec_fd: %m");
3676 if (r
== -ENOENT
&& (command
->flags
& EXEC_COMMAND_IGNORE_FAILURE
)) {
3677 log_struct_errno(LOG_INFO
, r
,
3678 "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR
,
3680 LOG_UNIT_INVOCATION_ID(unit
),
3681 LOG_UNIT_MESSAGE(unit
, "Executable %s missing, skipping: %m",
3683 "EXECUTABLE=%s", command
->path
);
3687 *exit_status
= EXIT_EXEC
;
3688 return log_unit_error_errno(unit
, r
, "Failed to execute command: %m");
3691 static int exec_context_load_environment(const Unit
*unit
, const ExecContext
*c
, char ***l
);
3692 static int exec_context_named_iofds(const ExecContext
*c
, const ExecParameters
*p
, int named_iofds
[3]);
3694 int exec_spawn(Unit
*unit
,
3695 ExecCommand
*command
,
3696 const ExecContext
*context
,
3697 const ExecParameters
*params
,
3698 ExecRuntime
*runtime
,
3699 DynamicCreds
*dcreds
,
3702 int socket_fd
, r
, named_iofds
[3] = { -1, -1, -1 }, *fds
= NULL
;
3703 _cleanup_free_
char *subcgroup_path
= NULL
;
3704 _cleanup_strv_free_
char **files_env
= NULL
;
3705 size_t n_storage_fds
= 0, n_socket_fds
= 0;
3706 _cleanup_free_
char *line
= NULL
;
3714 assert(params
->fds
|| (params
->n_socket_fds
+ params
->n_storage_fds
<= 0));
3716 if (context
->std_input
== EXEC_INPUT_SOCKET
||
3717 context
->std_output
== EXEC_OUTPUT_SOCKET
||
3718 context
->std_error
== EXEC_OUTPUT_SOCKET
) {
3720 if (params
->n_socket_fds
> 1) {
3721 log_unit_error(unit
, "Got more than one socket.");
3725 if (params
->n_socket_fds
== 0) {
3726 log_unit_error(unit
, "Got no socket.");
3730 socket_fd
= params
->fds
[0];
3734 n_socket_fds
= params
->n_socket_fds
;
3735 n_storage_fds
= params
->n_storage_fds
;
3738 r
= exec_context_named_iofds(context
, params
, named_iofds
);
3740 return log_unit_error_errno(unit
, r
, "Failed to load a named file descriptor: %m");
3742 r
= exec_context_load_environment(unit
, context
, &files_env
);
3744 return log_unit_error_errno(unit
, r
, "Failed to load environment files: %m");
3746 line
= exec_command_line(command
->argv
);
3750 log_struct(LOG_DEBUG
,
3751 LOG_UNIT_MESSAGE(unit
, "About to execute: %s", line
),
3752 "EXECUTABLE=%s", command
->path
,
3754 LOG_UNIT_INVOCATION_ID(unit
));
3756 if (params
->cgroup_path
) {
3757 r
= exec_parameters_get_cgroup_path(params
, &subcgroup_path
);
3759 return log_unit_error_errno(unit
, r
, "Failed to acquire subcgroup path: %m");
3760 if (r
> 0) { /* We are using a child cgroup */
3761 r
= cg_create(SYSTEMD_CGROUP_CONTROLLER
, subcgroup_path
);
3763 return log_unit_error_errno(unit
, r
, "Failed to create control group '%s': %m", subcgroup_path
);
3769 return log_unit_error_errno(unit
, errno
, "Failed to fork: %m");
3772 int exit_status
= EXIT_SUCCESS
;
3774 r
= exec_child(unit
,
3786 unit
->manager
->user_lookup_fds
[1],
3790 log_struct_errno(LOG_ERR
, r
,
3791 "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR
,
3793 LOG_UNIT_INVOCATION_ID(unit
),
3794 LOG_UNIT_MESSAGE(unit
, "Failed at step %s spawning %s: %m",
3795 exit_status_to_string(exit_status
, EXIT_STATUS_SYSTEMD
),
3797 "EXECUTABLE=%s", command
->path
);
3802 log_unit_debug(unit
, "Forked %s as "PID_FMT
, command
->path
, pid
);
3804 /* We add the new process to the cgroup both in the child (so that we can be sure that no user code is ever
3805 * executed outside of the cgroup) and in the parent (so that we can be sure that when we kill the cgroup the
3806 * process will be killed too). */
3808 (void) cg_attach(SYSTEMD_CGROUP_CONTROLLER
, subcgroup_path
, pid
);
3810 exec_status_start(&command
->exec_status
, pid
);
3816 void exec_context_init(ExecContext
*c
) {
3817 ExecDirectoryType i
;
3822 c
->ioprio
= IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE
, 0);
3823 c
->cpu_sched_policy
= SCHED_OTHER
;
3824 c
->syslog_priority
= LOG_DAEMON
|LOG_INFO
;
3825 c
->syslog_level_prefix
= true;
3826 c
->ignore_sigpipe
= true;
3827 c
->timer_slack_nsec
= NSEC_INFINITY
;
3828 c
->personality
= PERSONALITY_INVALID
;
3829 for (i
= 0; i
< _EXEC_DIRECTORY_TYPE_MAX
; i
++)
3830 c
->directories
[i
].mode
= 0755;
3831 c
->capability_bounding_set
= CAP_ALL
;
3832 assert_cc(NAMESPACE_FLAGS_INITIAL
!= NAMESPACE_FLAGS_ALL
);
3833 c
->restrict_namespaces
= NAMESPACE_FLAGS_INITIAL
;
3834 c
->log_level_max
= -1;
3837 void exec_context_done(ExecContext
*c
) {
3838 ExecDirectoryType i
;
3843 c
->environment
= strv_free(c
->environment
);
3844 c
->environment_files
= strv_free(c
->environment_files
);
3845 c
->pass_environment
= strv_free(c
->pass_environment
);
3846 c
->unset_environment
= strv_free(c
->unset_environment
);
3848 rlimit_free_all(c
->rlimit
);
3850 for (l
= 0; l
< 3; l
++) {
3851 c
->stdio_fdname
[l
] = mfree(c
->stdio_fdname
[l
]);
3852 c
->stdio_file
[l
] = mfree(c
->stdio_file
[l
]);
3855 c
->working_directory
= mfree(c
->working_directory
);
3856 c
->root_directory
= mfree(c
->root_directory
);
3857 c
->root_image
= mfree(c
->root_image
);
3858 c
->tty_path
= mfree(c
->tty_path
);
3859 c
->syslog_identifier
= mfree(c
->syslog_identifier
);
3860 c
->user
= mfree(c
->user
);
3861 c
->group
= mfree(c
->group
);
3863 c
->supplementary_groups
= strv_free(c
->supplementary_groups
);
3865 c
->pam_name
= mfree(c
->pam_name
);
3867 c
->read_only_paths
= strv_free(c
->read_only_paths
);
3868 c
->read_write_paths
= strv_free(c
->read_write_paths
);
3869 c
->inaccessible_paths
= strv_free(c
->inaccessible_paths
);
3871 bind_mount_free_many(c
->bind_mounts
, c
->n_bind_mounts
);
3872 c
->bind_mounts
= NULL
;
3873 c
->n_bind_mounts
= 0;
3874 temporary_filesystem_free_many(c
->temporary_filesystems
, c
->n_temporary_filesystems
);
3875 c
->temporary_filesystems
= NULL
;
3876 c
->n_temporary_filesystems
= 0;
3878 c
->cpuset
= cpu_set_mfree(c
->cpuset
);
3880 c
->utmp_id
= mfree(c
->utmp_id
);
3881 c
->selinux_context
= mfree(c
->selinux_context
);
3882 c
->apparmor_profile
= mfree(c
->apparmor_profile
);
3883 c
->smack_process_label
= mfree(c
->smack_process_label
);
3885 c
->syscall_filter
= hashmap_free(c
->syscall_filter
);
3886 c
->syscall_archs
= set_free(c
->syscall_archs
);
3887 c
->address_families
= set_free(c
->address_families
);
3889 for (i
= 0; i
< _EXEC_DIRECTORY_TYPE_MAX
; i
++)
3890 c
->directories
[i
].paths
= strv_free(c
->directories
[i
].paths
);
3892 c
->log_level_max
= -1;
3894 exec_context_free_log_extra_fields(c
);
3896 c
->log_rate_limit_interval_usec
= 0;
3897 c
->log_rate_limit_burst
= 0;
3899 c
->stdin_data
= mfree(c
->stdin_data
);
3900 c
->stdin_data_size
= 0;
3902 c
->network_namespace_path
= mfree(c
->network_namespace_path
);
3905 int exec_context_destroy_runtime_directory(const ExecContext
*c
, const char *runtime_prefix
) {
3910 if (!runtime_prefix
)
3913 STRV_FOREACH(i
, c
->directories
[EXEC_DIRECTORY_RUNTIME
].paths
) {
3914 _cleanup_free_
char *p
;
3916 p
= path_join(runtime_prefix
, *i
);
3920 /* We execute this synchronously, since we need to be sure this is gone when we start the
3922 (void) rm_rf(p
, REMOVE_ROOT
);
3928 static void exec_command_done(ExecCommand
*c
) {
3931 c
->path
= mfree(c
->path
);
3932 c
->argv
= strv_free(c
->argv
);
3935 void exec_command_done_array(ExecCommand
*c
, size_t n
) {
3938 for (i
= 0; i
< n
; i
++)
3939 exec_command_done(c
+i
);
3942 ExecCommand
* exec_command_free_list(ExecCommand
*c
) {
3946 LIST_REMOVE(command
, c
, i
);
3947 exec_command_done(i
);
3954 void exec_command_free_array(ExecCommand
**c
, size_t n
) {
3957 for (i
= 0; i
< n
; i
++)
3958 c
[i
] = exec_command_free_list(c
[i
]);
3961 void exec_command_reset_status_array(ExecCommand
*c
, size_t n
) {
3964 for (i
= 0; i
< n
; i
++)
3965 exec_status_reset(&c
[i
].exec_status
);
3968 void exec_command_reset_status_list_array(ExecCommand
**c
, size_t n
) {
3971 for (i
= 0; i
< n
; i
++) {
3974 LIST_FOREACH(command
, z
, c
[i
])
3975 exec_status_reset(&z
->exec_status
);
3979 typedef struct InvalidEnvInfo
{
3984 static void invalid_env(const char *p
, void *userdata
) {
3985 InvalidEnvInfo
*info
= userdata
;
3987 log_unit_error(info
->unit
, "Ignoring invalid environment assignment '%s': %s", p
, info
->path
);
3990 const char* exec_context_fdname(const ExecContext
*c
, int fd_index
) {
3996 if (c
->std_input
!= EXEC_INPUT_NAMED_FD
)
3999 return c
->stdio_fdname
[STDIN_FILENO
] ?: "stdin";
4002 if (c
->std_output
!= EXEC_OUTPUT_NAMED_FD
)
4005 return c
->stdio_fdname
[STDOUT_FILENO
] ?: "stdout";
4008 if (c
->std_error
!= EXEC_OUTPUT_NAMED_FD
)
4011 return c
->stdio_fdname
[STDERR_FILENO
] ?: "stderr";
4018 static int exec_context_named_iofds(const ExecContext
*c
, const ExecParameters
*p
, int named_iofds
[static 3]) {
4020 const char* stdio_fdname
[3];
4026 targets
= (c
->std_input
== EXEC_INPUT_NAMED_FD
) +
4027 (c
->std_output
== EXEC_OUTPUT_NAMED_FD
) +
4028 (c
->std_error
== EXEC_OUTPUT_NAMED_FD
);
4030 for (i
= 0; i
< 3; i
++)
4031 stdio_fdname
[i
] = exec_context_fdname(c
, i
);
4033 n_fds
= p
->n_storage_fds
+ p
->n_socket_fds
;
4035 for (i
= 0; i
< n_fds
&& targets
> 0; i
++)
4036 if (named_iofds
[STDIN_FILENO
] < 0 &&
4037 c
->std_input
== EXEC_INPUT_NAMED_FD
&&
4038 stdio_fdname
[STDIN_FILENO
] &&
4039 streq(p
->fd_names
[i
], stdio_fdname
[STDIN_FILENO
])) {
4041 named_iofds
[STDIN_FILENO
] = p
->fds
[i
];
4044 } else if (named_iofds
[STDOUT_FILENO
] < 0 &&
4045 c
->std_output
== EXEC_OUTPUT_NAMED_FD
&&
4046 stdio_fdname
[STDOUT_FILENO
] &&
4047 streq(p
->fd_names
[i
], stdio_fdname
[STDOUT_FILENO
])) {
4049 named_iofds
[STDOUT_FILENO
] = p
->fds
[i
];
4052 } else if (named_iofds
[STDERR_FILENO
] < 0 &&
4053 c
->std_error
== EXEC_OUTPUT_NAMED_FD
&&
4054 stdio_fdname
[STDERR_FILENO
] &&
4055 streq(p
->fd_names
[i
], stdio_fdname
[STDERR_FILENO
])) {
4057 named_iofds
[STDERR_FILENO
] = p
->fds
[i
];
4061 return targets
== 0 ? 0 : -ENOENT
;
4064 static int exec_context_load_environment(const Unit
*unit
, const ExecContext
*c
, char ***l
) {
4065 char **i
, **r
= NULL
;
4070 STRV_FOREACH(i
, c
->environment_files
) {
4074 bool ignore
= false;
4076 _cleanup_globfree_ glob_t pglob
= {};
4085 if (!path_is_absolute(fn
)) {
4093 /* Filename supports globbing, take all matching files */
4094 k
= safe_glob(fn
, 0, &pglob
);
4103 /* When we don't match anything, -ENOENT should be returned */
4104 assert(pglob
.gl_pathc
> 0);
4106 for (n
= 0; n
< pglob
.gl_pathc
; n
++) {
4107 k
= load_env_file(NULL
, pglob
.gl_pathv
[n
], &p
);
4115 /* Log invalid environment variables with filename */
4117 InvalidEnvInfo info
= {
4119 .path
= pglob
.gl_pathv
[n
]
4122 p
= strv_env_clean_with_callback(p
, invalid_env
, &info
);
4130 m
= strv_env_merge(2, r
, p
);
4146 static bool tty_may_match_dev_console(const char *tty
) {
4147 _cleanup_free_
char *resolved
= NULL
;
4152 tty
= skip_dev_prefix(tty
);
4154 /* trivial identity? */
4155 if (streq(tty
, "console"))
4158 if (resolve_dev_console(&resolved
) < 0)
4159 return true; /* if we could not resolve, assume it may */
4161 /* "tty0" means the active VC, so it may be the same sometimes */
4162 return path_equal(resolved
, tty
) || (streq(resolved
, "tty0") && tty_is_vc(tty
));
4165 static bool exec_context_may_touch_tty(const ExecContext
*ec
) {
4168 return ec
->tty_reset
||
4170 ec
->tty_vt_disallocate
||
4171 is_terminal_input(ec
->std_input
) ||
4172 is_terminal_output(ec
->std_output
) ||
4173 is_terminal_output(ec
->std_error
);
4176 bool exec_context_may_touch_console(const ExecContext
*ec
) {
4178 return exec_context_may_touch_tty(ec
) &&
4179 tty_may_match_dev_console(exec_context_tty_path(ec
));
4182 static void strv_fprintf(FILE *f
, char **l
) {
4188 fprintf(f
, " %s", *g
);
4191 void exec_context_dump(const ExecContext
*c
, FILE* f
, const char *prefix
) {
4192 ExecDirectoryType dt
;
4200 prefix
= strempty(prefix
);
4204 "%sWorkingDirectory: %s\n"
4205 "%sRootDirectory: %s\n"
4206 "%sNonBlocking: %s\n"
4207 "%sPrivateTmp: %s\n"
4208 "%sPrivateDevices: %s\n"
4209 "%sProtectKernelTunables: %s\n"
4210 "%sProtectKernelModules: %s\n"
4211 "%sProtectControlGroups: %s\n"
4212 "%sPrivateNetwork: %s\n"
4213 "%sPrivateUsers: %s\n"
4214 "%sProtectHome: %s\n"
4215 "%sProtectSystem: %s\n"
4216 "%sMountAPIVFS: %s\n"
4217 "%sIgnoreSIGPIPE: %s\n"
4218 "%sMemoryDenyWriteExecute: %s\n"
4219 "%sRestrictRealtime: %s\n"
4220 "%sKeyringMode: %s\n"
4221 "%sProtectHostname: %s\n",
4223 prefix
, c
->working_directory
? c
->working_directory
: "/",
4224 prefix
, c
->root_directory
? c
->root_directory
: "/",
4225 prefix
, yes_no(c
->non_blocking
),
4226 prefix
, yes_no(c
->private_tmp
),
4227 prefix
, yes_no(c
->private_devices
),
4228 prefix
, yes_no(c
->protect_kernel_tunables
),
4229 prefix
, yes_no(c
->protect_kernel_modules
),
4230 prefix
, yes_no(c
->protect_control_groups
),
4231 prefix
, yes_no(c
->private_network
),
4232 prefix
, yes_no(c
->private_users
),
4233 prefix
, protect_home_to_string(c
->protect_home
),
4234 prefix
, protect_system_to_string(c
->protect_system
),
4235 prefix
, yes_no(c
->mount_apivfs
),
4236 prefix
, yes_no(c
->ignore_sigpipe
),
4237 prefix
, yes_no(c
->memory_deny_write_execute
),
4238 prefix
, yes_no(c
->restrict_realtime
),
4239 prefix
, exec_keyring_mode_to_string(c
->keyring_mode
),
4240 prefix
, yes_no(c
->protect_hostname
));
4243 fprintf(f
, "%sRootImage: %s\n", prefix
, c
->root_image
);
4245 STRV_FOREACH(e
, c
->environment
)
4246 fprintf(f
, "%sEnvironment: %s\n", prefix
, *e
);
4248 STRV_FOREACH(e
, c
->environment_files
)
4249 fprintf(f
, "%sEnvironmentFile: %s\n", prefix
, *e
);
4251 STRV_FOREACH(e
, c
->pass_environment
)
4252 fprintf(f
, "%sPassEnvironment: %s\n", prefix
, *e
);
4254 STRV_FOREACH(e
, c
->unset_environment
)
4255 fprintf(f
, "%sUnsetEnvironment: %s\n", prefix
, *e
);
4257 fprintf(f
, "%sRuntimeDirectoryPreserve: %s\n", prefix
, exec_preserve_mode_to_string(c
->runtime_directory_preserve_mode
));
4259 for (dt
= 0; dt
< _EXEC_DIRECTORY_TYPE_MAX
; dt
++) {
4260 fprintf(f
, "%s%sMode: %04o\n", prefix
, exec_directory_type_to_string(dt
), c
->directories
[dt
].mode
);
4262 STRV_FOREACH(d
, c
->directories
[dt
].paths
)
4263 fprintf(f
, "%s%s: %s\n", prefix
, exec_directory_type_to_string(dt
), *d
);
4271 if (c
->oom_score_adjust_set
)
4273 "%sOOMScoreAdjust: %i\n",
4274 prefix
, c
->oom_score_adjust
);
4276 for (i
= 0; i
< RLIM_NLIMITS
; i
++)
4278 fprintf(f
, "%sLimit%s: " RLIM_FMT
"\n",
4279 prefix
, rlimit_to_string(i
), c
->rlimit
[i
]->rlim_max
);
4280 fprintf(f
, "%sLimit%sSoft: " RLIM_FMT
"\n",
4281 prefix
, rlimit_to_string(i
), c
->rlimit
[i
]->rlim_cur
);
4284 if (c
->ioprio_set
) {
4285 _cleanup_free_
char *class_str
= NULL
;
4287 r
= ioprio_class_to_string_alloc(IOPRIO_PRIO_CLASS(c
->ioprio
), &class_str
);
4289 fprintf(f
, "%sIOSchedulingClass: %s\n", prefix
, class_str
);
4291 fprintf(f
, "%sIOPriority: %lu\n", prefix
, IOPRIO_PRIO_DATA(c
->ioprio
));
4294 if (c
->cpu_sched_set
) {
4295 _cleanup_free_
char *policy_str
= NULL
;
4297 r
= sched_policy_to_string_alloc(c
->cpu_sched_policy
, &policy_str
);
4299 fprintf(f
, "%sCPUSchedulingPolicy: %s\n", prefix
, policy_str
);
4302 "%sCPUSchedulingPriority: %i\n"
4303 "%sCPUSchedulingResetOnFork: %s\n",
4304 prefix
, c
->cpu_sched_priority
,
4305 prefix
, yes_no(c
->cpu_sched_reset_on_fork
));
4309 fprintf(f
, "%sCPUAffinity:", prefix
);
4310 for (i
= 0; i
< c
->cpuset_ncpus
; i
++)
4311 if (CPU_ISSET_S(i
, CPU_ALLOC_SIZE(c
->cpuset_ncpus
), c
->cpuset
))
4312 fprintf(f
, " %u", i
);
4316 if (c
->timer_slack_nsec
!= NSEC_INFINITY
)
4317 fprintf(f
, "%sTimerSlackNSec: "NSEC_FMT
"\n", prefix
, c
->timer_slack_nsec
);
4320 "%sStandardInput: %s\n"
4321 "%sStandardOutput: %s\n"
4322 "%sStandardError: %s\n",
4323 prefix
, exec_input_to_string(c
->std_input
),
4324 prefix
, exec_output_to_string(c
->std_output
),
4325 prefix
, exec_output_to_string(c
->std_error
));
4327 if (c
->std_input
== EXEC_INPUT_NAMED_FD
)
4328 fprintf(f
, "%sStandardInputFileDescriptorName: %s\n", prefix
, c
->stdio_fdname
[STDIN_FILENO
]);
4329 if (c
->std_output
== EXEC_OUTPUT_NAMED_FD
)
4330 fprintf(f
, "%sStandardOutputFileDescriptorName: %s\n", prefix
, c
->stdio_fdname
[STDOUT_FILENO
]);
4331 if (c
->std_error
== EXEC_OUTPUT_NAMED_FD
)
4332 fprintf(f
, "%sStandardErrorFileDescriptorName: %s\n", prefix
, c
->stdio_fdname
[STDERR_FILENO
]);
4334 if (c
->std_input
== EXEC_INPUT_FILE
)
4335 fprintf(f
, "%sStandardInputFile: %s\n", prefix
, c
->stdio_file
[STDIN_FILENO
]);
4336 if (c
->std_output
== EXEC_OUTPUT_FILE
)
4337 fprintf(f
, "%sStandardOutputFile: %s\n", prefix
, c
->stdio_file
[STDOUT_FILENO
]);
4338 if (c
->std_output
== EXEC_OUTPUT_FILE_APPEND
)
4339 fprintf(f
, "%sStandardOutputFileToAppend: %s\n", prefix
, c
->stdio_file
[STDOUT_FILENO
]);
4340 if (c
->std_error
== EXEC_OUTPUT_FILE
)
4341 fprintf(f
, "%sStandardErrorFile: %s\n", prefix
, c
->stdio_file
[STDERR_FILENO
]);
4342 if (c
->std_error
== EXEC_OUTPUT_FILE_APPEND
)
4343 fprintf(f
, "%sStandardErrorFileToAppend: %s\n", prefix
, c
->stdio_file
[STDERR_FILENO
]);
4349 "%sTTYVHangup: %s\n"
4350 "%sTTYVTDisallocate: %s\n",
4351 prefix
, c
->tty_path
,
4352 prefix
, yes_no(c
->tty_reset
),
4353 prefix
, yes_no(c
->tty_vhangup
),
4354 prefix
, yes_no(c
->tty_vt_disallocate
));
4356 if (IN_SET(c
->std_output
,
4359 EXEC_OUTPUT_JOURNAL
,
4360 EXEC_OUTPUT_SYSLOG_AND_CONSOLE
,
4361 EXEC_OUTPUT_KMSG_AND_CONSOLE
,
4362 EXEC_OUTPUT_JOURNAL_AND_CONSOLE
) ||
4363 IN_SET(c
->std_error
,
4366 EXEC_OUTPUT_JOURNAL
,
4367 EXEC_OUTPUT_SYSLOG_AND_CONSOLE
,
4368 EXEC_OUTPUT_KMSG_AND_CONSOLE
,
4369 EXEC_OUTPUT_JOURNAL_AND_CONSOLE
)) {
4371 _cleanup_free_
char *fac_str
= NULL
, *lvl_str
= NULL
;
4373 r
= log_facility_unshifted_to_string_alloc(c
->syslog_priority
>> 3, &fac_str
);
4375 fprintf(f
, "%sSyslogFacility: %s\n", prefix
, fac_str
);
4377 r
= log_level_to_string_alloc(LOG_PRI(c
->syslog_priority
), &lvl_str
);
4379 fprintf(f
, "%sSyslogLevel: %s\n", prefix
, lvl_str
);
4382 if (c
->log_level_max
>= 0) {
4383 _cleanup_free_
char *t
= NULL
;
4385 (void) log_level_to_string_alloc(c
->log_level_max
, &t
);
4387 fprintf(f
, "%sLogLevelMax: %s\n", prefix
, strna(t
));
4390 if (c
->log_rate_limit_interval_usec
> 0) {
4391 char buf_timespan
[FORMAT_TIMESPAN_MAX
];
4394 "%sLogRateLimitIntervalSec: %s\n",
4395 prefix
, format_timespan(buf_timespan
, sizeof(buf_timespan
), c
->log_rate_limit_interval_usec
, USEC_PER_SEC
));
4398 if (c
->log_rate_limit_burst
> 0)
4399 fprintf(f
, "%sLogRateLimitBurst: %u\n", prefix
, c
->log_rate_limit_burst
);
4401 if (c
->n_log_extra_fields
> 0) {
4404 for (j
= 0; j
< c
->n_log_extra_fields
; j
++) {
4405 fprintf(f
, "%sLogExtraFields: ", prefix
);
4406 fwrite(c
->log_extra_fields
[j
].iov_base
,
4407 1, c
->log_extra_fields
[j
].iov_len
,
4413 if (c
->secure_bits
) {
4414 _cleanup_free_
char *str
= NULL
;
4416 r
= secure_bits_to_string_alloc(c
->secure_bits
, &str
);
4418 fprintf(f
, "%sSecure Bits: %s\n", prefix
, str
);
4421 if (c
->capability_bounding_set
!= CAP_ALL
) {
4422 _cleanup_free_
char *str
= NULL
;
4424 r
= capability_set_to_string_alloc(c
->capability_bounding_set
, &str
);
4426 fprintf(f
, "%sCapabilityBoundingSet: %s\n", prefix
, str
);
4429 if (c
->capability_ambient_set
!= 0) {
4430 _cleanup_free_
char *str
= NULL
;
4432 r
= capability_set_to_string_alloc(c
->capability_ambient_set
, &str
);
4434 fprintf(f
, "%sAmbientCapabilities: %s\n", prefix
, str
);
4438 fprintf(f
, "%sUser: %s\n", prefix
, c
->user
);
4440 fprintf(f
, "%sGroup: %s\n", prefix
, c
->group
);
4442 fprintf(f
, "%sDynamicUser: %s\n", prefix
, yes_no(c
->dynamic_user
));
4444 if (!strv_isempty(c
->supplementary_groups
)) {
4445 fprintf(f
, "%sSupplementaryGroups:", prefix
);
4446 strv_fprintf(f
, c
->supplementary_groups
);
4451 fprintf(f
, "%sPAMName: %s\n", prefix
, c
->pam_name
);
4453 if (!strv_isempty(c
->read_write_paths
)) {
4454 fprintf(f
, "%sReadWritePaths:", prefix
);
4455 strv_fprintf(f
, c
->read_write_paths
);
4459 if (!strv_isempty(c
->read_only_paths
)) {
4460 fprintf(f
, "%sReadOnlyPaths:", prefix
);
4461 strv_fprintf(f
, c
->read_only_paths
);
4465 if (!strv_isempty(c
->inaccessible_paths
)) {
4466 fprintf(f
, "%sInaccessiblePaths:", prefix
);
4467 strv_fprintf(f
, c
->inaccessible_paths
);
4471 if (c
->n_bind_mounts
> 0)
4472 for (i
= 0; i
< c
->n_bind_mounts
; i
++)
4473 fprintf(f
, "%s%s: %s%s:%s:%s\n", prefix
,
4474 c
->bind_mounts
[i
].read_only
? "BindReadOnlyPaths" : "BindPaths",
4475 c
->bind_mounts
[i
].ignore_enoent
? "-": "",
4476 c
->bind_mounts
[i
].source
,
4477 c
->bind_mounts
[i
].destination
,
4478 c
->bind_mounts
[i
].recursive
? "rbind" : "norbind");
4480 if (c
->n_temporary_filesystems
> 0)
4481 for (i
= 0; i
< c
->n_temporary_filesystems
; i
++) {
4482 TemporaryFileSystem
*t
= c
->temporary_filesystems
+ i
;
4484 fprintf(f
, "%sTemporaryFileSystem: %s%s%s\n", prefix
,
4486 isempty(t
->options
) ? "" : ":",
4487 strempty(t
->options
));
4492 "%sUtmpIdentifier: %s\n",
4493 prefix
, c
->utmp_id
);
4495 if (c
->selinux_context
)
4497 "%sSELinuxContext: %s%s\n",
4498 prefix
, c
->selinux_context_ignore
? "-" : "", c
->selinux_context
);
4500 if (c
->apparmor_profile
)
4502 "%sAppArmorProfile: %s%s\n",
4503 prefix
, c
->apparmor_profile_ignore
? "-" : "", c
->apparmor_profile
);
4505 if (c
->smack_process_label
)
4507 "%sSmackProcessLabel: %s%s\n",
4508 prefix
, c
->smack_process_label_ignore
? "-" : "", c
->smack_process_label
);
4510 if (c
->personality
!= PERSONALITY_INVALID
)
4512 "%sPersonality: %s\n",
4513 prefix
, strna(personality_to_string(c
->personality
)));
4516 "%sLockPersonality: %s\n",
4517 prefix
, yes_no(c
->lock_personality
));
4519 if (c
->syscall_filter
) {
4527 "%sSystemCallFilter: ",
4530 if (!c
->syscall_whitelist
)
4534 HASHMAP_FOREACH_KEY(val
, id
, c
->syscall_filter
, j
) {
4535 _cleanup_free_
char *name
= NULL
;
4536 const char *errno_name
= NULL
;
4537 int num
= PTR_TO_INT(val
);
4544 name
= seccomp_syscall_resolve_num_arch(SCMP_ARCH_NATIVE
, PTR_TO_INT(id
) - 1);
4545 fputs(strna(name
), f
);
4548 errno_name
= errno_to_name(num
);
4550 fprintf(f
, ":%s", errno_name
);
4552 fprintf(f
, ":%d", num
);
4560 if (c
->syscall_archs
) {
4567 "%sSystemCallArchitectures:",
4571 SET_FOREACH(id
, c
->syscall_archs
, j
)
4572 fprintf(f
, " %s", strna(seccomp_arch_to_string(PTR_TO_UINT32(id
) - 1)));
4577 if (exec_context_restrict_namespaces_set(c
)) {
4578 _cleanup_free_
char *s
= NULL
;
4580 r
= namespace_flags_to_string(c
->restrict_namespaces
, &s
);
4582 fprintf(f
, "%sRestrictNamespaces: %s\n",
4586 if (c
->network_namespace_path
)
4588 "%sNetworkNamespacePath: %s\n",
4589 prefix
, c
->network_namespace_path
);
4591 if (c
->syscall_errno
> 0) {
4592 const char *errno_name
;
4594 fprintf(f
, "%sSystemCallErrorNumber: ", prefix
);
4596 errno_name
= errno_to_name(c
->syscall_errno
);
4598 fprintf(f
, "%s\n", errno_name
);
4600 fprintf(f
, "%d\n", c
->syscall_errno
);
4604 bool exec_context_maintains_privileges(const ExecContext
*c
) {
4607 /* Returns true if the process forked off would run under
4608 * an unchanged UID or as root. */
4613 if (streq(c
->user
, "root") || streq(c
->user
, "0"))
4619 int exec_context_get_effective_ioprio(const ExecContext
*c
) {
4627 p
= ioprio_get(IOPRIO_WHO_PROCESS
, 0);
4629 return IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE
, 4);
4634 void exec_context_free_log_extra_fields(ExecContext
*c
) {
4639 for (l
= 0; l
< c
->n_log_extra_fields
; l
++)
4640 free(c
->log_extra_fields
[l
].iov_base
);
4641 c
->log_extra_fields
= mfree(c
->log_extra_fields
);
4642 c
->n_log_extra_fields
= 0;
4645 void exec_context_revert_tty(ExecContext
*c
) {
4650 /* First, reset the TTY (possibly kicking everybody else from the TTY) */
4651 exec_context_tty_reset(c
, NULL
);
4653 /* And then undo what chown_terminal() did earlier. Note that we only do this if we have a path
4654 * configured. If the TTY was passed to us as file descriptor we assume the TTY is opened and managed
4655 * by whoever passed it to us and thus knows better when and how to chmod()/chown() it back. */
4657 if (exec_context_may_touch_tty(c
)) {
4660 path
= exec_context_tty_path(c
);
4662 r
= chmod_and_chown(path
, TTY_MODE
, 0, TTY_GID
);
4663 if (r
< 0 && r
!= -ENOENT
)
4664 log_warning_errno(r
, "Failed to reset TTY ownership/access mode of %s, ignoring: %m", path
);
4669 void exec_status_start(ExecStatus
*s
, pid_t pid
) {
4676 dual_timestamp_get(&s
->start_timestamp
);
4679 void exec_status_exit(ExecStatus
*s
, const ExecContext
*context
, pid_t pid
, int code
, int status
) {
4682 if (s
->pid
!= pid
) {
4688 dual_timestamp_get(&s
->exit_timestamp
);
4693 if (context
&& context
->utmp_id
)
4694 (void) utmp_put_dead_process(context
->utmp_id
, pid
, code
, status
);
4697 void exec_status_reset(ExecStatus
*s
) {
4700 *s
= (ExecStatus
) {};
4703 void exec_status_dump(const ExecStatus
*s
, FILE *f
, const char *prefix
) {
4704 char buf
[FORMAT_TIMESTAMP_MAX
];
4712 prefix
= strempty(prefix
);
4715 "%sPID: "PID_FMT
"\n",
4718 if (dual_timestamp_is_set(&s
->start_timestamp
))
4720 "%sStart Timestamp: %s\n",
4721 prefix
, format_timestamp(buf
, sizeof(buf
), s
->start_timestamp
.realtime
));
4723 if (dual_timestamp_is_set(&s
->exit_timestamp
))
4725 "%sExit Timestamp: %s\n"
4727 "%sExit Status: %i\n",
4728 prefix
, format_timestamp(buf
, sizeof(buf
), s
->exit_timestamp
.realtime
),
4729 prefix
, sigchld_code_to_string(s
->code
),
4733 static char *exec_command_line(char **argv
) {
4741 STRV_FOREACH(a
, argv
)
4749 STRV_FOREACH(a
, argv
) {
4756 if (strpbrk(*a
, WHITESPACE
)) {
4767 /* FIXME: this doesn't really handle arguments that have
4768 * spaces and ticks in them */
4773 static void exec_command_dump(ExecCommand
*c
, FILE *f
, const char *prefix
) {
4774 _cleanup_free_
char *cmd
= NULL
;
4775 const char *prefix2
;
4780 prefix
= strempty(prefix
);
4781 prefix2
= strjoina(prefix
, "\t");
4783 cmd
= exec_command_line(c
->argv
);
4785 "%sCommand Line: %s\n",
4786 prefix
, cmd
? cmd
: strerror(ENOMEM
));
4788 exec_status_dump(&c
->exec_status
, f
, prefix2
);
4791 void exec_command_dump_list(ExecCommand
*c
, FILE *f
, const char *prefix
) {
4794 prefix
= strempty(prefix
);
4796 LIST_FOREACH(command
, c
, c
)
4797 exec_command_dump(c
, f
, prefix
);
4800 void exec_command_append_list(ExecCommand
**l
, ExecCommand
*e
) {
4807 /* It's kind of important, that we keep the order here */
4808 LIST_FIND_TAIL(command
, *l
, end
);
4809 LIST_INSERT_AFTER(command
, *l
, end
, e
);
4814 int exec_command_set(ExecCommand
*c
, const char *path
, ...) {
4822 l
= strv_new_ap(path
, ap
);
4834 free_and_replace(c
->path
, p
);
4836 return strv_free_and_replace(c
->argv
, l
);
4839 int exec_command_append(ExecCommand
*c
, const char *path
, ...) {
4840 _cleanup_strv_free_
char **l
= NULL
;
4848 l
= strv_new_ap(path
, ap
);
4854 r
= strv_extend_strv(&c
->argv
, l
, false);
4861 static void *remove_tmpdir_thread(void *p
) {
4862 _cleanup_free_
char *path
= p
;
4864 (void) rm_rf(path
, REMOVE_ROOT
|REMOVE_PHYSICAL
);
4868 static ExecRuntime
* exec_runtime_free(ExecRuntime
*rt
, bool destroy
) {
4875 (void) hashmap_remove(rt
->manager
->exec_runtime_by_id
, rt
->id
);
4877 /* When destroy is true, then rm_rf tmp_dir and var_tmp_dir. */
4878 if (destroy
&& rt
->tmp_dir
) {
4879 log_debug("Spawning thread to nuke %s", rt
->tmp_dir
);
4881 r
= asynchronous_job(remove_tmpdir_thread
, rt
->tmp_dir
);
4883 log_warning_errno(r
, "Failed to nuke %s: %m", rt
->tmp_dir
);
4890 if (destroy
&& rt
->var_tmp_dir
) {
4891 log_debug("Spawning thread to nuke %s", rt
->var_tmp_dir
);
4893 r
= asynchronous_job(remove_tmpdir_thread
, rt
->var_tmp_dir
);
4895 log_warning_errno(r
, "Failed to nuke %s: %m", rt
->var_tmp_dir
);
4896 free(rt
->var_tmp_dir
);
4899 rt
->var_tmp_dir
= NULL
;
4902 rt
->id
= mfree(rt
->id
);
4903 rt
->tmp_dir
= mfree(rt
->tmp_dir
);
4904 rt
->var_tmp_dir
= mfree(rt
->var_tmp_dir
);
4905 safe_close_pair(rt
->netns_storage_socket
);
4909 static void exec_runtime_freep(ExecRuntime
**rt
) {
4910 (void) exec_runtime_free(*rt
, false);
4913 static int exec_runtime_allocate(ExecRuntime
**ret
) {
4918 n
= new(ExecRuntime
, 1);
4922 *n
= (ExecRuntime
) {
4923 .netns_storage_socket
= { -1, -1 },
4930 static int exec_runtime_add(
4933 const char *tmp_dir
,
4934 const char *var_tmp_dir
,
4935 const int netns_storage_socket
[2],
4936 ExecRuntime
**ret
) {
4938 _cleanup_(exec_runtime_freep
) ExecRuntime
*rt
= NULL
;
4944 r
= hashmap_ensure_allocated(&m
->exec_runtime_by_id
, &string_hash_ops
);
4948 r
= exec_runtime_allocate(&rt
);
4952 rt
->id
= strdup(id
);
4957 rt
->tmp_dir
= strdup(tmp_dir
);
4961 /* When tmp_dir is set, then we require var_tmp_dir is also set. */
4962 assert(var_tmp_dir
);
4963 rt
->var_tmp_dir
= strdup(var_tmp_dir
);
4964 if (!rt
->var_tmp_dir
)
4968 if (netns_storage_socket
) {
4969 rt
->netns_storage_socket
[0] = netns_storage_socket
[0];
4970 rt
->netns_storage_socket
[1] = netns_storage_socket
[1];
4973 r
= hashmap_put(m
->exec_runtime_by_id
, rt
->id
, rt
);
4982 /* do not remove created ExecRuntime object when the operation succeeds. */
4987 static int exec_runtime_make(Manager
*m
, const ExecContext
*c
, const char *id
, ExecRuntime
**ret
) {
4988 _cleanup_free_
char *tmp_dir
= NULL
, *var_tmp_dir
= NULL
;
4989 _cleanup_close_pair_
int netns_storage_socket
[2] = { -1, -1 };
4996 /* It is not necessary to create ExecRuntime object. */
4997 if (!c
->private_network
&& !c
->private_tmp
&& !c
->network_namespace_path
)
5000 if (c
->private_tmp
) {
5001 r
= setup_tmp_dirs(id
, &tmp_dir
, &var_tmp_dir
);
5006 if (c
->private_network
|| c
->network_namespace_path
) {
5007 if (socketpair(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
, 0, netns_storage_socket
) < 0)
5011 r
= exec_runtime_add(m
, id
, tmp_dir
, var_tmp_dir
, netns_storage_socket
, ret
);
5016 netns_storage_socket
[0] = netns_storage_socket
[1] = -1;
5020 int exec_runtime_acquire(Manager
*m
, const ExecContext
*c
, const char *id
, bool create
, ExecRuntime
**ret
) {
5028 rt
= hashmap_get(m
->exec_runtime_by_id
, id
);
5030 /* We already have a ExecRuntime object, let's increase the ref count and reuse it */
5036 /* If not found, then create a new object. */
5037 r
= exec_runtime_make(m
, c
, id
, &rt
);
5039 /* When r == 0, it is not necessary to create ExecRuntime object. */
5043 /* increment reference counter. */
5049 ExecRuntime
*exec_runtime_unref(ExecRuntime
*rt
, bool destroy
) {
5053 assert(rt
->n_ref
> 0);
5059 return exec_runtime_free(rt
, destroy
);
5062 int exec_runtime_serialize(const Manager
*m
, FILE *f
, FDSet
*fds
) {
5070 HASHMAP_FOREACH(rt
, m
->exec_runtime_by_id
, i
) {
5071 fprintf(f
, "exec-runtime=%s", rt
->id
);
5074 fprintf(f
, " tmp-dir=%s", rt
->tmp_dir
);
5076 if (rt
->var_tmp_dir
)
5077 fprintf(f
, " var-tmp-dir=%s", rt
->var_tmp_dir
);
5079 if (rt
->netns_storage_socket
[0] >= 0) {
5082 copy
= fdset_put_dup(fds
, rt
->netns_storage_socket
[0]);
5086 fprintf(f
, " netns-socket-0=%i", copy
);
5089 if (rt
->netns_storage_socket
[1] >= 0) {
5092 copy
= fdset_put_dup(fds
, rt
->netns_storage_socket
[1]);
5096 fprintf(f
, " netns-socket-1=%i", copy
);
5105 int exec_runtime_deserialize_compat(Unit
*u
, const char *key
, const char *value
, FDSet
*fds
) {
5106 _cleanup_(exec_runtime_freep
) ExecRuntime
*rt_create
= NULL
;
5110 /* This is for the migration from old (v237 or earlier) deserialization text.
5111 * Due to the bug #7790, this may not work with the units that use JoinsNamespaceOf=.
5112 * Even if the ExecRuntime object originally created by the other unit, we cannot judge
5113 * so or not from the serialized text, then we always creates a new object owned by this. */
5119 /* Manager manages ExecRuntime objects by the unit id.
5120 * So, we omit the serialized text when the unit does not have id (yet?)... */
5121 if (isempty(u
->id
)) {
5122 log_unit_debug(u
, "Invocation ID not found. Dropping runtime parameter.");
5126 r
= hashmap_ensure_allocated(&u
->manager
->exec_runtime_by_id
, &string_hash_ops
);
5128 log_unit_debug_errno(u
, r
, "Failed to allocate storage for runtime parameter: %m");
5132 rt
= hashmap_get(u
->manager
->exec_runtime_by_id
, u
->id
);
5134 r
= exec_runtime_allocate(&rt_create
);
5138 rt_create
->id
= strdup(u
->id
);
5145 if (streq(key
, "tmp-dir")) {
5148 copy
= strdup(value
);
5152 free_and_replace(rt
->tmp_dir
, copy
);
5154 } else if (streq(key
, "var-tmp-dir")) {
5157 copy
= strdup(value
);
5161 free_and_replace(rt
->var_tmp_dir
, copy
);
5163 } else if (streq(key
, "netns-socket-0")) {
5166 if (safe_atoi(value
, &fd
) < 0 || !fdset_contains(fds
, fd
)) {
5167 log_unit_debug(u
, "Failed to parse netns socket value: %s", value
);
5171 safe_close(rt
->netns_storage_socket
[0]);
5172 rt
->netns_storage_socket
[0] = fdset_remove(fds
, fd
);
5174 } else if (streq(key
, "netns-socket-1")) {
5177 if (safe_atoi(value
, &fd
) < 0 || !fdset_contains(fds
, fd
)) {
5178 log_unit_debug(u
, "Failed to parse netns socket value: %s", value
);
5182 safe_close(rt
->netns_storage_socket
[1]);
5183 rt
->netns_storage_socket
[1] = fdset_remove(fds
, fd
);
5187 /* If the object is newly created, then put it to the hashmap which manages ExecRuntime objects. */
5189 r
= hashmap_put(u
->manager
->exec_runtime_by_id
, rt_create
->id
, rt_create
);
5191 log_unit_debug_errno(u
, r
, "Failed to put runtime parameter to manager's storage: %m");
5195 rt_create
->manager
= u
->manager
;
5204 void exec_runtime_deserialize_one(Manager
*m
, const char *value
, FDSet
*fds
) {
5205 char *id
= NULL
, *tmp_dir
= NULL
, *var_tmp_dir
= NULL
;
5206 int r
, fd0
= -1, fd1
= -1;
5207 const char *p
, *v
= value
;
5214 n
= strcspn(v
, " ");
5215 id
= strndupa(v
, n
);
5220 v
= startswith(p
, "tmp-dir=");
5222 n
= strcspn(v
, " ");
5223 tmp_dir
= strndupa(v
, n
);
5229 v
= startswith(p
, "var-tmp-dir=");
5231 n
= strcspn(v
, " ");
5232 var_tmp_dir
= strndupa(v
, n
);
5238 v
= startswith(p
, "netns-socket-0=");
5242 n
= strcspn(v
, " ");
5243 buf
= strndupa(v
, n
);
5244 if (safe_atoi(buf
, &fd0
) < 0 || !fdset_contains(fds
, fd0
)) {
5245 log_debug("Unable to process exec-runtime netns fd specification.");
5248 fd0
= fdset_remove(fds
, fd0
);
5254 v
= startswith(p
, "netns-socket-1=");
5258 n
= strcspn(v
, " ");
5259 buf
= strndupa(v
, n
);
5260 if (safe_atoi(buf
, &fd1
) < 0 || !fdset_contains(fds
, fd1
)) {
5261 log_debug("Unable to process exec-runtime netns fd specification.");
5264 fd1
= fdset_remove(fds
, fd1
);
5269 r
= exec_runtime_add(m
, id
, tmp_dir
, var_tmp_dir
, (int[]) { fd0
, fd1
}, NULL
);
5271 log_debug_errno(r
, "Failed to add exec-runtime: %m");
5274 void exec_runtime_vacuum(Manager
*m
) {
5280 /* Free unreferenced ExecRuntime objects. This is used after manager deserialization process. */
5282 HASHMAP_FOREACH(rt
, m
->exec_runtime_by_id
, i
) {
5286 (void) exec_runtime_free(rt
, false);
5290 void exec_params_clear(ExecParameters
*p
) {
5294 strv_free(p
->environment
);
5297 static const char* const exec_input_table
[_EXEC_INPUT_MAX
] = {
5298 [EXEC_INPUT_NULL
] = "null",
5299 [EXEC_INPUT_TTY
] = "tty",
5300 [EXEC_INPUT_TTY_FORCE
] = "tty-force",
5301 [EXEC_INPUT_TTY_FAIL
] = "tty-fail",
5302 [EXEC_INPUT_SOCKET
] = "socket",
5303 [EXEC_INPUT_NAMED_FD
] = "fd",
5304 [EXEC_INPUT_DATA
] = "data",
5305 [EXEC_INPUT_FILE
] = "file",
5308 DEFINE_STRING_TABLE_LOOKUP(exec_input
, ExecInput
);
5310 static const char* const exec_output_table
[_EXEC_OUTPUT_MAX
] = {
5311 [EXEC_OUTPUT_INHERIT
] = "inherit",
5312 [EXEC_OUTPUT_NULL
] = "null",
5313 [EXEC_OUTPUT_TTY
] = "tty",
5314 [EXEC_OUTPUT_SYSLOG
] = "syslog",
5315 [EXEC_OUTPUT_SYSLOG_AND_CONSOLE
] = "syslog+console",
5316 [EXEC_OUTPUT_KMSG
] = "kmsg",
5317 [EXEC_OUTPUT_KMSG_AND_CONSOLE
] = "kmsg+console",
5318 [EXEC_OUTPUT_JOURNAL
] = "journal",
5319 [EXEC_OUTPUT_JOURNAL_AND_CONSOLE
] = "journal+console",
5320 [EXEC_OUTPUT_SOCKET
] = "socket",
5321 [EXEC_OUTPUT_NAMED_FD
] = "fd",
5322 [EXEC_OUTPUT_FILE
] = "file",
5323 [EXEC_OUTPUT_FILE_APPEND
] = "append",
5326 DEFINE_STRING_TABLE_LOOKUP(exec_output
, ExecOutput
);
5328 static const char* const exec_utmp_mode_table
[_EXEC_UTMP_MODE_MAX
] = {
5329 [EXEC_UTMP_INIT
] = "init",
5330 [EXEC_UTMP_LOGIN
] = "login",
5331 [EXEC_UTMP_USER
] = "user",
5334 DEFINE_STRING_TABLE_LOOKUP(exec_utmp_mode
, ExecUtmpMode
);
5336 static const char* const exec_preserve_mode_table
[_EXEC_PRESERVE_MODE_MAX
] = {
5337 [EXEC_PRESERVE_NO
] = "no",
5338 [EXEC_PRESERVE_YES
] = "yes",
5339 [EXEC_PRESERVE_RESTART
] = "restart",
5342 DEFINE_STRING_TABLE_LOOKUP_WITH_BOOLEAN(exec_preserve_mode
, ExecPreserveMode
, EXEC_PRESERVE_YES
);
5344 static const char* const exec_directory_type_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
5345 [EXEC_DIRECTORY_RUNTIME
] = "RuntimeDirectory",
5346 [EXEC_DIRECTORY_STATE
] = "StateDirectory",
5347 [EXEC_DIRECTORY_CACHE
] = "CacheDirectory",
5348 [EXEC_DIRECTORY_LOGS
] = "LogsDirectory",
5349 [EXEC_DIRECTORY_CONFIGURATION
] = "ConfigurationDirectory",
5352 DEFINE_STRING_TABLE_LOOKUP(exec_directory_type
, ExecDirectoryType
);
5354 static const char* const exec_directory_env_name_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
5355 [EXEC_DIRECTORY_RUNTIME
] = "RUNTIME_DIRECTORY",
5356 [EXEC_DIRECTORY_STATE
] = "STATE_DIRECTORY",
5357 [EXEC_DIRECTORY_CACHE
] = "CACHE_DIRECTORY",
5358 [EXEC_DIRECTORY_LOGS
] = "LOGS_DIRECTORY",
5359 [EXEC_DIRECTORY_CONFIGURATION
] = "CONFIGURATION_DIRECTORY",
5362 DEFINE_PRIVATE_STRING_TABLE_LOOKUP_TO_STRING(exec_directory_env_name
, ExecDirectoryType
);
5364 static const char* const exec_keyring_mode_table
[_EXEC_KEYRING_MODE_MAX
] = {
5365 [EXEC_KEYRING_INHERIT
] = "inherit",
5366 [EXEC_KEYRING_PRIVATE
] = "private",
5367 [EXEC_KEYRING_SHARED
] = "shared",
5370 DEFINE_STRING_TABLE_LOOKUP(exec_keyring_mode
, ExecKeyringMode
);