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
= strjoin(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 _cleanup_free_
char *private_root
= NULL
;
2091 /* So, here's one extra complication when dealing with DynamicUser=1 units. In that case we
2092 * want to avoid leaving a directory around fully accessible that is owned by a dynamic user
2093 * whose UID is later on reused. To lock this down we use the same trick used by container
2094 * managers to prohibit host users to get access to files of the same UID in containers: we
2095 * place everything inside a directory that has an access mode of 0700 and is owned root:root,
2096 * so that it acts as security boundary for unprivileged host code. We then use fs namespacing
2097 * to make this directory permeable for the service itself.
2099 * Specifically: for a service which wants a special directory "foo/" we first create a
2100 * directory "private/" with access mode 0700 owned by root:root. Then we place "foo" inside of
2101 * that directory (i.e. "private/foo/"), and make "foo" a symlink to "private/foo". This way,
2102 * privileged host users can access "foo/" as usual, but unprivileged host users can't look
2103 * into it. Inside of the namespaceof the container "private/" is replaced by a more liberally
2104 * accessible tmpfs, into which the host's "private/foo/" is mounted under the same name, thus
2105 * disabling the access boundary for the service and making sure it only gets access to the
2106 * dirs it needs but no others. Tricky? Yes, absolutely, but it works!
2108 * Note that we don't do this for EXEC_DIRECTORY_CONFIGURATION as that's assumed not to be
2109 * owned by the service itself.
2110 * Also, note that we don't do this for EXEC_DIRECTORY_RUNTIME as that's often used for sharing
2111 * files or sockets with other services. */
2113 private_root
= strjoin(params
->prefix
[type
], "/private");
2114 if (!private_root
) {
2119 /* First set up private root if it doesn't exist yet, with access mode 0700 and owned by root:root */
2120 r
= mkdir_safe_label(private_root
, 0700, 0, 0, MKDIR_WARN_MODE
);
2124 pp
= strjoin(private_root
, "/", *rt
);
2130 /* Create all directories between the configured directory and this private root, and mark them 0755 */
2131 r
= mkdir_parents_label(pp
, 0755);
2135 if (is_dir(p
, false) > 0 &&
2136 (laccess(pp
, F_OK
) < 0 && errno
== ENOENT
)) {
2138 /* Hmm, the private directory doesn't exist yet, but the normal one exists? If so, move
2139 * it over. Most likely the service has been upgraded from one that didn't use
2140 * DynamicUser=1, to one that does. */
2142 if (rename(p
, pp
) < 0) {
2147 /* Otherwise, create the actual directory for the service */
2149 r
= mkdir_label(pp
, context
->directories
[type
].mode
);
2150 if (r
< 0 && r
!= -EEXIST
)
2154 /* And link it up from the original place */
2155 r
= symlink_idempotent(pp
, p
, true);
2159 /* Lock down the access mode */
2160 if (chmod(pp
, context
->directories
[type
].mode
) < 0) {
2165 r
= mkdir_label(p
, context
->directories
[type
].mode
);
2166 if (r
< 0 && r
!= -EEXIST
)
2171 if (stat(p
, &st
) < 0) {
2175 if (((st
.st_mode
^ context
->directories
[type
].mode
) & 07777) != 0)
2176 log_warning("%s \'%s\' already exists but the mode is different. "
2177 "(filesystem: %o %sMode: %o)",
2178 exec_directory_type_to_string(type
), *rt
,
2179 st
.st_mode
& 07777, exec_directory_type_to_string(type
), context
->directories
[type
].mode
& 07777);
2180 if (!context
->dynamic_user
)
2185 /* Don't change the owner of the configuration directory, as in the common case it is not written to by
2186 * a service, and shall not be writable. */
2187 if (type
== EXEC_DIRECTORY_CONFIGURATION
)
2190 /* Then, change the ownership of the whole tree, if necessary */
2191 r
= path_chown_recursive(pp
?: p
, uid
, gid
);
2199 *exit_status
= exit_status_table
[type
];
2204 static int setup_smack(
2205 const ExecContext
*context
,
2206 const ExecCommand
*command
) {
2213 if (context
->smack_process_label
) {
2214 r
= mac_smack_apply_pid(0, context
->smack_process_label
);
2218 #ifdef SMACK_DEFAULT_PROCESS_LABEL
2220 _cleanup_free_
char *exec_label
= NULL
;
2222 r
= mac_smack_read(command
->path
, SMACK_ATTR_EXEC
, &exec_label
);
2223 if (r
< 0 && !IN_SET(r
, -ENODATA
, -EOPNOTSUPP
))
2226 r
= mac_smack_apply_pid(0, exec_label
? : SMACK_DEFAULT_PROCESS_LABEL
);
2236 static int compile_bind_mounts(
2237 const ExecContext
*context
,
2238 const ExecParameters
*params
,
2239 BindMount
**ret_bind_mounts
,
2240 size_t *ret_n_bind_mounts
,
2241 char ***ret_empty_directories
) {
2243 _cleanup_strv_free_
char **empty_directories
= NULL
;
2244 BindMount
*bind_mounts
;
2246 ExecDirectoryType t
;
2251 assert(ret_bind_mounts
);
2252 assert(ret_n_bind_mounts
);
2253 assert(ret_empty_directories
);
2255 n
= context
->n_bind_mounts
;
2256 for (t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
2257 if (!params
->prefix
[t
])
2260 n
+= strv_length(context
->directories
[t
].paths
);
2264 *ret_bind_mounts
= NULL
;
2265 *ret_n_bind_mounts
= 0;
2266 *ret_empty_directories
= NULL
;
2270 bind_mounts
= new(BindMount
, n
);
2274 for (i
= 0; i
< context
->n_bind_mounts
; i
++) {
2275 BindMount
*item
= context
->bind_mounts
+ i
;
2278 s
= strdup(item
->source
);
2284 d
= strdup(item
->destination
);
2291 bind_mounts
[h
++] = (BindMount
) {
2294 .read_only
= item
->read_only
,
2295 .recursive
= item
->recursive
,
2296 .ignore_enoent
= item
->ignore_enoent
,
2300 for (t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
2303 if (!params
->prefix
[t
])
2306 if (strv_isempty(context
->directories
[t
].paths
))
2309 if (context
->dynamic_user
&&
2310 !IN_SET(t
, EXEC_DIRECTORY_RUNTIME
, EXEC_DIRECTORY_CONFIGURATION
) &&
2311 !(context
->root_directory
|| context
->root_image
)) {
2314 /* So this is for a dynamic user, and we need to make sure the process can access its own
2315 * directory. For that we overmount the usually inaccessible "private" subdirectory with a
2316 * tmpfs that makes it accessible and is empty except for the submounts we do this for. */
2318 private_root
= strjoin(params
->prefix
[t
], "/private");
2319 if (!private_root
) {
2324 r
= strv_consume(&empty_directories
, private_root
);
2329 STRV_FOREACH(suffix
, context
->directories
[t
].paths
) {
2332 if (context
->dynamic_user
&&
2333 !IN_SET(t
, EXEC_DIRECTORY_RUNTIME
, EXEC_DIRECTORY_CONFIGURATION
))
2334 s
= strjoin(params
->prefix
[t
], "/private/", *suffix
);
2336 s
= strjoin(params
->prefix
[t
], "/", *suffix
);
2342 if (context
->dynamic_user
&&
2343 !IN_SET(t
, EXEC_DIRECTORY_RUNTIME
, EXEC_DIRECTORY_CONFIGURATION
) &&
2344 (context
->root_directory
|| context
->root_image
))
2345 /* When RootDirectory= or RootImage= are set, then the symbolic link to the private
2346 * directory is not created on the root directory. So, let's bind-mount the directory
2347 * on the 'non-private' place. */
2348 d
= strjoin(params
->prefix
[t
], "/", *suffix
);
2357 bind_mounts
[h
++] = (BindMount
) {
2362 .ignore_enoent
= false,
2369 *ret_bind_mounts
= bind_mounts
;
2370 *ret_n_bind_mounts
= n
;
2371 *ret_empty_directories
= TAKE_PTR(empty_directories
);
2376 bind_mount_free_many(bind_mounts
, h
);
2380 static int apply_mount_namespace(
2382 const ExecCommand
*command
,
2383 const ExecContext
*context
,
2384 const ExecParameters
*params
,
2385 const ExecRuntime
*runtime
) {
2387 _cleanup_strv_free_
char **empty_directories
= NULL
;
2388 char *tmp
= NULL
, *var
= NULL
;
2389 const char *root_dir
= NULL
, *root_image
= NULL
;
2390 NamespaceInfo ns_info
;
2391 bool needs_sandboxing
;
2392 BindMount
*bind_mounts
= NULL
;
2393 size_t n_bind_mounts
= 0;
2398 /* The runtime struct only contains the parent of the private /tmp,
2399 * which is non-accessible to world users. Inside of it there's a /tmp
2400 * that is sticky, and that's the one we want to use here. */
2402 if (context
->private_tmp
&& runtime
) {
2403 if (runtime
->tmp_dir
)
2404 tmp
= strjoina(runtime
->tmp_dir
, "/tmp");
2405 if (runtime
->var_tmp_dir
)
2406 var
= strjoina(runtime
->var_tmp_dir
, "/tmp");
2409 if (params
->flags
& EXEC_APPLY_CHROOT
) {
2410 root_image
= context
->root_image
;
2413 root_dir
= context
->root_directory
;
2416 r
= compile_bind_mounts(context
, params
, &bind_mounts
, &n_bind_mounts
, &empty_directories
);
2420 needs_sandboxing
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && !(command
->flags
& EXEC_COMMAND_FULLY_PRIVILEGED
);
2421 if (needs_sandboxing
)
2422 ns_info
= (NamespaceInfo
) {
2423 .ignore_protect_paths
= false,
2424 .private_dev
= context
->private_devices
,
2425 .protect_control_groups
= context
->protect_control_groups
,
2426 .protect_kernel_tunables
= context
->protect_kernel_tunables
,
2427 .protect_kernel_modules
= context
->protect_kernel_modules
,
2428 .protect_hostname
= context
->protect_hostname
,
2429 .mount_apivfs
= context
->mount_apivfs
,
2430 .private_mounts
= context
->private_mounts
,
2432 else if (!context
->dynamic_user
&& root_dir
)
2434 * If DynamicUser=no and RootDirectory= is set then lets pass a relaxed
2435 * sandbox info, otherwise enforce it, don't ignore protected paths and
2436 * fail if we are enable to apply the sandbox inside the mount namespace.
2438 ns_info
= (NamespaceInfo
) {
2439 .ignore_protect_paths
= true,
2442 ns_info
= (NamespaceInfo
) {};
2444 if (context
->mount_flags
== MS_SHARED
)
2445 log_unit_debug(u
, "shared mount propagation hidden by other fs namespacing unit settings: ignoring");
2447 r
= setup_namespace(root_dir
, root_image
,
2448 &ns_info
, context
->read_write_paths
,
2449 needs_sandboxing
? context
->read_only_paths
: NULL
,
2450 needs_sandboxing
? context
->inaccessible_paths
: NULL
,
2454 context
->temporary_filesystems
,
2455 context
->n_temporary_filesystems
,
2458 needs_sandboxing
? context
->protect_home
: PROTECT_HOME_NO
,
2459 needs_sandboxing
? context
->protect_system
: PROTECT_SYSTEM_NO
,
2460 context
->mount_flags
,
2461 DISSECT_IMAGE_DISCARD_ON_LOOP
);
2463 bind_mount_free_many(bind_mounts
, n_bind_mounts
);
2465 /* If we couldn't set up the namespace this is probably due to a missing capability. setup_namespace() reports
2466 * that with a special, recognizable error ENOANO. In this case, silently proceeed, but only if exclusively
2467 * sandboxing options were used, i.e. nothing such as RootDirectory= or BindMount= that would result in a
2468 * completely different execution environment. */
2470 if (n_bind_mounts
== 0 &&
2471 context
->n_temporary_filesystems
== 0 &&
2472 !root_dir
&& !root_image
&&
2473 !context
->dynamic_user
) {
2474 log_unit_debug(u
, "Failed to set up namespace, assuming containerized execution and ignoring.");
2478 log_unit_debug(u
, "Failed to set up namespace, and refusing to continue since the selected namespacing options alter mount environment non-trivially.\n"
2479 "Bind mounts: %zu, temporary filesystems: %zu, root directory: %s, root image: %s, dynamic user: %s",
2480 n_bind_mounts
, context
->n_temporary_filesystems
, yes_no(root_dir
), yes_no(root_image
), yes_no(context
->dynamic_user
));
2488 static int apply_working_directory(
2489 const ExecContext
*context
,
2490 const ExecParameters
*params
,
2492 const bool needs_mount_ns
,
2498 assert(exit_status
);
2500 if (context
->working_directory_home
) {
2503 *exit_status
= EXIT_CHDIR
;
2509 } else if (context
->working_directory
)
2510 wd
= context
->working_directory
;
2514 if (params
->flags
& EXEC_APPLY_CHROOT
) {
2515 if (!needs_mount_ns
&& context
->root_directory
)
2516 if (chroot(context
->root_directory
) < 0) {
2517 *exit_status
= EXIT_CHROOT
;
2523 d
= prefix_roota(context
->root_directory
, wd
);
2525 if (chdir(d
) < 0 && !context
->working_directory_missing_ok
) {
2526 *exit_status
= EXIT_CHDIR
;
2533 static int setup_keyring(
2535 const ExecContext
*context
,
2536 const ExecParameters
*p
,
2537 uid_t uid
, gid_t gid
) {
2539 key_serial_t keyring
;
2548 /* Let's set up a new per-service "session" kernel keyring for each system service. This has the benefit that
2549 * each service runs with its own keyring shared among all processes of the service, but with no hook-up beyond
2550 * that scope, and in particular no link to the per-UID keyring. If we don't do this the keyring will be
2551 * automatically created on-demand and then linked to the per-UID keyring, by the kernel. The kernel's built-in
2552 * on-demand behaviour is very appropriate for login users, but probably not so much for system services, where
2553 * UIDs are not necessarily specific to a service but reused (at least in the case of UID 0). */
2555 if (context
->keyring_mode
== EXEC_KEYRING_INHERIT
)
2558 /* Acquiring a reference to the user keyring is nasty. We briefly change identity in order to get things set up
2559 * properly by the kernel. If we don't do that then we can't create it atomically, and that sucks for parallel
2560 * execution. This mimics what pam_keyinit does, too. Setting up session keyring, to be owned by the right user
2561 * & group is just as nasty as acquiring a reference to the user keyring. */
2563 saved_uid
= getuid();
2564 saved_gid
= getgid();
2566 if (gid_is_valid(gid
) && gid
!= saved_gid
) {
2567 if (setregid(gid
, -1) < 0)
2568 return log_unit_error_errno(u
, errno
, "Failed to change GID for user keyring: %m");
2571 if (uid_is_valid(uid
) && uid
!= saved_uid
) {
2572 if (setreuid(uid
, -1) < 0) {
2573 r
= log_unit_error_errno(u
, errno
, "Failed to change UID for user keyring: %m");
2578 keyring
= keyctl(KEYCTL_JOIN_SESSION_KEYRING
, 0, 0, 0, 0);
2579 if (keyring
== -1) {
2580 if (errno
== ENOSYS
)
2581 log_unit_debug_errno(u
, errno
, "Kernel keyring not supported, ignoring.");
2582 else if (IN_SET(errno
, EACCES
, EPERM
))
2583 log_unit_debug_errno(u
, errno
, "Kernel keyring access prohibited, ignoring.");
2584 else if (errno
== EDQUOT
)
2585 log_unit_debug_errno(u
, errno
, "Out of kernel keyrings to allocate, ignoring.");
2587 r
= log_unit_error_errno(u
, errno
, "Setting up kernel keyring failed: %m");
2592 /* When requested link the user keyring into the session keyring. */
2593 if (context
->keyring_mode
== EXEC_KEYRING_SHARED
) {
2595 if (keyctl(KEYCTL_LINK
,
2596 KEY_SPEC_USER_KEYRING
,
2597 KEY_SPEC_SESSION_KEYRING
, 0, 0) < 0) {
2598 r
= log_unit_error_errno(u
, errno
, "Failed to link user keyring into session keyring: %m");
2603 /* Restore uid/gid back */
2604 if (uid_is_valid(uid
) && uid
!= saved_uid
) {
2605 if (setreuid(saved_uid
, -1) < 0) {
2606 r
= log_unit_error_errno(u
, errno
, "Failed to change UID back for user keyring: %m");
2611 if (gid_is_valid(gid
) && gid
!= saved_gid
) {
2612 if (setregid(saved_gid
, -1) < 0)
2613 return log_unit_error_errno(u
, errno
, "Failed to change GID back for user keyring: %m");
2616 /* Populate they keyring with the invocation ID by default, as original saved_uid. */
2617 if (!sd_id128_is_null(u
->invocation_id
)) {
2620 key
= add_key("user", "invocation_id", &u
->invocation_id
, sizeof(u
->invocation_id
), KEY_SPEC_SESSION_KEYRING
);
2622 log_unit_debug_errno(u
, errno
, "Failed to add invocation ID to keyring, ignoring: %m");
2624 if (keyctl(KEYCTL_SETPERM
, key
,
2625 KEY_POS_VIEW
|KEY_POS_READ
|KEY_POS_SEARCH
|
2626 KEY_USR_VIEW
|KEY_USR_READ
|KEY_USR_SEARCH
, 0, 0) < 0)
2627 r
= log_unit_error_errno(u
, errno
, "Failed to restrict invocation ID permission: %m");
2632 /* Revert back uid & gid for the the last time, and exit */
2633 /* no extra logging, as only the first already reported error matters */
2634 if (getuid() != saved_uid
)
2635 (void) setreuid(saved_uid
, -1);
2637 if (getgid() != saved_gid
)
2638 (void) setregid(saved_gid
, -1);
2643 static void append_socket_pair(int *array
, size_t *n
, const int pair
[static 2]) {
2651 array
[(*n
)++] = pair
[0];
2653 array
[(*n
)++] = pair
[1];
2656 static int close_remaining_fds(
2657 const ExecParameters
*params
,
2658 const ExecRuntime
*runtime
,
2659 const DynamicCreds
*dcreds
,
2663 int *fds
, size_t n_fds
) {
2665 size_t n_dont_close
= 0;
2666 int dont_close
[n_fds
+ 12];
2670 if (params
->stdin_fd
>= 0)
2671 dont_close
[n_dont_close
++] = params
->stdin_fd
;
2672 if (params
->stdout_fd
>= 0)
2673 dont_close
[n_dont_close
++] = params
->stdout_fd
;
2674 if (params
->stderr_fd
>= 0)
2675 dont_close
[n_dont_close
++] = params
->stderr_fd
;
2678 dont_close
[n_dont_close
++] = socket_fd
;
2680 dont_close
[n_dont_close
++] = exec_fd
;
2682 memcpy(dont_close
+ n_dont_close
, fds
, sizeof(int) * n_fds
);
2683 n_dont_close
+= n_fds
;
2687 append_socket_pair(dont_close
, &n_dont_close
, runtime
->netns_storage_socket
);
2691 append_socket_pair(dont_close
, &n_dont_close
, dcreds
->user
->storage_socket
);
2693 append_socket_pair(dont_close
, &n_dont_close
, dcreds
->group
->storage_socket
);
2696 if (user_lookup_fd
>= 0)
2697 dont_close
[n_dont_close
++] = user_lookup_fd
;
2699 return close_all_fds(dont_close
, n_dont_close
);
2702 static int send_user_lookup(
2710 /* Send the resolved UID/GID to PID 1 after we learnt it. We send a single datagram, containing the UID/GID
2711 * data as well as the unit name. Note that we suppress sending this if no user/group to resolve was
2714 if (user_lookup_fd
< 0)
2717 if (!uid_is_valid(uid
) && !gid_is_valid(gid
))
2720 if (writev(user_lookup_fd
,
2722 IOVEC_INIT(&uid
, sizeof(uid
)),
2723 IOVEC_INIT(&gid
, sizeof(gid
)),
2724 IOVEC_INIT_STRING(unit
->id
) }, 3) < 0)
2730 static int acquire_home(const ExecContext
*c
, uid_t uid
, const char** home
, char **buf
) {
2737 /* If WorkingDirectory=~ is set, try to acquire a usable home directory. */
2742 if (!c
->working_directory_home
)
2745 r
= get_home_dir(buf
);
2753 static int compile_suggested_paths(const ExecContext
*c
, const ExecParameters
*p
, char ***ret
) {
2754 _cleanup_strv_free_
char ** list
= NULL
;
2755 ExecDirectoryType t
;
2762 assert(c
->dynamic_user
);
2764 /* Compile a list of paths that it might make sense to read the owning UID from to use as initial candidate for
2765 * dynamic UID allocation, in order to save us from doing costly recursive chown()s of the special
2768 for (t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
2771 if (t
== EXEC_DIRECTORY_CONFIGURATION
)
2777 STRV_FOREACH(i
, c
->directories
[t
].paths
) {
2780 if (t
== EXEC_DIRECTORY_RUNTIME
)
2781 e
= strjoin(p
->prefix
[t
], "/", *i
);
2783 e
= strjoin(p
->prefix
[t
], "/private/", *i
);
2787 r
= strv_consume(&list
, e
);
2793 *ret
= TAKE_PTR(list
);
2798 static char *exec_command_line(char **argv
);
2800 static int exec_parameters_get_cgroup_path(const ExecParameters
*params
, char **ret
) {
2801 bool using_subcgroup
;
2807 if (!params
->cgroup_path
)
2810 /* If we are called for a unit where cgroup delegation is on, and the payload created its own populated
2811 * subcgroup (which we expect it to do, after all it asked for delegation), then we cannot place the control
2812 * processes started after the main unit's process in the unit's main cgroup because it is now an inner one,
2813 * and inner cgroups may not contain processes. Hence, if delegation is on, and this is a control process,
2814 * let's use ".control" as subcgroup instead. Note that we do so only for ExecStartPost=, ExecReload=,
2815 * ExecStop=, ExecStopPost=, i.e. for the commands where the main process is already forked. For ExecStartPre=
2816 * this is not necessary, the cgroup is still empty. We distinguish these cases with the EXEC_CONTROL_CGROUP
2817 * flag, which is only passed for the former statements, not for the latter. */
2819 using_subcgroup
= FLAGS_SET(params
->flags
, EXEC_CONTROL_CGROUP
|EXEC_CGROUP_DELEGATE
|EXEC_IS_CONTROL
);
2820 if (using_subcgroup
)
2821 p
= strjoin(params
->cgroup_path
, "/.control");
2823 p
= strdup(params
->cgroup_path
);
2828 return using_subcgroup
;
2831 static int exec_child(
2833 const ExecCommand
*command
,
2834 const ExecContext
*context
,
2835 const ExecParameters
*params
,
2836 ExecRuntime
*runtime
,
2837 DynamicCreds
*dcreds
,
2841 size_t n_socket_fds
,
2842 size_t n_storage_fds
,
2847 _cleanup_strv_free_
char **our_env
= NULL
, **pass_env
= NULL
, **accum_env
= NULL
, **replaced_argv
= NULL
;
2848 int *fds_with_exec_fd
, n_fds_with_exec_fd
, r
, ngids
= 0, exec_fd
= -1;
2849 _cleanup_free_ gid_t
*supplementary_gids
= NULL
;
2850 const char *username
= NULL
, *groupname
= NULL
;
2851 _cleanup_free_
char *home_buffer
= NULL
;
2852 const char *home
= NULL
, *shell
= NULL
;
2853 char **final_argv
= NULL
;
2854 dev_t journal_stream_dev
= 0;
2855 ino_t journal_stream_ino
= 0;
2856 bool needs_sandboxing
, /* Do we need to set up full sandboxing? (i.e. all namespacing, all MAC stuff, caps, yadda yadda */
2857 needs_setuid
, /* Do we need to do the actual setresuid()/setresgid() calls? */
2858 needs_mount_namespace
, /* Do we need to set up a mount namespace for this kernel? */
2859 needs_ambient_hack
; /* Do we need to apply the ambient capabilities hack? */
2861 _cleanup_free_
char *mac_selinux_context_net
= NULL
;
2862 bool use_selinux
= false;
2865 bool use_smack
= false;
2868 bool use_apparmor
= false;
2870 uid_t uid
= UID_INVALID
;
2871 gid_t gid
= GID_INVALID
;
2873 ExecDirectoryType dt
;
2880 assert(exit_status
);
2882 rename_process_from_path(command
->path
);
2884 /* We reset exactly these signals, since they are the
2885 * only ones we set to SIG_IGN in the main daemon. All
2886 * others we leave untouched because we set them to
2887 * SIG_DFL or a valid handler initially, both of which
2888 * will be demoted to SIG_DFL. */
2889 (void) default_signals(SIGNALS_CRASH_HANDLER
,
2890 SIGNALS_IGNORE
, -1);
2892 if (context
->ignore_sigpipe
)
2893 (void) ignore_signals(SIGPIPE
, -1);
2895 r
= reset_signal_mask();
2897 *exit_status
= EXIT_SIGNAL_MASK
;
2898 return log_unit_error_errno(unit
, r
, "Failed to set process signal mask: %m");
2901 if (params
->idle_pipe
)
2902 do_idle_pipe_dance(params
->idle_pipe
);
2904 /* Close fds we don't need very early to make sure we don't block init reexecution because it cannot bind its
2905 * sockets. Among the fds we close are the logging fds, and we want to keep them closed, so that we don't have
2906 * any fds open we don't really want open during the transition. In order to make logging work, we switch the
2907 * log subsystem into open_when_needed mode, so that it reopens the logs on every single log call. */
2910 log_set_open_when_needed(true);
2912 /* In case anything used libc syslog(), close this here, too */
2915 n_fds
= n_socket_fds
+ n_storage_fds
;
2916 r
= close_remaining_fds(params
, runtime
, dcreds
, user_lookup_fd
, socket_fd
, params
->exec_fd
, fds
, n_fds
);
2918 *exit_status
= EXIT_FDS
;
2919 return log_unit_error_errno(unit
, r
, "Failed to close unwanted file descriptors: %m");
2922 if (!context
->same_pgrp
)
2924 *exit_status
= EXIT_SETSID
;
2925 return log_unit_error_errno(unit
, errno
, "Failed to create new process session: %m");
2928 exec_context_tty_reset(context
, params
);
2930 if (unit_shall_confirm_spawn(unit
)) {
2931 const char *vc
= params
->confirm_spawn
;
2932 _cleanup_free_
char *cmdline
= NULL
;
2934 cmdline
= exec_command_line(command
->argv
);
2936 *exit_status
= EXIT_MEMORY
;
2940 r
= ask_for_confirmation(vc
, unit
, cmdline
);
2941 if (r
!= CONFIRM_EXECUTE
) {
2942 if (r
== CONFIRM_PRETEND_SUCCESS
) {
2943 *exit_status
= EXIT_SUCCESS
;
2946 *exit_status
= EXIT_CONFIRM
;
2947 log_unit_error(unit
, "Execution cancelled by the user");
2952 /* We are about to invoke NSS and PAM modules. Let's tell them what we are doing here, maybe they care. This is
2953 * used by nss-resolve to disable itself when we are about to start systemd-resolved, to avoid deadlocks. Note
2954 * that these env vars do not survive the execve(), which means they really only apply to the PAM and NSS
2955 * invocations themselves. Also note that while we'll only invoke NSS modules involved in user management they
2956 * might internally call into other NSS modules that are involved in hostname resolution, we never know. */
2957 if (setenv("SYSTEMD_ACTIVATION_UNIT", unit
->id
, true) != 0 ||
2958 setenv("SYSTEMD_ACTIVATION_SCOPE", MANAGER_IS_SYSTEM(unit
->manager
) ? "system" : "user", true) != 0) {
2959 *exit_status
= EXIT_MEMORY
;
2960 return log_unit_error_errno(unit
, errno
, "Failed to update environment: %m");
2963 if (context
->dynamic_user
&& dcreds
) {
2964 _cleanup_strv_free_
char **suggested_paths
= NULL
;
2966 /* On top of that, make sure we bypass our own NSS module nss-systemd comprehensively for any NSS
2967 * checks, if DynamicUser=1 is used, as we shouldn't create a feedback loop with ourselves here.*/
2968 if (putenv((char*) "SYSTEMD_NSS_DYNAMIC_BYPASS=1") != 0) {
2969 *exit_status
= EXIT_USER
;
2970 return log_unit_error_errno(unit
, errno
, "Failed to update environment: %m");
2973 r
= compile_suggested_paths(context
, params
, &suggested_paths
);
2975 *exit_status
= EXIT_MEMORY
;
2979 r
= dynamic_creds_realize(dcreds
, suggested_paths
, &uid
, &gid
);
2981 *exit_status
= EXIT_USER
;
2983 log_unit_error(unit
, "Failed to update dynamic user credentials: User or group with specified name already exists.");
2986 return log_unit_error_errno(unit
, r
, "Failed to update dynamic user credentials: %m");
2989 if (!uid_is_valid(uid
)) {
2990 *exit_status
= EXIT_USER
;
2991 log_unit_error(unit
, "UID validation failed for \""UID_FMT
"\"", uid
);
2995 if (!gid_is_valid(gid
)) {
2996 *exit_status
= EXIT_USER
;
2997 log_unit_error(unit
, "GID validation failed for \""GID_FMT
"\"", gid
);
3002 username
= dcreds
->user
->name
;
3005 r
= get_fixed_user(context
, &username
, &uid
, &gid
, &home
, &shell
);
3007 *exit_status
= EXIT_USER
;
3008 return log_unit_error_errno(unit
, r
, "Failed to determine user credentials: %m");
3011 r
= get_fixed_group(context
, &groupname
, &gid
);
3013 *exit_status
= EXIT_GROUP
;
3014 return log_unit_error_errno(unit
, r
, "Failed to determine group credentials: %m");
3018 /* Initialize user supplementary groups and get SupplementaryGroups= ones */
3019 r
= get_supplementary_groups(context
, username
, groupname
, gid
,
3020 &supplementary_gids
, &ngids
);
3022 *exit_status
= EXIT_GROUP
;
3023 return log_unit_error_errno(unit
, r
, "Failed to determine supplementary groups: %m");
3026 r
= send_user_lookup(unit
, user_lookup_fd
, uid
, gid
);
3028 *exit_status
= EXIT_USER
;
3029 return log_unit_error_errno(unit
, r
, "Failed to send user credentials to PID1: %m");
3032 user_lookup_fd
= safe_close(user_lookup_fd
);
3034 r
= acquire_home(context
, uid
, &home
, &home_buffer
);
3036 *exit_status
= EXIT_CHDIR
;
3037 return log_unit_error_errno(unit
, r
, "Failed to determine $HOME for user: %m");
3040 /* If a socket is connected to STDIN/STDOUT/STDERR, we
3041 * must sure to drop O_NONBLOCK */
3043 (void) fd_nonblock(socket_fd
, false);
3045 /* Journald will try to look-up our cgroup in order to populate _SYSTEMD_CGROUP and _SYSTEMD_UNIT fields.
3046 * Hence we need to migrate to the target cgroup from init.scope before connecting to journald */
3047 if (params
->cgroup_path
) {
3048 _cleanup_free_
char *p
= NULL
;
3050 r
= exec_parameters_get_cgroup_path(params
, &p
);
3052 *exit_status
= EXIT_CGROUP
;
3053 return log_unit_error_errno(unit
, r
, "Failed to acquire cgroup path: %m");
3056 r
= cg_attach_everywhere(params
->cgroup_supported
, p
, 0, NULL
, NULL
);
3058 *exit_status
= EXIT_CGROUP
;
3059 return log_unit_error_errno(unit
, r
, "Failed to attach to cgroup %s: %m", p
);
3063 if (context
->network_namespace_path
&& runtime
&& runtime
->netns_storage_socket
[0] >= 0) {
3064 r
= open_netns_path(runtime
->netns_storage_socket
, context
->network_namespace_path
);
3066 *exit_status
= EXIT_NETWORK
;
3067 return log_unit_error_errno(unit
, r
, "Failed to open network namespace path %s: %m", context
->network_namespace_path
);
3071 r
= setup_input(context
, params
, socket_fd
, named_iofds
);
3073 *exit_status
= EXIT_STDIN
;
3074 return log_unit_error_errno(unit
, r
, "Failed to set up standard input: %m");
3077 r
= setup_output(unit
, context
, params
, STDOUT_FILENO
, socket_fd
, named_iofds
, basename(command
->path
), uid
, gid
, &journal_stream_dev
, &journal_stream_ino
);
3079 *exit_status
= EXIT_STDOUT
;
3080 return log_unit_error_errno(unit
, r
, "Failed to set up standard output: %m");
3083 r
= setup_output(unit
, context
, params
, STDERR_FILENO
, socket_fd
, named_iofds
, basename(command
->path
), uid
, gid
, &journal_stream_dev
, &journal_stream_ino
);
3085 *exit_status
= EXIT_STDERR
;
3086 return log_unit_error_errno(unit
, r
, "Failed to set up standard error output: %m");
3089 if (context
->oom_score_adjust_set
) {
3090 /* When we can't make this change due to EPERM, then let's silently skip over it. User namespaces
3091 * prohibit write access to this file, and we shouldn't trip up over that. */
3092 r
= set_oom_score_adjust(context
->oom_score_adjust
);
3093 if (IN_SET(r
, -EPERM
, -EACCES
))
3094 log_unit_debug_errno(unit
, r
, "Failed to adjust OOM setting, assuming containerized execution, ignoring: %m");
3096 *exit_status
= EXIT_OOM_ADJUST
;
3097 return log_unit_error_errno(unit
, r
, "Failed to adjust OOM setting: %m");
3101 if (context
->nice_set
)
3102 if (setpriority(PRIO_PROCESS
, 0, context
->nice
) < 0) {
3103 *exit_status
= EXIT_NICE
;
3104 return log_unit_error_errno(unit
, errno
, "Failed to set up process scheduling priority (nice level): %m");
3107 if (context
->cpu_sched_set
) {
3108 struct sched_param param
= {
3109 .sched_priority
= context
->cpu_sched_priority
,
3112 r
= sched_setscheduler(0,
3113 context
->cpu_sched_policy
|
3114 (context
->cpu_sched_reset_on_fork
?
3115 SCHED_RESET_ON_FORK
: 0),
3118 *exit_status
= EXIT_SETSCHEDULER
;
3119 return log_unit_error_errno(unit
, errno
, "Failed to set up CPU scheduling: %m");
3123 if (context
->cpuset
)
3124 if (sched_setaffinity(0, CPU_ALLOC_SIZE(context
->cpuset_ncpus
), context
->cpuset
) < 0) {
3125 *exit_status
= EXIT_CPUAFFINITY
;
3126 return log_unit_error_errno(unit
, errno
, "Failed to set up CPU affinity: %m");
3129 if (context
->ioprio_set
)
3130 if (ioprio_set(IOPRIO_WHO_PROCESS
, 0, context
->ioprio
) < 0) {
3131 *exit_status
= EXIT_IOPRIO
;
3132 return log_unit_error_errno(unit
, errno
, "Failed to set up IO scheduling priority: %m");
3135 if (context
->timer_slack_nsec
!= NSEC_INFINITY
)
3136 if (prctl(PR_SET_TIMERSLACK
, context
->timer_slack_nsec
) < 0) {
3137 *exit_status
= EXIT_TIMERSLACK
;
3138 return log_unit_error_errno(unit
, errno
, "Failed to set up timer slack: %m");
3141 if (context
->personality
!= PERSONALITY_INVALID
) {
3142 r
= safe_personality(context
->personality
);
3144 *exit_status
= EXIT_PERSONALITY
;
3145 return log_unit_error_errno(unit
, r
, "Failed to set up execution domain (personality): %m");
3149 if (context
->utmp_id
)
3150 utmp_put_init_process(context
->utmp_id
, getpid_cached(), getsid(0),
3152 context
->utmp_mode
== EXEC_UTMP_INIT
? INIT_PROCESS
:
3153 context
->utmp_mode
== EXEC_UTMP_LOGIN
? LOGIN_PROCESS
:
3157 if (context
->user
) {
3158 r
= chown_terminal(STDIN_FILENO
, uid
);
3160 *exit_status
= EXIT_STDIN
;
3161 return log_unit_error_errno(unit
, r
, "Failed to change ownership of terminal: %m");
3165 /* If delegation is enabled we'll pass ownership of the cgroup to the user of the new process. On cgroup v1
3166 * this is only about systemd's own hierarchy, i.e. not the controller hierarchies, simply because that's not
3167 * safe. On cgroup v2 there's only one hierarchy anyway, and delegation is safe there, hence in that case only
3168 * touch a single hierarchy too. */
3169 if (params
->cgroup_path
&& context
->user
&& (params
->flags
& EXEC_CGROUP_DELEGATE
)) {
3170 r
= cg_set_access(SYSTEMD_CGROUP_CONTROLLER
, params
->cgroup_path
, uid
, gid
);
3172 *exit_status
= EXIT_CGROUP
;
3173 return log_unit_error_errno(unit
, r
, "Failed to adjust control group access: %m");
3177 for (dt
= 0; dt
< _EXEC_DIRECTORY_TYPE_MAX
; dt
++) {
3178 r
= setup_exec_directory(context
, params
, uid
, gid
, dt
, exit_status
);
3180 return log_unit_error_errno(unit
, r
, "Failed to set up special execution directory in %s: %m", params
->prefix
[dt
]);
3183 r
= build_environment(
3195 *exit_status
= EXIT_MEMORY
;
3199 r
= build_pass_environment(context
, &pass_env
);
3201 *exit_status
= EXIT_MEMORY
;
3205 accum_env
= strv_env_merge(5,
3206 params
->environment
,
3209 context
->environment
,
3213 *exit_status
= EXIT_MEMORY
;
3216 accum_env
= strv_env_clean(accum_env
);
3218 (void) umask(context
->umask
);
3220 r
= setup_keyring(unit
, context
, params
, uid
, gid
);
3222 *exit_status
= EXIT_KEYRING
;
3223 return log_unit_error_errno(unit
, r
, "Failed to set up kernel keyring: %m");
3226 /* We need sandboxing if the caller asked us to apply it and the command isn't explicitly excepted from it */
3227 needs_sandboxing
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && !(command
->flags
& EXEC_COMMAND_FULLY_PRIVILEGED
);
3229 /* 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 */
3230 needs_ambient_hack
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && (command
->flags
& EXEC_COMMAND_AMBIENT_MAGIC
) && !ambient_capabilities_supported();
3232 /* 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 */
3233 if (needs_ambient_hack
)
3234 needs_setuid
= false;
3236 needs_setuid
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && !(command
->flags
& (EXEC_COMMAND_FULLY_PRIVILEGED
|EXEC_COMMAND_NO_SETUID
));
3238 if (needs_sandboxing
) {
3239 /* MAC enablement checks need to be done before a new mount ns is created, as they rely on /sys being
3240 * present. The actual MAC context application will happen later, as late as possible, to avoid
3241 * impacting our own code paths. */
3244 use_selinux
= mac_selinux_use();
3247 use_smack
= mac_smack_use();
3250 use_apparmor
= mac_apparmor_use();
3254 if (needs_sandboxing
) {
3257 /* Let's set the resource limits before we call into PAM, so that pam_limits wins over what
3258 * is set here. (See below.) */
3260 r
= setrlimit_closest_all((const struct rlimit
* const *) context
->rlimit
, &which_failed
);
3262 *exit_status
= EXIT_LIMITS
;
3263 return log_unit_error_errno(unit
, r
, "Failed to adjust resource limit RLIMIT_%s: %m", rlimit_to_string(which_failed
));
3269 /* Let's call into PAM after we set up our own idea of resource limits to that pam_limits
3270 * wins here. (See above.) */
3272 if (context
->pam_name
&& username
) {
3273 r
= setup_pam(context
->pam_name
, username
, uid
, gid
, context
->tty_path
, &accum_env
, fds
, n_fds
);
3275 *exit_status
= EXIT_PAM
;
3276 return log_unit_error_errno(unit
, r
, "Failed to set up PAM session: %m");
3281 if ((context
->private_network
|| context
->network_namespace_path
) && runtime
&& runtime
->netns_storage_socket
[0] >= 0) {
3283 if (ns_type_supported(NAMESPACE_NET
)) {
3284 r
= setup_netns(runtime
->netns_storage_socket
);
3286 *exit_status
= EXIT_NETWORK
;
3287 return log_unit_error_errno(unit
, r
, "Failed to set up network namespacing: %m");
3289 } else if (context
->network_namespace_path
) {
3290 *exit_status
= EXIT_NETWORK
;
3291 return log_unit_error_errno(unit
, SYNTHETIC_ERRNO(EOPNOTSUPP
), "NetworkNamespacePath= is not supported, refusing.");
3293 log_unit_warning(unit
, "PrivateNetwork=yes is configured, but the kernel does not support network namespaces, ignoring.");
3296 needs_mount_namespace
= exec_needs_mount_namespace(context
, params
, runtime
);
3297 if (needs_mount_namespace
) {
3298 r
= apply_mount_namespace(unit
, command
, context
, params
, runtime
);
3300 *exit_status
= EXIT_NAMESPACE
;
3301 return log_unit_error_errno(unit
, r
, "Failed to set up mount namespacing: %m");
3305 if (context
->protect_hostname
) {
3306 if (ns_type_supported(NAMESPACE_UTS
)) {
3307 if (unshare(CLONE_NEWUTS
) < 0) {
3308 *exit_status
= EXIT_NAMESPACE
;
3309 return log_unit_error_errno(unit
, errno
, "Failed to set up UTS namespacing: %m");
3312 log_unit_warning(unit
, "ProtectHostname=yes is configured, but the kernel does not support UTS namespaces, ignoring namespace setup.");
3314 r
= seccomp_protect_hostname();
3316 *exit_status
= EXIT_SECCOMP
;
3317 return log_unit_error_errno(unit
, r
, "Failed to apply hostname restrictions: %m");
3322 /* Drop groups as early as possbile */
3324 r
= enforce_groups(gid
, supplementary_gids
, ngids
);
3326 *exit_status
= EXIT_GROUP
;
3327 return log_unit_error_errno(unit
, r
, "Changing group credentials failed: %m");
3331 if (needs_sandboxing
) {
3333 if (use_selinux
&& params
->selinux_context_net
&& socket_fd
>= 0) {
3334 r
= mac_selinux_get_child_mls_label(socket_fd
, command
->path
, context
->selinux_context
, &mac_selinux_context_net
);
3336 *exit_status
= EXIT_SELINUX_CONTEXT
;
3337 return log_unit_error_errno(unit
, r
, "Failed to determine SELinux context: %m");
3342 if (context
->private_users
) {
3343 r
= setup_private_users(uid
, gid
);
3345 *exit_status
= EXIT_USER
;
3346 return log_unit_error_errno(unit
, r
, "Failed to set up user namespacing: %m");
3351 /* We repeat the fd closing here, to make sure that nothing is leaked from the PAM modules. Note that we are
3352 * more aggressive this time since socket_fd and the netns fds we don't need anymore. We do keep the exec_fd
3353 * however if we have it as we want to keep it open until the final execve(). */
3355 if (params
->exec_fd
>= 0) {
3356 exec_fd
= params
->exec_fd
;
3358 if (exec_fd
< 3 + (int) n_fds
) {
3361 /* Let's move the exec fd far up, so that it's outside of the fd range we want to pass to the
3362 * process we are about to execute. */
3364 moved_fd
= fcntl(exec_fd
, F_DUPFD_CLOEXEC
, 3 + (int) n_fds
);
3366 *exit_status
= EXIT_FDS
;
3367 return log_unit_error_errno(unit
, errno
, "Couldn't move exec fd up: %m");
3370 safe_close(exec_fd
);
3373 /* This fd should be FD_CLOEXEC already, but let's make sure. */
3374 r
= fd_cloexec(exec_fd
, true);
3376 *exit_status
= EXIT_FDS
;
3377 return log_unit_error_errno(unit
, r
, "Failed to make exec fd FD_CLOEXEC: %m");
3381 fds_with_exec_fd
= newa(int, n_fds
+ 1);
3382 memcpy_safe(fds_with_exec_fd
, fds
, n_fds
* sizeof(int));
3383 fds_with_exec_fd
[n_fds
] = exec_fd
;
3384 n_fds_with_exec_fd
= n_fds
+ 1;
3386 fds_with_exec_fd
= fds
;
3387 n_fds_with_exec_fd
= n_fds
;
3390 r
= close_all_fds(fds_with_exec_fd
, n_fds_with_exec_fd
);
3392 r
= shift_fds(fds
, n_fds
);
3394 r
= flags_fds(fds
, n_socket_fds
, n_storage_fds
, context
->non_blocking
);
3396 *exit_status
= EXIT_FDS
;
3397 return log_unit_error_errno(unit
, r
, "Failed to adjust passed file descriptors: %m");
3400 /* At this point, the fds we want to pass to the program are all ready and set up, with O_CLOEXEC turned off
3401 * and at the right fd numbers. The are no other fds open, with one exception: the exec_fd if it is defined,
3402 * and it has O_CLOEXEC set, after all we want it to be closed by the execve(), so that our parent knows we
3405 secure_bits
= context
->secure_bits
;
3407 if (needs_sandboxing
) {
3410 /* Set the RTPRIO resource limit to 0, but only if nothing else was explicitly
3411 * requested. (Note this is placed after the general resource limit initialization, see
3412 * above, in order to take precedence.) */
3413 if (context
->restrict_realtime
&& !context
->rlimit
[RLIMIT_RTPRIO
]) {
3414 if (setrlimit(RLIMIT_RTPRIO
, &RLIMIT_MAKE_CONST(0)) < 0) {
3415 *exit_status
= EXIT_LIMITS
;
3416 return log_unit_error_errno(unit
, errno
, "Failed to adjust RLIMIT_RTPRIO resource limit: %m");
3421 /* LSM Smack needs the capability CAP_MAC_ADMIN to change the current execution security context of the
3422 * process. This is the latest place before dropping capabilities. Other MAC context are set later. */
3424 r
= setup_smack(context
, command
);
3426 *exit_status
= EXIT_SMACK_PROCESS_LABEL
;
3427 return log_unit_error_errno(unit
, r
, "Failed to set SMACK process label: %m");
3432 bset
= context
->capability_bounding_set
;
3433 /* If the ambient caps hack is enabled (which means the kernel can't do them, and the user asked for
3434 * our magic fallback), then let's add some extra caps, so that the service can drop privs of its own,
3435 * instead of us doing that */
3436 if (needs_ambient_hack
)
3437 bset
|= (UINT64_C(1) << CAP_SETPCAP
) |
3438 (UINT64_C(1) << CAP_SETUID
) |
3439 (UINT64_C(1) << CAP_SETGID
);
3441 if (!cap_test_all(bset
)) {
3442 r
= capability_bounding_set_drop(bset
, false);
3444 *exit_status
= EXIT_CAPABILITIES
;
3445 return log_unit_error_errno(unit
, r
, "Failed to drop capabilities: %m");
3449 /* This is done before enforce_user, but ambient set
3450 * does not survive over setresuid() if keep_caps is not set. */
3451 if (!needs_ambient_hack
&&
3452 context
->capability_ambient_set
!= 0) {
3453 r
= capability_ambient_set_apply(context
->capability_ambient_set
, true);
3455 *exit_status
= EXIT_CAPABILITIES
;
3456 return log_unit_error_errno(unit
, r
, "Failed to apply ambient capabilities (before UID change): %m");
3462 if (context
->user
) {
3463 r
= enforce_user(context
, uid
);
3465 *exit_status
= EXIT_USER
;
3466 return log_unit_error_errno(unit
, r
, "Failed to change UID to " UID_FMT
": %m", uid
);
3469 if (!needs_ambient_hack
&&
3470 context
->capability_ambient_set
!= 0) {
3472 /* Fix the ambient capabilities after user change. */
3473 r
= capability_ambient_set_apply(context
->capability_ambient_set
, false);
3475 *exit_status
= EXIT_CAPABILITIES
;
3476 return log_unit_error_errno(unit
, r
, "Failed to apply ambient capabilities (after UID change): %m");
3479 /* If we were asked to change user and ambient capabilities
3480 * were requested, we had to add keep-caps to the securebits
3481 * so that we would maintain the inherited capability set
3482 * through the setresuid(). Make sure that the bit is added
3483 * also to the context secure_bits so that we don't try to
3484 * drop the bit away next. */
3486 secure_bits
|= 1<<SECURE_KEEP_CAPS
;
3491 /* Apply working directory here, because the working directory might be on NFS and only the user running
3492 * this service might have the correct privilege to change to the working directory */
3493 r
= apply_working_directory(context
, params
, home
, needs_mount_namespace
, exit_status
);
3495 return log_unit_error_errno(unit
, r
, "Changing to the requested working directory failed: %m");
3497 if (needs_sandboxing
) {
3498 /* Apply other MAC contexts late, but before seccomp syscall filtering, as those should really be last to
3499 * influence our own codepaths as little as possible. Moreover, applying MAC contexts usually requires
3500 * syscalls that are subject to seccomp filtering, hence should probably be applied before the syscalls
3501 * are restricted. */
3505 char *exec_context
= mac_selinux_context_net
?: context
->selinux_context
;
3508 r
= setexeccon(exec_context
);
3510 *exit_status
= EXIT_SELINUX_CONTEXT
;
3511 return log_unit_error_errno(unit
, r
, "Failed to change SELinux context to %s: %m", exec_context
);
3518 if (use_apparmor
&& context
->apparmor_profile
) {
3519 r
= aa_change_onexec(context
->apparmor_profile
);
3520 if (r
< 0 && !context
->apparmor_profile_ignore
) {
3521 *exit_status
= EXIT_APPARMOR_PROFILE
;
3522 return log_unit_error_errno(unit
, errno
, "Failed to prepare AppArmor profile change to %s: %m", context
->apparmor_profile
);
3527 /* PR_GET_SECUREBITS is not privileged, while PR_SET_SECUREBITS is. So to suppress potential EPERMs
3528 * we'll try not to call PR_SET_SECUREBITS unless necessary. */
3529 if (prctl(PR_GET_SECUREBITS
) != secure_bits
)
3530 if (prctl(PR_SET_SECUREBITS
, secure_bits
) < 0) {
3531 *exit_status
= EXIT_SECUREBITS
;
3532 return log_unit_error_errno(unit
, errno
, "Failed to set process secure bits: %m");
3535 if (context_has_no_new_privileges(context
))
3536 if (prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0) < 0) {
3537 *exit_status
= EXIT_NO_NEW_PRIVILEGES
;
3538 return log_unit_error_errno(unit
, errno
, "Failed to disable new privileges: %m");
3542 r
= apply_address_families(unit
, context
);
3544 *exit_status
= EXIT_ADDRESS_FAMILIES
;
3545 return log_unit_error_errno(unit
, r
, "Failed to restrict address families: %m");
3548 r
= apply_memory_deny_write_execute(unit
, context
);
3550 *exit_status
= EXIT_SECCOMP
;
3551 return log_unit_error_errno(unit
, r
, "Failed to disable writing to executable memory: %m");
3554 r
= apply_restrict_realtime(unit
, context
);
3556 *exit_status
= EXIT_SECCOMP
;
3557 return log_unit_error_errno(unit
, r
, "Failed to apply realtime restrictions: %m");
3560 r
= apply_restrict_namespaces(unit
, context
);
3562 *exit_status
= EXIT_SECCOMP
;
3563 return log_unit_error_errno(unit
, r
, "Failed to apply namespace restrictions: %m");
3566 r
= apply_protect_sysctl(unit
, context
);
3568 *exit_status
= EXIT_SECCOMP
;
3569 return log_unit_error_errno(unit
, r
, "Failed to apply sysctl restrictions: %m");
3572 r
= apply_protect_kernel_modules(unit
, context
);
3574 *exit_status
= EXIT_SECCOMP
;
3575 return log_unit_error_errno(unit
, r
, "Failed to apply module loading restrictions: %m");
3578 r
= apply_private_devices(unit
, context
);
3580 *exit_status
= EXIT_SECCOMP
;
3581 return log_unit_error_errno(unit
, r
, "Failed to set up private devices: %m");
3584 r
= apply_syscall_archs(unit
, context
);
3586 *exit_status
= EXIT_SECCOMP
;
3587 return log_unit_error_errno(unit
, r
, "Failed to apply syscall architecture restrictions: %m");
3590 r
= apply_lock_personality(unit
, context
);
3592 *exit_status
= EXIT_SECCOMP
;
3593 return log_unit_error_errno(unit
, r
, "Failed to lock personalities: %m");
3596 /* This really should remain the last step before the execve(), to make sure our own code is unaffected
3597 * by the filter as little as possible. */
3598 r
= apply_syscall_filter(unit
, context
, needs_ambient_hack
);
3600 *exit_status
= EXIT_SECCOMP
;
3601 return log_unit_error_errno(unit
, r
, "Failed to apply system call filters: %m");
3606 if (!strv_isempty(context
->unset_environment
)) {
3609 ee
= strv_env_delete(accum_env
, 1, context
->unset_environment
);
3611 *exit_status
= EXIT_MEMORY
;
3615 strv_free_and_replace(accum_env
, ee
);
3618 if (!FLAGS_SET(command
->flags
, EXEC_COMMAND_NO_ENV_EXPAND
)) {
3619 replaced_argv
= replace_env_argv(command
->argv
, accum_env
);
3620 if (!replaced_argv
) {
3621 *exit_status
= EXIT_MEMORY
;
3624 final_argv
= replaced_argv
;
3626 final_argv
= command
->argv
;
3628 if (DEBUG_LOGGING
) {
3629 _cleanup_free_
char *line
;
3631 line
= exec_command_line(final_argv
);
3633 log_struct(LOG_DEBUG
,
3634 "EXECUTABLE=%s", command
->path
,
3635 LOG_UNIT_MESSAGE(unit
, "Executing: %s", line
),
3637 LOG_UNIT_INVOCATION_ID(unit
));
3643 /* We have finished with all our initializations. Let's now let the manager know that. From this point
3644 * on, if the manager sees POLLHUP on the exec_fd, then execve() was successful. */
3646 if (write(exec_fd
, &hot
, sizeof(hot
)) < 0) {
3647 *exit_status
= EXIT_EXEC
;
3648 return log_unit_error_errno(unit
, errno
, "Failed to enable exec_fd: %m");
3652 execve(command
->path
, final_argv
, accum_env
);
3658 /* The execve() failed. This means the exec_fd is still open. Which means we need to tell the manager
3659 * that POLLHUP on it no longer means execve() succeeded. */
3661 if (write(exec_fd
, &hot
, sizeof(hot
)) < 0) {
3662 *exit_status
= EXIT_EXEC
;
3663 return log_unit_error_errno(unit
, errno
, "Failed to disable exec_fd: %m");
3667 if (r
== -ENOENT
&& (command
->flags
& EXEC_COMMAND_IGNORE_FAILURE
)) {
3668 log_struct_errno(LOG_INFO
, r
,
3669 "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR
,
3671 LOG_UNIT_INVOCATION_ID(unit
),
3672 LOG_UNIT_MESSAGE(unit
, "Executable %s missing, skipping: %m",
3674 "EXECUTABLE=%s", command
->path
);
3678 *exit_status
= EXIT_EXEC
;
3679 return log_unit_error_errno(unit
, r
, "Failed to execute command: %m");
3682 static int exec_context_load_environment(const Unit
*unit
, const ExecContext
*c
, char ***l
);
3683 static int exec_context_named_iofds(const ExecContext
*c
, const ExecParameters
*p
, int named_iofds
[3]);
3685 int exec_spawn(Unit
*unit
,
3686 ExecCommand
*command
,
3687 const ExecContext
*context
,
3688 const ExecParameters
*params
,
3689 ExecRuntime
*runtime
,
3690 DynamicCreds
*dcreds
,
3693 int socket_fd
, r
, named_iofds
[3] = { -1, -1, -1 }, *fds
= NULL
;
3694 _cleanup_free_
char *subcgroup_path
= NULL
;
3695 _cleanup_strv_free_
char **files_env
= NULL
;
3696 size_t n_storage_fds
= 0, n_socket_fds
= 0;
3697 _cleanup_free_
char *line
= NULL
;
3705 assert(params
->fds
|| (params
->n_socket_fds
+ params
->n_storage_fds
<= 0));
3707 if (context
->std_input
== EXEC_INPUT_SOCKET
||
3708 context
->std_output
== EXEC_OUTPUT_SOCKET
||
3709 context
->std_error
== EXEC_OUTPUT_SOCKET
) {
3711 if (params
->n_socket_fds
> 1) {
3712 log_unit_error(unit
, "Got more than one socket.");
3716 if (params
->n_socket_fds
== 0) {
3717 log_unit_error(unit
, "Got no socket.");
3721 socket_fd
= params
->fds
[0];
3725 n_socket_fds
= params
->n_socket_fds
;
3726 n_storage_fds
= params
->n_storage_fds
;
3729 r
= exec_context_named_iofds(context
, params
, named_iofds
);
3731 return log_unit_error_errno(unit
, r
, "Failed to load a named file descriptor: %m");
3733 r
= exec_context_load_environment(unit
, context
, &files_env
);
3735 return log_unit_error_errno(unit
, r
, "Failed to load environment files: %m");
3737 line
= exec_command_line(command
->argv
);
3741 log_struct(LOG_DEBUG
,
3742 LOG_UNIT_MESSAGE(unit
, "About to execute: %s", line
),
3743 "EXECUTABLE=%s", command
->path
,
3745 LOG_UNIT_INVOCATION_ID(unit
));
3747 if (params
->cgroup_path
) {
3748 r
= exec_parameters_get_cgroup_path(params
, &subcgroup_path
);
3750 return log_unit_error_errno(unit
, r
, "Failed to acquire subcgroup path: %m");
3751 if (r
> 0) { /* We are using a child cgroup */
3752 r
= cg_create(SYSTEMD_CGROUP_CONTROLLER
, subcgroup_path
);
3754 return log_unit_error_errno(unit
, r
, "Failed to create control group '%s': %m", subcgroup_path
);
3760 return log_unit_error_errno(unit
, errno
, "Failed to fork: %m");
3763 int exit_status
= EXIT_SUCCESS
;
3765 r
= exec_child(unit
,
3777 unit
->manager
->user_lookup_fds
[1],
3781 log_struct_errno(LOG_ERR
, r
,
3782 "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR
,
3784 LOG_UNIT_INVOCATION_ID(unit
),
3785 LOG_UNIT_MESSAGE(unit
, "Failed at step %s spawning %s: %m",
3786 exit_status_to_string(exit_status
, EXIT_STATUS_SYSTEMD
),
3788 "EXECUTABLE=%s", command
->path
);
3793 log_unit_debug(unit
, "Forked %s as "PID_FMT
, command
->path
, pid
);
3795 /* We add the new process to the cgroup both in the child (so that we can be sure that no user code is ever
3796 * executed outside of the cgroup) and in the parent (so that we can be sure that when we kill the cgroup the
3797 * process will be killed too). */
3799 (void) cg_attach(SYSTEMD_CGROUP_CONTROLLER
, subcgroup_path
, pid
);
3801 exec_status_start(&command
->exec_status
, pid
);
3807 void exec_context_init(ExecContext
*c
) {
3808 ExecDirectoryType i
;
3813 c
->ioprio
= IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE
, 0);
3814 c
->cpu_sched_policy
= SCHED_OTHER
;
3815 c
->syslog_priority
= LOG_DAEMON
|LOG_INFO
;
3816 c
->syslog_level_prefix
= true;
3817 c
->ignore_sigpipe
= true;
3818 c
->timer_slack_nsec
= NSEC_INFINITY
;
3819 c
->personality
= PERSONALITY_INVALID
;
3820 for (i
= 0; i
< _EXEC_DIRECTORY_TYPE_MAX
; i
++)
3821 c
->directories
[i
].mode
= 0755;
3822 c
->capability_bounding_set
= CAP_ALL
;
3823 assert_cc(NAMESPACE_FLAGS_INITIAL
!= NAMESPACE_FLAGS_ALL
);
3824 c
->restrict_namespaces
= NAMESPACE_FLAGS_INITIAL
;
3825 c
->log_level_max
= -1;
3828 void exec_context_done(ExecContext
*c
) {
3829 ExecDirectoryType i
;
3834 c
->environment
= strv_free(c
->environment
);
3835 c
->environment_files
= strv_free(c
->environment_files
);
3836 c
->pass_environment
= strv_free(c
->pass_environment
);
3837 c
->unset_environment
= strv_free(c
->unset_environment
);
3839 rlimit_free_all(c
->rlimit
);
3841 for (l
= 0; l
< 3; l
++) {
3842 c
->stdio_fdname
[l
] = mfree(c
->stdio_fdname
[l
]);
3843 c
->stdio_file
[l
] = mfree(c
->stdio_file
[l
]);
3846 c
->working_directory
= mfree(c
->working_directory
);
3847 c
->root_directory
= mfree(c
->root_directory
);
3848 c
->root_image
= mfree(c
->root_image
);
3849 c
->tty_path
= mfree(c
->tty_path
);
3850 c
->syslog_identifier
= mfree(c
->syslog_identifier
);
3851 c
->user
= mfree(c
->user
);
3852 c
->group
= mfree(c
->group
);
3854 c
->supplementary_groups
= strv_free(c
->supplementary_groups
);
3856 c
->pam_name
= mfree(c
->pam_name
);
3858 c
->read_only_paths
= strv_free(c
->read_only_paths
);
3859 c
->read_write_paths
= strv_free(c
->read_write_paths
);
3860 c
->inaccessible_paths
= strv_free(c
->inaccessible_paths
);
3862 bind_mount_free_many(c
->bind_mounts
, c
->n_bind_mounts
);
3863 c
->bind_mounts
= NULL
;
3864 c
->n_bind_mounts
= 0;
3865 temporary_filesystem_free_many(c
->temporary_filesystems
, c
->n_temporary_filesystems
);
3866 c
->temporary_filesystems
= NULL
;
3867 c
->n_temporary_filesystems
= 0;
3869 c
->cpuset
= cpu_set_mfree(c
->cpuset
);
3871 c
->utmp_id
= mfree(c
->utmp_id
);
3872 c
->selinux_context
= mfree(c
->selinux_context
);
3873 c
->apparmor_profile
= mfree(c
->apparmor_profile
);
3874 c
->smack_process_label
= mfree(c
->smack_process_label
);
3876 c
->syscall_filter
= hashmap_free(c
->syscall_filter
);
3877 c
->syscall_archs
= set_free(c
->syscall_archs
);
3878 c
->address_families
= set_free(c
->address_families
);
3880 for (i
= 0; i
< _EXEC_DIRECTORY_TYPE_MAX
; i
++)
3881 c
->directories
[i
].paths
= strv_free(c
->directories
[i
].paths
);
3883 c
->log_level_max
= -1;
3885 exec_context_free_log_extra_fields(c
);
3887 c
->log_rate_limit_interval_usec
= 0;
3888 c
->log_rate_limit_burst
= 0;
3890 c
->stdin_data
= mfree(c
->stdin_data
);
3891 c
->stdin_data_size
= 0;
3893 c
->network_namespace_path
= mfree(c
->network_namespace_path
);
3896 int exec_context_destroy_runtime_directory(const ExecContext
*c
, const char *runtime_prefix
) {
3901 if (!runtime_prefix
)
3904 STRV_FOREACH(i
, c
->directories
[EXEC_DIRECTORY_RUNTIME
].paths
) {
3905 _cleanup_free_
char *p
;
3907 p
= path_join(runtime_prefix
, *i
);
3911 /* We execute this synchronously, since we need to be sure this is gone when we start the
3913 (void) rm_rf(p
, REMOVE_ROOT
);
3919 static void exec_command_done(ExecCommand
*c
) {
3922 c
->path
= mfree(c
->path
);
3923 c
->argv
= strv_free(c
->argv
);
3926 void exec_command_done_array(ExecCommand
*c
, size_t n
) {
3929 for (i
= 0; i
< n
; i
++)
3930 exec_command_done(c
+i
);
3933 ExecCommand
* exec_command_free_list(ExecCommand
*c
) {
3937 LIST_REMOVE(command
, c
, i
);
3938 exec_command_done(i
);
3945 void exec_command_free_array(ExecCommand
**c
, size_t n
) {
3948 for (i
= 0; i
< n
; i
++)
3949 c
[i
] = exec_command_free_list(c
[i
]);
3952 void exec_command_reset_status_array(ExecCommand
*c
, size_t n
) {
3955 for (i
= 0; i
< n
; i
++)
3956 exec_status_reset(&c
[i
].exec_status
);
3959 void exec_command_reset_status_list_array(ExecCommand
**c
, size_t n
) {
3962 for (i
= 0; i
< n
; i
++) {
3965 LIST_FOREACH(command
, z
, c
[i
])
3966 exec_status_reset(&z
->exec_status
);
3970 typedef struct InvalidEnvInfo
{
3975 static void invalid_env(const char *p
, void *userdata
) {
3976 InvalidEnvInfo
*info
= userdata
;
3978 log_unit_error(info
->unit
, "Ignoring invalid environment assignment '%s': %s", p
, info
->path
);
3981 const char* exec_context_fdname(const ExecContext
*c
, int fd_index
) {
3987 if (c
->std_input
!= EXEC_INPUT_NAMED_FD
)
3990 return c
->stdio_fdname
[STDIN_FILENO
] ?: "stdin";
3993 if (c
->std_output
!= EXEC_OUTPUT_NAMED_FD
)
3996 return c
->stdio_fdname
[STDOUT_FILENO
] ?: "stdout";
3999 if (c
->std_error
!= EXEC_OUTPUT_NAMED_FD
)
4002 return c
->stdio_fdname
[STDERR_FILENO
] ?: "stderr";
4009 static int exec_context_named_iofds(const ExecContext
*c
, const ExecParameters
*p
, int named_iofds
[static 3]) {
4011 const char* stdio_fdname
[3];
4017 targets
= (c
->std_input
== EXEC_INPUT_NAMED_FD
) +
4018 (c
->std_output
== EXEC_OUTPUT_NAMED_FD
) +
4019 (c
->std_error
== EXEC_OUTPUT_NAMED_FD
);
4021 for (i
= 0; i
< 3; i
++)
4022 stdio_fdname
[i
] = exec_context_fdname(c
, i
);
4024 n_fds
= p
->n_storage_fds
+ p
->n_socket_fds
;
4026 for (i
= 0; i
< n_fds
&& targets
> 0; i
++)
4027 if (named_iofds
[STDIN_FILENO
] < 0 &&
4028 c
->std_input
== EXEC_INPUT_NAMED_FD
&&
4029 stdio_fdname
[STDIN_FILENO
] &&
4030 streq(p
->fd_names
[i
], stdio_fdname
[STDIN_FILENO
])) {
4032 named_iofds
[STDIN_FILENO
] = p
->fds
[i
];
4035 } else if (named_iofds
[STDOUT_FILENO
] < 0 &&
4036 c
->std_output
== EXEC_OUTPUT_NAMED_FD
&&
4037 stdio_fdname
[STDOUT_FILENO
] &&
4038 streq(p
->fd_names
[i
], stdio_fdname
[STDOUT_FILENO
])) {
4040 named_iofds
[STDOUT_FILENO
] = p
->fds
[i
];
4043 } else if (named_iofds
[STDERR_FILENO
] < 0 &&
4044 c
->std_error
== EXEC_OUTPUT_NAMED_FD
&&
4045 stdio_fdname
[STDERR_FILENO
] &&
4046 streq(p
->fd_names
[i
], stdio_fdname
[STDERR_FILENO
])) {
4048 named_iofds
[STDERR_FILENO
] = p
->fds
[i
];
4052 return targets
== 0 ? 0 : -ENOENT
;
4055 static int exec_context_load_environment(const Unit
*unit
, const ExecContext
*c
, char ***l
) {
4056 char **i
, **r
= NULL
;
4061 STRV_FOREACH(i
, c
->environment_files
) {
4065 bool ignore
= false;
4067 _cleanup_globfree_ glob_t pglob
= {};
4076 if (!path_is_absolute(fn
)) {
4084 /* Filename supports globbing, take all matching files */
4085 k
= safe_glob(fn
, 0, &pglob
);
4094 /* When we don't match anything, -ENOENT should be returned */
4095 assert(pglob
.gl_pathc
> 0);
4097 for (n
= 0; n
< pglob
.gl_pathc
; n
++) {
4098 k
= load_env_file(NULL
, pglob
.gl_pathv
[n
], &p
);
4106 /* Log invalid environment variables with filename */
4108 InvalidEnvInfo info
= {
4110 .path
= pglob
.gl_pathv
[n
]
4113 p
= strv_env_clean_with_callback(p
, invalid_env
, &info
);
4121 m
= strv_env_merge(2, r
, p
);
4137 static bool tty_may_match_dev_console(const char *tty
) {
4138 _cleanup_free_
char *resolved
= NULL
;
4143 tty
= skip_dev_prefix(tty
);
4145 /* trivial identity? */
4146 if (streq(tty
, "console"))
4149 if (resolve_dev_console(&resolved
) < 0)
4150 return true; /* if we could not resolve, assume it may */
4152 /* "tty0" means the active VC, so it may be the same sometimes */
4153 return streq(resolved
, tty
) || (streq(resolved
, "tty0") && tty_is_vc(tty
));
4156 bool exec_context_may_touch_console(const ExecContext
*ec
) {
4158 return (ec
->tty_reset
||
4160 ec
->tty_vt_disallocate
||
4161 is_terminal_input(ec
->std_input
) ||
4162 is_terminal_output(ec
->std_output
) ||
4163 is_terminal_output(ec
->std_error
)) &&
4164 tty_may_match_dev_console(exec_context_tty_path(ec
));
4167 static void strv_fprintf(FILE *f
, char **l
) {
4173 fprintf(f
, " %s", *g
);
4176 void exec_context_dump(const ExecContext
*c
, FILE* f
, const char *prefix
) {
4177 ExecDirectoryType dt
;
4185 prefix
= strempty(prefix
);
4189 "%sWorkingDirectory: %s\n"
4190 "%sRootDirectory: %s\n"
4191 "%sNonBlocking: %s\n"
4192 "%sPrivateTmp: %s\n"
4193 "%sPrivateDevices: %s\n"
4194 "%sProtectKernelTunables: %s\n"
4195 "%sProtectKernelModules: %s\n"
4196 "%sProtectControlGroups: %s\n"
4197 "%sPrivateNetwork: %s\n"
4198 "%sPrivateUsers: %s\n"
4199 "%sProtectHome: %s\n"
4200 "%sProtectSystem: %s\n"
4201 "%sMountAPIVFS: %s\n"
4202 "%sIgnoreSIGPIPE: %s\n"
4203 "%sMemoryDenyWriteExecute: %s\n"
4204 "%sRestrictRealtime: %s\n"
4205 "%sKeyringMode: %s\n"
4206 "%sProtectHostname: %s\n",
4208 prefix
, c
->working_directory
? c
->working_directory
: "/",
4209 prefix
, c
->root_directory
? c
->root_directory
: "/",
4210 prefix
, yes_no(c
->non_blocking
),
4211 prefix
, yes_no(c
->private_tmp
),
4212 prefix
, yes_no(c
->private_devices
),
4213 prefix
, yes_no(c
->protect_kernel_tunables
),
4214 prefix
, yes_no(c
->protect_kernel_modules
),
4215 prefix
, yes_no(c
->protect_control_groups
),
4216 prefix
, yes_no(c
->private_network
),
4217 prefix
, yes_no(c
->private_users
),
4218 prefix
, protect_home_to_string(c
->protect_home
),
4219 prefix
, protect_system_to_string(c
->protect_system
),
4220 prefix
, yes_no(c
->mount_apivfs
),
4221 prefix
, yes_no(c
->ignore_sigpipe
),
4222 prefix
, yes_no(c
->memory_deny_write_execute
),
4223 prefix
, yes_no(c
->restrict_realtime
),
4224 prefix
, exec_keyring_mode_to_string(c
->keyring_mode
),
4225 prefix
, yes_no(c
->protect_hostname
));
4228 fprintf(f
, "%sRootImage: %s\n", prefix
, c
->root_image
);
4230 STRV_FOREACH(e
, c
->environment
)
4231 fprintf(f
, "%sEnvironment: %s\n", prefix
, *e
);
4233 STRV_FOREACH(e
, c
->environment_files
)
4234 fprintf(f
, "%sEnvironmentFile: %s\n", prefix
, *e
);
4236 STRV_FOREACH(e
, c
->pass_environment
)
4237 fprintf(f
, "%sPassEnvironment: %s\n", prefix
, *e
);
4239 STRV_FOREACH(e
, c
->unset_environment
)
4240 fprintf(f
, "%sUnsetEnvironment: %s\n", prefix
, *e
);
4242 fprintf(f
, "%sRuntimeDirectoryPreserve: %s\n", prefix
, exec_preserve_mode_to_string(c
->runtime_directory_preserve_mode
));
4244 for (dt
= 0; dt
< _EXEC_DIRECTORY_TYPE_MAX
; dt
++) {
4245 fprintf(f
, "%s%sMode: %04o\n", prefix
, exec_directory_type_to_string(dt
), c
->directories
[dt
].mode
);
4247 STRV_FOREACH(d
, c
->directories
[dt
].paths
)
4248 fprintf(f
, "%s%s: %s\n", prefix
, exec_directory_type_to_string(dt
), *d
);
4256 if (c
->oom_score_adjust_set
)
4258 "%sOOMScoreAdjust: %i\n",
4259 prefix
, c
->oom_score_adjust
);
4261 for (i
= 0; i
< RLIM_NLIMITS
; i
++)
4263 fprintf(f
, "%sLimit%s: " RLIM_FMT
"\n",
4264 prefix
, rlimit_to_string(i
), c
->rlimit
[i
]->rlim_max
);
4265 fprintf(f
, "%sLimit%sSoft: " RLIM_FMT
"\n",
4266 prefix
, rlimit_to_string(i
), c
->rlimit
[i
]->rlim_cur
);
4269 if (c
->ioprio_set
) {
4270 _cleanup_free_
char *class_str
= NULL
;
4272 r
= ioprio_class_to_string_alloc(IOPRIO_PRIO_CLASS(c
->ioprio
), &class_str
);
4274 fprintf(f
, "%sIOSchedulingClass: %s\n", prefix
, class_str
);
4276 fprintf(f
, "%sIOPriority: %lu\n", prefix
, IOPRIO_PRIO_DATA(c
->ioprio
));
4279 if (c
->cpu_sched_set
) {
4280 _cleanup_free_
char *policy_str
= NULL
;
4282 r
= sched_policy_to_string_alloc(c
->cpu_sched_policy
, &policy_str
);
4284 fprintf(f
, "%sCPUSchedulingPolicy: %s\n", prefix
, policy_str
);
4287 "%sCPUSchedulingPriority: %i\n"
4288 "%sCPUSchedulingResetOnFork: %s\n",
4289 prefix
, c
->cpu_sched_priority
,
4290 prefix
, yes_no(c
->cpu_sched_reset_on_fork
));
4294 fprintf(f
, "%sCPUAffinity:", prefix
);
4295 for (i
= 0; i
< c
->cpuset_ncpus
; i
++)
4296 if (CPU_ISSET_S(i
, CPU_ALLOC_SIZE(c
->cpuset_ncpus
), c
->cpuset
))
4297 fprintf(f
, " %u", i
);
4301 if (c
->timer_slack_nsec
!= NSEC_INFINITY
)
4302 fprintf(f
, "%sTimerSlackNSec: "NSEC_FMT
"\n", prefix
, c
->timer_slack_nsec
);
4305 "%sStandardInput: %s\n"
4306 "%sStandardOutput: %s\n"
4307 "%sStandardError: %s\n",
4308 prefix
, exec_input_to_string(c
->std_input
),
4309 prefix
, exec_output_to_string(c
->std_output
),
4310 prefix
, exec_output_to_string(c
->std_error
));
4312 if (c
->std_input
== EXEC_INPUT_NAMED_FD
)
4313 fprintf(f
, "%sStandardInputFileDescriptorName: %s\n", prefix
, c
->stdio_fdname
[STDIN_FILENO
]);
4314 if (c
->std_output
== EXEC_OUTPUT_NAMED_FD
)
4315 fprintf(f
, "%sStandardOutputFileDescriptorName: %s\n", prefix
, c
->stdio_fdname
[STDOUT_FILENO
]);
4316 if (c
->std_error
== EXEC_OUTPUT_NAMED_FD
)
4317 fprintf(f
, "%sStandardErrorFileDescriptorName: %s\n", prefix
, c
->stdio_fdname
[STDERR_FILENO
]);
4319 if (c
->std_input
== EXEC_INPUT_FILE
)
4320 fprintf(f
, "%sStandardInputFile: %s\n", prefix
, c
->stdio_file
[STDIN_FILENO
]);
4321 if (c
->std_output
== EXEC_OUTPUT_FILE
)
4322 fprintf(f
, "%sStandardOutputFile: %s\n", prefix
, c
->stdio_file
[STDOUT_FILENO
]);
4323 if (c
->std_output
== EXEC_OUTPUT_FILE_APPEND
)
4324 fprintf(f
, "%sStandardOutputFileToAppend: %s\n", prefix
, c
->stdio_file
[STDOUT_FILENO
]);
4325 if (c
->std_error
== EXEC_OUTPUT_FILE
)
4326 fprintf(f
, "%sStandardErrorFile: %s\n", prefix
, c
->stdio_file
[STDERR_FILENO
]);
4327 if (c
->std_error
== EXEC_OUTPUT_FILE_APPEND
)
4328 fprintf(f
, "%sStandardErrorFileToAppend: %s\n", prefix
, c
->stdio_file
[STDERR_FILENO
]);
4334 "%sTTYVHangup: %s\n"
4335 "%sTTYVTDisallocate: %s\n",
4336 prefix
, c
->tty_path
,
4337 prefix
, yes_no(c
->tty_reset
),
4338 prefix
, yes_no(c
->tty_vhangup
),
4339 prefix
, yes_no(c
->tty_vt_disallocate
));
4341 if (IN_SET(c
->std_output
,
4344 EXEC_OUTPUT_JOURNAL
,
4345 EXEC_OUTPUT_SYSLOG_AND_CONSOLE
,
4346 EXEC_OUTPUT_KMSG_AND_CONSOLE
,
4347 EXEC_OUTPUT_JOURNAL_AND_CONSOLE
) ||
4348 IN_SET(c
->std_error
,
4351 EXEC_OUTPUT_JOURNAL
,
4352 EXEC_OUTPUT_SYSLOG_AND_CONSOLE
,
4353 EXEC_OUTPUT_KMSG_AND_CONSOLE
,
4354 EXEC_OUTPUT_JOURNAL_AND_CONSOLE
)) {
4356 _cleanup_free_
char *fac_str
= NULL
, *lvl_str
= NULL
;
4358 r
= log_facility_unshifted_to_string_alloc(c
->syslog_priority
>> 3, &fac_str
);
4360 fprintf(f
, "%sSyslogFacility: %s\n", prefix
, fac_str
);
4362 r
= log_level_to_string_alloc(LOG_PRI(c
->syslog_priority
), &lvl_str
);
4364 fprintf(f
, "%sSyslogLevel: %s\n", prefix
, lvl_str
);
4367 if (c
->log_level_max
>= 0) {
4368 _cleanup_free_
char *t
= NULL
;
4370 (void) log_level_to_string_alloc(c
->log_level_max
, &t
);
4372 fprintf(f
, "%sLogLevelMax: %s\n", prefix
, strna(t
));
4375 if (c
->log_rate_limit_interval_usec
> 0) {
4376 char buf_timespan
[FORMAT_TIMESPAN_MAX
];
4379 "%sLogRateLimitIntervalSec: %s\n",
4380 prefix
, format_timespan(buf_timespan
, sizeof(buf_timespan
), c
->log_rate_limit_interval_usec
, USEC_PER_SEC
));
4383 if (c
->log_rate_limit_burst
> 0)
4384 fprintf(f
, "%sLogRateLimitBurst: %u\n", prefix
, c
->log_rate_limit_burst
);
4386 if (c
->n_log_extra_fields
> 0) {
4389 for (j
= 0; j
< c
->n_log_extra_fields
; j
++) {
4390 fprintf(f
, "%sLogExtraFields: ", prefix
);
4391 fwrite(c
->log_extra_fields
[j
].iov_base
,
4392 1, c
->log_extra_fields
[j
].iov_len
,
4398 if (c
->secure_bits
) {
4399 _cleanup_free_
char *str
= NULL
;
4401 r
= secure_bits_to_string_alloc(c
->secure_bits
, &str
);
4403 fprintf(f
, "%sSecure Bits: %s\n", prefix
, str
);
4406 if (c
->capability_bounding_set
!= CAP_ALL
) {
4407 _cleanup_free_
char *str
= NULL
;
4409 r
= capability_set_to_string_alloc(c
->capability_bounding_set
, &str
);
4411 fprintf(f
, "%sCapabilityBoundingSet: %s\n", prefix
, str
);
4414 if (c
->capability_ambient_set
!= 0) {
4415 _cleanup_free_
char *str
= NULL
;
4417 r
= capability_set_to_string_alloc(c
->capability_ambient_set
, &str
);
4419 fprintf(f
, "%sAmbientCapabilities: %s\n", prefix
, str
);
4423 fprintf(f
, "%sUser: %s\n", prefix
, c
->user
);
4425 fprintf(f
, "%sGroup: %s\n", prefix
, c
->group
);
4427 fprintf(f
, "%sDynamicUser: %s\n", prefix
, yes_no(c
->dynamic_user
));
4429 if (!strv_isempty(c
->supplementary_groups
)) {
4430 fprintf(f
, "%sSupplementaryGroups:", prefix
);
4431 strv_fprintf(f
, c
->supplementary_groups
);
4436 fprintf(f
, "%sPAMName: %s\n", prefix
, c
->pam_name
);
4438 if (!strv_isempty(c
->read_write_paths
)) {
4439 fprintf(f
, "%sReadWritePaths:", prefix
);
4440 strv_fprintf(f
, c
->read_write_paths
);
4444 if (!strv_isempty(c
->read_only_paths
)) {
4445 fprintf(f
, "%sReadOnlyPaths:", prefix
);
4446 strv_fprintf(f
, c
->read_only_paths
);
4450 if (!strv_isempty(c
->inaccessible_paths
)) {
4451 fprintf(f
, "%sInaccessiblePaths:", prefix
);
4452 strv_fprintf(f
, c
->inaccessible_paths
);
4456 if (c
->n_bind_mounts
> 0)
4457 for (i
= 0; i
< c
->n_bind_mounts
; i
++)
4458 fprintf(f
, "%s%s: %s%s:%s:%s\n", prefix
,
4459 c
->bind_mounts
[i
].read_only
? "BindReadOnlyPaths" : "BindPaths",
4460 c
->bind_mounts
[i
].ignore_enoent
? "-": "",
4461 c
->bind_mounts
[i
].source
,
4462 c
->bind_mounts
[i
].destination
,
4463 c
->bind_mounts
[i
].recursive
? "rbind" : "norbind");
4465 if (c
->n_temporary_filesystems
> 0)
4466 for (i
= 0; i
< c
->n_temporary_filesystems
; i
++) {
4467 TemporaryFileSystem
*t
= c
->temporary_filesystems
+ i
;
4469 fprintf(f
, "%sTemporaryFileSystem: %s%s%s\n", prefix
,
4471 isempty(t
->options
) ? "" : ":",
4472 strempty(t
->options
));
4477 "%sUtmpIdentifier: %s\n",
4478 prefix
, c
->utmp_id
);
4480 if (c
->selinux_context
)
4482 "%sSELinuxContext: %s%s\n",
4483 prefix
, c
->selinux_context_ignore
? "-" : "", c
->selinux_context
);
4485 if (c
->apparmor_profile
)
4487 "%sAppArmorProfile: %s%s\n",
4488 prefix
, c
->apparmor_profile_ignore
? "-" : "", c
->apparmor_profile
);
4490 if (c
->smack_process_label
)
4492 "%sSmackProcessLabel: %s%s\n",
4493 prefix
, c
->smack_process_label_ignore
? "-" : "", c
->smack_process_label
);
4495 if (c
->personality
!= PERSONALITY_INVALID
)
4497 "%sPersonality: %s\n",
4498 prefix
, strna(personality_to_string(c
->personality
)));
4501 "%sLockPersonality: %s\n",
4502 prefix
, yes_no(c
->lock_personality
));
4504 if (c
->syscall_filter
) {
4512 "%sSystemCallFilter: ",
4515 if (!c
->syscall_whitelist
)
4519 HASHMAP_FOREACH_KEY(val
, id
, c
->syscall_filter
, j
) {
4520 _cleanup_free_
char *name
= NULL
;
4521 const char *errno_name
= NULL
;
4522 int num
= PTR_TO_INT(val
);
4529 name
= seccomp_syscall_resolve_num_arch(SCMP_ARCH_NATIVE
, PTR_TO_INT(id
) - 1);
4530 fputs(strna(name
), f
);
4533 errno_name
= errno_to_name(num
);
4535 fprintf(f
, ":%s", errno_name
);
4537 fprintf(f
, ":%d", num
);
4545 if (c
->syscall_archs
) {
4552 "%sSystemCallArchitectures:",
4556 SET_FOREACH(id
, c
->syscall_archs
, j
)
4557 fprintf(f
, " %s", strna(seccomp_arch_to_string(PTR_TO_UINT32(id
) - 1)));
4562 if (exec_context_restrict_namespaces_set(c
)) {
4563 _cleanup_free_
char *s
= NULL
;
4565 r
= namespace_flags_to_string(c
->restrict_namespaces
, &s
);
4567 fprintf(f
, "%sRestrictNamespaces: %s\n",
4571 if (c
->network_namespace_path
)
4573 "%sNetworkNamespacePath: %s\n",
4574 prefix
, c
->network_namespace_path
);
4576 if (c
->syscall_errno
> 0) {
4577 const char *errno_name
;
4579 fprintf(f
, "%sSystemCallErrorNumber: ", prefix
);
4581 errno_name
= errno_to_name(c
->syscall_errno
);
4583 fprintf(f
, "%s\n", errno_name
);
4585 fprintf(f
, "%d\n", c
->syscall_errno
);
4589 bool exec_context_maintains_privileges(const ExecContext
*c
) {
4592 /* Returns true if the process forked off would run under
4593 * an unchanged UID or as root. */
4598 if (streq(c
->user
, "root") || streq(c
->user
, "0"))
4604 int exec_context_get_effective_ioprio(const ExecContext
*c
) {
4612 p
= ioprio_get(IOPRIO_WHO_PROCESS
, 0);
4614 return IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE
, 4);
4619 void exec_context_free_log_extra_fields(ExecContext
*c
) {
4624 for (l
= 0; l
< c
->n_log_extra_fields
; l
++)
4625 free(c
->log_extra_fields
[l
].iov_base
);
4626 c
->log_extra_fields
= mfree(c
->log_extra_fields
);
4627 c
->n_log_extra_fields
= 0;
4630 void exec_status_start(ExecStatus
*s
, pid_t pid
) {
4637 dual_timestamp_get(&s
->start_timestamp
);
4640 void exec_status_exit(ExecStatus
*s
, const ExecContext
*context
, pid_t pid
, int code
, int status
) {
4643 if (s
->pid
!= pid
) {
4649 dual_timestamp_get(&s
->exit_timestamp
);
4655 if (context
->utmp_id
)
4656 (void) utmp_put_dead_process(context
->utmp_id
, pid
, code
, status
);
4658 exec_context_tty_reset(context
, NULL
);
4662 void exec_status_reset(ExecStatus
*s
) {
4665 *s
= (ExecStatus
) {};
4668 void exec_status_dump(const ExecStatus
*s
, FILE *f
, const char *prefix
) {
4669 char buf
[FORMAT_TIMESTAMP_MAX
];
4677 prefix
= strempty(prefix
);
4680 "%sPID: "PID_FMT
"\n",
4683 if (dual_timestamp_is_set(&s
->start_timestamp
))
4685 "%sStart Timestamp: %s\n",
4686 prefix
, format_timestamp(buf
, sizeof(buf
), s
->start_timestamp
.realtime
));
4688 if (dual_timestamp_is_set(&s
->exit_timestamp
))
4690 "%sExit Timestamp: %s\n"
4692 "%sExit Status: %i\n",
4693 prefix
, format_timestamp(buf
, sizeof(buf
), s
->exit_timestamp
.realtime
),
4694 prefix
, sigchld_code_to_string(s
->code
),
4698 static char *exec_command_line(char **argv
) {
4706 STRV_FOREACH(a
, argv
)
4714 STRV_FOREACH(a
, argv
) {
4721 if (strpbrk(*a
, WHITESPACE
)) {
4732 /* FIXME: this doesn't really handle arguments that have
4733 * spaces and ticks in them */
4738 static void exec_command_dump(ExecCommand
*c
, FILE *f
, const char *prefix
) {
4739 _cleanup_free_
char *cmd
= NULL
;
4740 const char *prefix2
;
4745 prefix
= strempty(prefix
);
4746 prefix2
= strjoina(prefix
, "\t");
4748 cmd
= exec_command_line(c
->argv
);
4750 "%sCommand Line: %s\n",
4751 prefix
, cmd
? cmd
: strerror(ENOMEM
));
4753 exec_status_dump(&c
->exec_status
, f
, prefix2
);
4756 void exec_command_dump_list(ExecCommand
*c
, FILE *f
, const char *prefix
) {
4759 prefix
= strempty(prefix
);
4761 LIST_FOREACH(command
, c
, c
)
4762 exec_command_dump(c
, f
, prefix
);
4765 void exec_command_append_list(ExecCommand
**l
, ExecCommand
*e
) {
4772 /* It's kind of important, that we keep the order here */
4773 LIST_FIND_TAIL(command
, *l
, end
);
4774 LIST_INSERT_AFTER(command
, *l
, end
, e
);
4779 int exec_command_set(ExecCommand
*c
, const char *path
, ...) {
4787 l
= strv_new_ap(path
, ap
);
4799 free_and_replace(c
->path
, p
);
4801 return strv_free_and_replace(c
->argv
, l
);
4804 int exec_command_append(ExecCommand
*c
, const char *path
, ...) {
4805 _cleanup_strv_free_
char **l
= NULL
;
4813 l
= strv_new_ap(path
, ap
);
4819 r
= strv_extend_strv(&c
->argv
, l
, false);
4826 static void *remove_tmpdir_thread(void *p
) {
4827 _cleanup_free_
char *path
= p
;
4829 (void) rm_rf(path
, REMOVE_ROOT
|REMOVE_PHYSICAL
);
4833 static ExecRuntime
* exec_runtime_free(ExecRuntime
*rt
, bool destroy
) {
4840 (void) hashmap_remove(rt
->manager
->exec_runtime_by_id
, rt
->id
);
4842 /* When destroy is true, then rm_rf tmp_dir and var_tmp_dir. */
4843 if (destroy
&& rt
->tmp_dir
) {
4844 log_debug("Spawning thread to nuke %s", rt
->tmp_dir
);
4846 r
= asynchronous_job(remove_tmpdir_thread
, rt
->tmp_dir
);
4848 log_warning_errno(r
, "Failed to nuke %s: %m", rt
->tmp_dir
);
4855 if (destroy
&& rt
->var_tmp_dir
) {
4856 log_debug("Spawning thread to nuke %s", rt
->var_tmp_dir
);
4858 r
= asynchronous_job(remove_tmpdir_thread
, rt
->var_tmp_dir
);
4860 log_warning_errno(r
, "Failed to nuke %s: %m", rt
->var_tmp_dir
);
4861 free(rt
->var_tmp_dir
);
4864 rt
->var_tmp_dir
= NULL
;
4867 rt
->id
= mfree(rt
->id
);
4868 rt
->tmp_dir
= mfree(rt
->tmp_dir
);
4869 rt
->var_tmp_dir
= mfree(rt
->var_tmp_dir
);
4870 safe_close_pair(rt
->netns_storage_socket
);
4874 static void exec_runtime_freep(ExecRuntime
**rt
) {
4875 (void) exec_runtime_free(*rt
, false);
4878 static int exec_runtime_allocate(ExecRuntime
**ret
) {
4883 n
= new(ExecRuntime
, 1);
4887 *n
= (ExecRuntime
) {
4888 .netns_storage_socket
= { -1, -1 },
4895 static int exec_runtime_add(
4898 const char *tmp_dir
,
4899 const char *var_tmp_dir
,
4900 const int netns_storage_socket
[2],
4901 ExecRuntime
**ret
) {
4903 _cleanup_(exec_runtime_freep
) ExecRuntime
*rt
= NULL
;
4909 r
= hashmap_ensure_allocated(&m
->exec_runtime_by_id
, &string_hash_ops
);
4913 r
= exec_runtime_allocate(&rt
);
4917 rt
->id
= strdup(id
);
4922 rt
->tmp_dir
= strdup(tmp_dir
);
4926 /* When tmp_dir is set, then we require var_tmp_dir is also set. */
4927 assert(var_tmp_dir
);
4928 rt
->var_tmp_dir
= strdup(var_tmp_dir
);
4929 if (!rt
->var_tmp_dir
)
4933 if (netns_storage_socket
) {
4934 rt
->netns_storage_socket
[0] = netns_storage_socket
[0];
4935 rt
->netns_storage_socket
[1] = netns_storage_socket
[1];
4938 r
= hashmap_put(m
->exec_runtime_by_id
, rt
->id
, rt
);
4947 /* do not remove created ExecRuntime object when the operation succeeds. */
4952 static int exec_runtime_make(Manager
*m
, const ExecContext
*c
, const char *id
, ExecRuntime
**ret
) {
4953 _cleanup_free_
char *tmp_dir
= NULL
, *var_tmp_dir
= NULL
;
4954 _cleanup_close_pair_
int netns_storage_socket
[2] = { -1, -1 };
4961 /* It is not necessary to create ExecRuntime object. */
4962 if (!c
->private_network
&& !c
->private_tmp
&& !c
->network_namespace_path
)
4965 if (c
->private_tmp
) {
4966 r
= setup_tmp_dirs(id
, &tmp_dir
, &var_tmp_dir
);
4971 if (c
->private_network
|| c
->network_namespace_path
) {
4972 if (socketpair(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
, 0, netns_storage_socket
) < 0)
4976 r
= exec_runtime_add(m
, id
, tmp_dir
, var_tmp_dir
, netns_storage_socket
, ret
);
4981 netns_storage_socket
[0] = netns_storage_socket
[1] = -1;
4985 int exec_runtime_acquire(Manager
*m
, const ExecContext
*c
, const char *id
, bool create
, ExecRuntime
**ret
) {
4993 rt
= hashmap_get(m
->exec_runtime_by_id
, id
);
4995 /* We already have a ExecRuntime object, let's increase the ref count and reuse it */
5001 /* If not found, then create a new object. */
5002 r
= exec_runtime_make(m
, c
, id
, &rt
);
5004 /* When r == 0, it is not necessary to create ExecRuntime object. */
5008 /* increment reference counter. */
5014 ExecRuntime
*exec_runtime_unref(ExecRuntime
*rt
, bool destroy
) {
5018 assert(rt
->n_ref
> 0);
5024 return exec_runtime_free(rt
, destroy
);
5027 int exec_runtime_serialize(const Manager
*m
, FILE *f
, FDSet
*fds
) {
5035 HASHMAP_FOREACH(rt
, m
->exec_runtime_by_id
, i
) {
5036 fprintf(f
, "exec-runtime=%s", rt
->id
);
5039 fprintf(f
, " tmp-dir=%s", rt
->tmp_dir
);
5041 if (rt
->var_tmp_dir
)
5042 fprintf(f
, " var-tmp-dir=%s", rt
->var_tmp_dir
);
5044 if (rt
->netns_storage_socket
[0] >= 0) {
5047 copy
= fdset_put_dup(fds
, rt
->netns_storage_socket
[0]);
5051 fprintf(f
, " netns-socket-0=%i", copy
);
5054 if (rt
->netns_storage_socket
[1] >= 0) {
5057 copy
= fdset_put_dup(fds
, rt
->netns_storage_socket
[1]);
5061 fprintf(f
, " netns-socket-1=%i", copy
);
5070 int exec_runtime_deserialize_compat(Unit
*u
, const char *key
, const char *value
, FDSet
*fds
) {
5071 _cleanup_(exec_runtime_freep
) ExecRuntime
*rt_create
= NULL
;
5075 /* This is for the migration from old (v237 or earlier) deserialization text.
5076 * Due to the bug #7790, this may not work with the units that use JoinsNamespaceOf=.
5077 * Even if the ExecRuntime object originally created by the other unit, we cannot judge
5078 * so or not from the serialized text, then we always creates a new object owned by this. */
5084 /* Manager manages ExecRuntime objects by the unit id.
5085 * So, we omit the serialized text when the unit does not have id (yet?)... */
5086 if (isempty(u
->id
)) {
5087 log_unit_debug(u
, "Invocation ID not found. Dropping runtime parameter.");
5091 r
= hashmap_ensure_allocated(&u
->manager
->exec_runtime_by_id
, &string_hash_ops
);
5093 log_unit_debug_errno(u
, r
, "Failed to allocate storage for runtime parameter: %m");
5097 rt
= hashmap_get(u
->manager
->exec_runtime_by_id
, u
->id
);
5099 r
= exec_runtime_allocate(&rt_create
);
5103 rt_create
->id
= strdup(u
->id
);
5110 if (streq(key
, "tmp-dir")) {
5113 copy
= strdup(value
);
5117 free_and_replace(rt
->tmp_dir
, copy
);
5119 } else if (streq(key
, "var-tmp-dir")) {
5122 copy
= strdup(value
);
5126 free_and_replace(rt
->var_tmp_dir
, copy
);
5128 } else if (streq(key
, "netns-socket-0")) {
5131 if (safe_atoi(value
, &fd
) < 0 || !fdset_contains(fds
, fd
)) {
5132 log_unit_debug(u
, "Failed to parse netns socket value: %s", value
);
5136 safe_close(rt
->netns_storage_socket
[0]);
5137 rt
->netns_storage_socket
[0] = fdset_remove(fds
, fd
);
5139 } else if (streq(key
, "netns-socket-1")) {
5142 if (safe_atoi(value
, &fd
) < 0 || !fdset_contains(fds
, fd
)) {
5143 log_unit_debug(u
, "Failed to parse netns socket value: %s", value
);
5147 safe_close(rt
->netns_storage_socket
[1]);
5148 rt
->netns_storage_socket
[1] = fdset_remove(fds
, fd
);
5152 /* If the object is newly created, then put it to the hashmap which manages ExecRuntime objects. */
5154 r
= hashmap_put(u
->manager
->exec_runtime_by_id
, rt_create
->id
, rt_create
);
5156 log_unit_debug_errno(u
, r
, "Failed to put runtime parameter to manager's storage: %m");
5160 rt_create
->manager
= u
->manager
;
5169 void exec_runtime_deserialize_one(Manager
*m
, const char *value
, FDSet
*fds
) {
5170 char *id
= NULL
, *tmp_dir
= NULL
, *var_tmp_dir
= NULL
;
5171 int r
, fd0
= -1, fd1
= -1;
5172 const char *p
, *v
= value
;
5179 n
= strcspn(v
, " ");
5180 id
= strndupa(v
, n
);
5185 v
= startswith(p
, "tmp-dir=");
5187 n
= strcspn(v
, " ");
5188 tmp_dir
= strndupa(v
, n
);
5194 v
= startswith(p
, "var-tmp-dir=");
5196 n
= strcspn(v
, " ");
5197 var_tmp_dir
= strndupa(v
, n
);
5203 v
= startswith(p
, "netns-socket-0=");
5207 n
= strcspn(v
, " ");
5208 buf
= strndupa(v
, n
);
5209 if (safe_atoi(buf
, &fd0
) < 0 || !fdset_contains(fds
, fd0
)) {
5210 log_debug("Unable to process exec-runtime netns fd specification.");
5213 fd0
= fdset_remove(fds
, fd0
);
5219 v
= startswith(p
, "netns-socket-1=");
5223 n
= strcspn(v
, " ");
5224 buf
= strndupa(v
, n
);
5225 if (safe_atoi(buf
, &fd1
) < 0 || !fdset_contains(fds
, fd1
)) {
5226 log_debug("Unable to process exec-runtime netns fd specification.");
5229 fd1
= fdset_remove(fds
, fd1
);
5234 r
= exec_runtime_add(m
, id
, tmp_dir
, var_tmp_dir
, (int[]) { fd0
, fd1
}, NULL
);
5236 log_debug_errno(r
, "Failed to add exec-runtime: %m");
5239 void exec_runtime_vacuum(Manager
*m
) {
5245 /* Free unreferenced ExecRuntime objects. This is used after manager deserialization process. */
5247 HASHMAP_FOREACH(rt
, m
->exec_runtime_by_id
, i
) {
5251 (void) exec_runtime_free(rt
, false);
5255 void exec_params_clear(ExecParameters
*p
) {
5259 strv_free(p
->environment
);
5262 static const char* const exec_input_table
[_EXEC_INPUT_MAX
] = {
5263 [EXEC_INPUT_NULL
] = "null",
5264 [EXEC_INPUT_TTY
] = "tty",
5265 [EXEC_INPUT_TTY_FORCE
] = "tty-force",
5266 [EXEC_INPUT_TTY_FAIL
] = "tty-fail",
5267 [EXEC_INPUT_SOCKET
] = "socket",
5268 [EXEC_INPUT_NAMED_FD
] = "fd",
5269 [EXEC_INPUT_DATA
] = "data",
5270 [EXEC_INPUT_FILE
] = "file",
5273 DEFINE_STRING_TABLE_LOOKUP(exec_input
, ExecInput
);
5275 static const char* const exec_output_table
[_EXEC_OUTPUT_MAX
] = {
5276 [EXEC_OUTPUT_INHERIT
] = "inherit",
5277 [EXEC_OUTPUT_NULL
] = "null",
5278 [EXEC_OUTPUT_TTY
] = "tty",
5279 [EXEC_OUTPUT_SYSLOG
] = "syslog",
5280 [EXEC_OUTPUT_SYSLOG_AND_CONSOLE
] = "syslog+console",
5281 [EXEC_OUTPUT_KMSG
] = "kmsg",
5282 [EXEC_OUTPUT_KMSG_AND_CONSOLE
] = "kmsg+console",
5283 [EXEC_OUTPUT_JOURNAL
] = "journal",
5284 [EXEC_OUTPUT_JOURNAL_AND_CONSOLE
] = "journal+console",
5285 [EXEC_OUTPUT_SOCKET
] = "socket",
5286 [EXEC_OUTPUT_NAMED_FD
] = "fd",
5287 [EXEC_OUTPUT_FILE
] = "file",
5288 [EXEC_OUTPUT_FILE_APPEND
] = "append",
5291 DEFINE_STRING_TABLE_LOOKUP(exec_output
, ExecOutput
);
5293 static const char* const exec_utmp_mode_table
[_EXEC_UTMP_MODE_MAX
] = {
5294 [EXEC_UTMP_INIT
] = "init",
5295 [EXEC_UTMP_LOGIN
] = "login",
5296 [EXEC_UTMP_USER
] = "user",
5299 DEFINE_STRING_TABLE_LOOKUP(exec_utmp_mode
, ExecUtmpMode
);
5301 static const char* const exec_preserve_mode_table
[_EXEC_PRESERVE_MODE_MAX
] = {
5302 [EXEC_PRESERVE_NO
] = "no",
5303 [EXEC_PRESERVE_YES
] = "yes",
5304 [EXEC_PRESERVE_RESTART
] = "restart",
5307 DEFINE_STRING_TABLE_LOOKUP_WITH_BOOLEAN(exec_preserve_mode
, ExecPreserveMode
, EXEC_PRESERVE_YES
);
5309 static const char* const exec_directory_type_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
5310 [EXEC_DIRECTORY_RUNTIME
] = "RuntimeDirectory",
5311 [EXEC_DIRECTORY_STATE
] = "StateDirectory",
5312 [EXEC_DIRECTORY_CACHE
] = "CacheDirectory",
5313 [EXEC_DIRECTORY_LOGS
] = "LogsDirectory",
5314 [EXEC_DIRECTORY_CONFIGURATION
] = "ConfigurationDirectory",
5317 DEFINE_STRING_TABLE_LOOKUP(exec_directory_type
, ExecDirectoryType
);
5319 static const char* const exec_directory_env_name_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
5320 [EXEC_DIRECTORY_RUNTIME
] = "RUNTIME_DIRECTORY",
5321 [EXEC_DIRECTORY_STATE
] = "STATE_DIRECTORY",
5322 [EXEC_DIRECTORY_CACHE
] = "CACHE_DIRECTORY",
5323 [EXEC_DIRECTORY_LOGS
] = "LOGS_DIRECTORY",
5324 [EXEC_DIRECTORY_CONFIGURATION
] = "CONFIGURATION_DIRECTORY",
5327 DEFINE_PRIVATE_STRING_TABLE_LOOKUP_TO_STRING(exec_directory_env_name
, ExecDirectoryType
);
5329 static const char* const exec_keyring_mode_table
[_EXEC_KEYRING_MODE_MAX
] = {
5330 [EXEC_KEYRING_INHERIT
] = "inherit",
5331 [EXEC_KEYRING_PRIVATE
] = "private",
5332 [EXEC_KEYRING_SHARED
] = "shared",
5335 DEFINE_STRING_TABLE_LOOKUP(exec_keyring_mode
, ExecKeyringMode
);