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 c
->restrict_suid_sgid
||
1408 exec_context_restrict_namespaces_set(c
) ||
1409 c
->protect_kernel_tunables
||
1410 c
->protect_kernel_modules
||
1411 c
->private_devices
||
1412 context_has_syscall_filters(c
) ||
1413 !set_isempty(c
->syscall_archs
) ||
1414 c
->lock_personality
||
1415 c
->protect_hostname
;
1420 static bool skip_seccomp_unavailable(const Unit
* u
, const char* msg
) {
1422 if (is_seccomp_available())
1425 log_unit_debug(u
, "SECCOMP features not detected in the kernel, skipping %s", msg
);
1429 static int apply_syscall_filter(const Unit
* u
, const ExecContext
*c
, bool needs_ambient_hack
) {
1430 uint32_t negative_action
, default_action
, action
;
1436 if (!context_has_syscall_filters(c
))
1439 if (skip_seccomp_unavailable(u
, "SystemCallFilter="))
1442 negative_action
= c
->syscall_errno
== 0 ? SCMP_ACT_KILL
: SCMP_ACT_ERRNO(c
->syscall_errno
);
1444 if (c
->syscall_whitelist
) {
1445 default_action
= negative_action
;
1446 action
= SCMP_ACT_ALLOW
;
1448 default_action
= SCMP_ACT_ALLOW
;
1449 action
= negative_action
;
1452 if (needs_ambient_hack
) {
1453 r
= seccomp_filter_set_add(c
->syscall_filter
, c
->syscall_whitelist
, syscall_filter_sets
+ SYSCALL_FILTER_SET_SETUID
);
1458 return seccomp_load_syscall_filter_set_raw(default_action
, c
->syscall_filter
, action
, false);
1461 static int apply_syscall_archs(const Unit
*u
, const ExecContext
*c
) {
1465 if (set_isempty(c
->syscall_archs
))
1468 if (skip_seccomp_unavailable(u
, "SystemCallArchitectures="))
1471 return seccomp_restrict_archs(c
->syscall_archs
);
1474 static int apply_address_families(const Unit
* u
, const ExecContext
*c
) {
1478 if (!context_has_address_families(c
))
1481 if (skip_seccomp_unavailable(u
, "RestrictAddressFamilies="))
1484 return seccomp_restrict_address_families(c
->address_families
, c
->address_families_whitelist
);
1487 static int apply_memory_deny_write_execute(const Unit
* u
, const ExecContext
*c
) {
1491 if (!c
->memory_deny_write_execute
)
1494 if (skip_seccomp_unavailable(u
, "MemoryDenyWriteExecute="))
1497 return seccomp_memory_deny_write_execute();
1500 static int apply_restrict_realtime(const Unit
* u
, const ExecContext
*c
) {
1504 if (!c
->restrict_realtime
)
1507 if (skip_seccomp_unavailable(u
, "RestrictRealtime="))
1510 return seccomp_restrict_realtime();
1513 static int apply_restrict_suid_sgid(const Unit
* u
, const ExecContext
*c
) {
1517 if (!c
->restrict_suid_sgid
)
1520 if (skip_seccomp_unavailable(u
, "RestrictSUIDSGID="))
1523 return seccomp_restrict_suid_sgid();
1526 static int apply_protect_sysctl(const Unit
*u
, const ExecContext
*c
) {
1530 /* Turn off the legacy sysctl() system call. Many distributions turn this off while building the kernel, but
1531 * let's protect even those systems where this is left on in the kernel. */
1533 if (!c
->protect_kernel_tunables
)
1536 if (skip_seccomp_unavailable(u
, "ProtectKernelTunables="))
1539 return seccomp_protect_sysctl();
1542 static int apply_protect_kernel_modules(const Unit
*u
, const ExecContext
*c
) {
1546 /* Turn off module syscalls on ProtectKernelModules=yes */
1548 if (!c
->protect_kernel_modules
)
1551 if (skip_seccomp_unavailable(u
, "ProtectKernelModules="))
1554 return seccomp_load_syscall_filter_set(SCMP_ACT_ALLOW
, syscall_filter_sets
+ SYSCALL_FILTER_SET_MODULE
, SCMP_ACT_ERRNO(EPERM
), false);
1557 static int apply_private_devices(const Unit
*u
, const ExecContext
*c
) {
1561 /* If PrivateDevices= is set, also turn off iopl and all @raw-io syscalls. */
1563 if (!c
->private_devices
)
1566 if (skip_seccomp_unavailable(u
, "PrivateDevices="))
1569 return seccomp_load_syscall_filter_set(SCMP_ACT_ALLOW
, syscall_filter_sets
+ SYSCALL_FILTER_SET_RAW_IO
, SCMP_ACT_ERRNO(EPERM
), false);
1572 static int apply_restrict_namespaces(const Unit
*u
, const ExecContext
*c
) {
1576 if (!exec_context_restrict_namespaces_set(c
))
1579 if (skip_seccomp_unavailable(u
, "RestrictNamespaces="))
1582 return seccomp_restrict_namespaces(c
->restrict_namespaces
);
1585 static int apply_lock_personality(const Unit
* u
, const ExecContext
*c
) {
1586 unsigned long personality
;
1592 if (!c
->lock_personality
)
1595 if (skip_seccomp_unavailable(u
, "LockPersonality="))
1598 personality
= c
->personality
;
1600 /* If personality is not specified, use either PER_LINUX or PER_LINUX32 depending on what is currently set. */
1601 if (personality
== PERSONALITY_INVALID
) {
1603 r
= opinionated_personality(&personality
);
1608 return seccomp_lock_personality(personality
);
1613 static void do_idle_pipe_dance(int idle_pipe
[static 4]) {
1616 idle_pipe
[1] = safe_close(idle_pipe
[1]);
1617 idle_pipe
[2] = safe_close(idle_pipe
[2]);
1619 if (idle_pipe
[0] >= 0) {
1622 r
= fd_wait_for_event(idle_pipe
[0], POLLHUP
, IDLE_TIMEOUT_USEC
);
1624 if (idle_pipe
[3] >= 0 && r
== 0 /* timeout */) {
1627 /* Signal systemd that we are bored and want to continue. */
1628 n
= write(idle_pipe
[3], "x", 1);
1630 /* Wait for systemd to react to the signal above. */
1631 fd_wait_for_event(idle_pipe
[0], POLLHUP
, IDLE_TIMEOUT2_USEC
);
1634 idle_pipe
[0] = safe_close(idle_pipe
[0]);
1638 idle_pipe
[3] = safe_close(idle_pipe
[3]);
1641 static const char *exec_directory_env_name_to_string(ExecDirectoryType t
);
1643 static int build_environment(
1645 const ExecContext
*c
,
1646 const ExecParameters
*p
,
1649 const char *username
,
1651 dev_t journal_stream_dev
,
1652 ino_t journal_stream_ino
,
1655 _cleanup_strv_free_
char **our_env
= NULL
;
1656 ExecDirectoryType t
;
1665 our_env
= new0(char*, 14 + _EXEC_DIRECTORY_TYPE_MAX
);
1670 _cleanup_free_
char *joined
= NULL
;
1672 if (asprintf(&x
, "LISTEN_PID="PID_FMT
, getpid_cached()) < 0)
1674 our_env
[n_env
++] = x
;
1676 if (asprintf(&x
, "LISTEN_FDS=%zu", n_fds
) < 0)
1678 our_env
[n_env
++] = x
;
1680 joined
= strv_join(p
->fd_names
, ":");
1684 x
= strjoin("LISTEN_FDNAMES=", joined
);
1687 our_env
[n_env
++] = x
;
1690 if ((p
->flags
& EXEC_SET_WATCHDOG
) && p
->watchdog_usec
> 0) {
1691 if (asprintf(&x
, "WATCHDOG_PID="PID_FMT
, getpid_cached()) < 0)
1693 our_env
[n_env
++] = x
;
1695 if (asprintf(&x
, "WATCHDOG_USEC="USEC_FMT
, p
->watchdog_usec
) < 0)
1697 our_env
[n_env
++] = x
;
1700 /* If this is D-Bus, tell the nss-systemd module, since it relies on being able to use D-Bus look up dynamic
1701 * users via PID 1, possibly dead-locking the dbus daemon. This way it will not use D-Bus to resolve names, but
1702 * check the database directly. */
1703 if (p
->flags
& EXEC_NSS_BYPASS_BUS
) {
1704 x
= strdup("SYSTEMD_NSS_BYPASS_BUS=1");
1707 our_env
[n_env
++] = x
;
1711 x
= strappend("HOME=", home
);
1715 path_simplify(x
+ 5, true);
1716 our_env
[n_env
++] = x
;
1720 x
= strappend("LOGNAME=", username
);
1723 our_env
[n_env
++] = x
;
1725 x
= strappend("USER=", username
);
1728 our_env
[n_env
++] = x
;
1732 x
= strappend("SHELL=", shell
);
1736 path_simplify(x
+ 6, true);
1737 our_env
[n_env
++] = x
;
1740 if (!sd_id128_is_null(u
->invocation_id
)) {
1741 if (asprintf(&x
, "INVOCATION_ID=" SD_ID128_FORMAT_STR
, SD_ID128_FORMAT_VAL(u
->invocation_id
)) < 0)
1744 our_env
[n_env
++] = x
;
1747 if (exec_context_needs_term(c
)) {
1748 const char *tty_path
, *term
= NULL
;
1750 tty_path
= exec_context_tty_path(c
);
1752 /* If we are forked off PID 1 and we are supposed to operate on /dev/console, then let's try to inherit
1753 * the $TERM set for PID 1. This is useful for containers so that the $TERM the container manager
1754 * passes to PID 1 ends up all the way in the console login shown. */
1756 if (path_equal(tty_path
, "/dev/console") && getppid() == 1)
1757 term
= getenv("TERM");
1759 term
= default_term_for_tty(tty_path
);
1761 x
= strappend("TERM=", term
);
1764 our_env
[n_env
++] = x
;
1767 if (journal_stream_dev
!= 0 && journal_stream_ino
!= 0) {
1768 if (asprintf(&x
, "JOURNAL_STREAM=" DEV_FMT
":" INO_FMT
, journal_stream_dev
, journal_stream_ino
) < 0)
1771 our_env
[n_env
++] = x
;
1774 for (t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
1775 _cleanup_free_
char *pre
= NULL
, *joined
= NULL
;
1781 if (strv_isempty(c
->directories
[t
].paths
))
1784 n
= exec_directory_env_name_to_string(t
);
1788 pre
= strjoin(p
->prefix
[t
], "/");
1792 joined
= strv_join_prefix(c
->directories
[t
].paths
, ":", pre
);
1796 x
= strjoin(n
, "=", joined
);
1800 our_env
[n_env
++] = x
;
1803 our_env
[n_env
++] = NULL
;
1804 assert(n_env
<= 14 + _EXEC_DIRECTORY_TYPE_MAX
);
1806 *ret
= TAKE_PTR(our_env
);
1811 static int build_pass_environment(const ExecContext
*c
, char ***ret
) {
1812 _cleanup_strv_free_
char **pass_env
= NULL
;
1813 size_t n_env
= 0, n_bufsize
= 0;
1816 STRV_FOREACH(i
, c
->pass_environment
) {
1817 _cleanup_free_
char *x
= NULL
;
1823 x
= strjoin(*i
, "=", v
);
1827 if (!GREEDY_REALLOC(pass_env
, n_bufsize
, n_env
+ 2))
1830 pass_env
[n_env
++] = TAKE_PTR(x
);
1831 pass_env
[n_env
] = NULL
;
1834 *ret
= TAKE_PTR(pass_env
);
1839 static bool exec_needs_mount_namespace(
1840 const ExecContext
*context
,
1841 const ExecParameters
*params
,
1842 const ExecRuntime
*runtime
) {
1847 if (context
->root_image
)
1850 if (!strv_isempty(context
->read_write_paths
) ||
1851 !strv_isempty(context
->read_only_paths
) ||
1852 !strv_isempty(context
->inaccessible_paths
))
1855 if (context
->n_bind_mounts
> 0)
1858 if (context
->n_temporary_filesystems
> 0)
1861 if (!IN_SET(context
->mount_flags
, 0, MS_SHARED
))
1864 if (context
->private_tmp
&& runtime
&& (runtime
->tmp_dir
|| runtime
->var_tmp_dir
))
1867 if (context
->private_devices
||
1868 context
->private_mounts
||
1869 context
->protect_system
!= PROTECT_SYSTEM_NO
||
1870 context
->protect_home
!= PROTECT_HOME_NO
||
1871 context
->protect_kernel_tunables
||
1872 context
->protect_kernel_modules
||
1873 context
->protect_control_groups
)
1876 if (context
->root_directory
) {
1877 ExecDirectoryType t
;
1879 if (context
->mount_apivfs
)
1882 for (t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
1883 if (!params
->prefix
[t
])
1886 if (!strv_isempty(context
->directories
[t
].paths
))
1891 if (context
->dynamic_user
&&
1892 (!strv_isempty(context
->directories
[EXEC_DIRECTORY_STATE
].paths
) ||
1893 !strv_isempty(context
->directories
[EXEC_DIRECTORY_CACHE
].paths
) ||
1894 !strv_isempty(context
->directories
[EXEC_DIRECTORY_LOGS
].paths
)))
1900 static int setup_private_users(uid_t uid
, gid_t gid
) {
1901 _cleanup_free_
char *uid_map
= NULL
, *gid_map
= NULL
;
1902 _cleanup_close_pair_
int errno_pipe
[2] = { -1, -1 };
1903 _cleanup_close_
int unshare_ready_fd
= -1;
1904 _cleanup_(sigkill_waitp
) pid_t pid
= 0;
1909 /* Set up a user namespace and map root to root, the selected UID/GID to itself, and everything else to
1910 * nobody. In order to be able to write this mapping we need CAP_SETUID in the original user namespace, which
1911 * we however lack after opening the user namespace. To work around this we fork() a temporary child process,
1912 * which waits for the parent to create the new user namespace while staying in the original namespace. The
1913 * child then writes the UID mapping, under full privileges. The parent waits for the child to finish and
1914 * continues execution normally. */
1916 if (uid
!= 0 && uid_is_valid(uid
)) {
1917 r
= asprintf(&uid_map
,
1918 "0 0 1\n" /* Map root → root */
1919 UID_FMT
" " UID_FMT
" 1\n", /* Map $UID → $UID */
1924 uid_map
= strdup("0 0 1\n"); /* The case where the above is the same */
1929 if (gid
!= 0 && gid_is_valid(gid
)) {
1930 r
= asprintf(&gid_map
,
1931 "0 0 1\n" /* Map root → root */
1932 GID_FMT
" " GID_FMT
" 1\n", /* Map $GID → $GID */
1937 gid_map
= strdup("0 0 1\n"); /* The case where the above is the same */
1942 /* Create a communication channel so that the parent can tell the child when it finished creating the user
1944 unshare_ready_fd
= eventfd(0, EFD_CLOEXEC
);
1945 if (unshare_ready_fd
< 0)
1948 /* Create a communication channel so that the child can tell the parent a proper error code in case it
1950 if (pipe2(errno_pipe
, O_CLOEXEC
) < 0)
1953 r
= safe_fork("(sd-userns)", FORK_RESET_SIGNALS
|FORK_DEATHSIG
, &pid
);
1957 _cleanup_close_
int fd
= -1;
1961 /* Child process, running in the original user namespace. Let's update the parent's UID/GID map from
1962 * here, after the parent opened its own user namespace. */
1965 errno_pipe
[0] = safe_close(errno_pipe
[0]);
1967 /* Wait until the parent unshared the user namespace */
1968 if (read(unshare_ready_fd
, &c
, sizeof(c
)) < 0) {
1973 /* Disable the setgroups() system call in the child user namespace, for good. */
1974 a
= procfs_file_alloca(ppid
, "setgroups");
1975 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
1977 if (errno
!= ENOENT
) {
1982 /* If the file is missing the kernel is too old, let's continue anyway. */
1984 if (write(fd
, "deny\n", 5) < 0) {
1989 fd
= safe_close(fd
);
1992 /* First write the GID map */
1993 a
= procfs_file_alloca(ppid
, "gid_map");
1994 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
1999 if (write(fd
, gid_map
, strlen(gid_map
)) < 0) {
2003 fd
= safe_close(fd
);
2005 /* The write the UID map */
2006 a
= procfs_file_alloca(ppid
, "uid_map");
2007 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
2012 if (write(fd
, uid_map
, strlen(uid_map
)) < 0) {
2017 _exit(EXIT_SUCCESS
);
2020 (void) write(errno_pipe
[1], &r
, sizeof(r
));
2021 _exit(EXIT_FAILURE
);
2024 errno_pipe
[1] = safe_close(errno_pipe
[1]);
2026 if (unshare(CLONE_NEWUSER
) < 0)
2029 /* Let the child know that the namespace is ready now */
2030 if (write(unshare_ready_fd
, &c
, sizeof(c
)) < 0)
2033 /* Try to read an error code from the child */
2034 n
= read(errno_pipe
[0], &r
, sizeof(r
));
2037 if (n
== sizeof(r
)) { /* an error code was sent to us */
2042 if (n
!= 0) /* on success we should have read 0 bytes */
2045 r
= wait_for_terminate_and_check("(sd-userns)", pid
, 0);
2049 if (r
!= EXIT_SUCCESS
) /* If something strange happened with the child, let's consider this fatal, too */
2055 static int setup_exec_directory(
2056 const ExecContext
*context
,
2057 const ExecParameters
*params
,
2060 ExecDirectoryType type
,
2063 static const int exit_status_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
2064 [EXEC_DIRECTORY_RUNTIME
] = EXIT_RUNTIME_DIRECTORY
,
2065 [EXEC_DIRECTORY_STATE
] = EXIT_STATE_DIRECTORY
,
2066 [EXEC_DIRECTORY_CACHE
] = EXIT_CACHE_DIRECTORY
,
2067 [EXEC_DIRECTORY_LOGS
] = EXIT_LOGS_DIRECTORY
,
2068 [EXEC_DIRECTORY_CONFIGURATION
] = EXIT_CONFIGURATION_DIRECTORY
,
2075 assert(type
>= 0 && type
< _EXEC_DIRECTORY_TYPE_MAX
);
2076 assert(exit_status
);
2078 if (!params
->prefix
[type
])
2081 if (params
->flags
& EXEC_CHOWN_DIRECTORIES
) {
2082 if (!uid_is_valid(uid
))
2084 if (!gid_is_valid(gid
))
2088 STRV_FOREACH(rt
, context
->directories
[type
].paths
) {
2089 _cleanup_free_
char *p
= NULL
, *pp
= NULL
;
2091 p
= path_join(params
->prefix
[type
], *rt
);
2097 r
= mkdir_parents_label(p
, 0755);
2101 if (context
->dynamic_user
&&
2102 (!IN_SET(type
, EXEC_DIRECTORY_RUNTIME
, EXEC_DIRECTORY_CONFIGURATION
) ||
2103 (type
== EXEC_DIRECTORY_RUNTIME
&& context
->runtime_directory_preserve_mode
!= EXEC_PRESERVE_NO
))) {
2104 _cleanup_free_
char *private_root
= NULL
;
2106 /* So, here's one extra complication when dealing with DynamicUser=1 units. In that case we
2107 * want to avoid leaving a directory around fully accessible that is owned by a dynamic user
2108 * whose UID is later on reused. To lock this down we use the same trick used by container
2109 * managers to prohibit host users to get access to files of the same UID in containers: we
2110 * place everything inside a directory that has an access mode of 0700 and is owned root:root,
2111 * so that it acts as security boundary for unprivileged host code. We then use fs namespacing
2112 * to make this directory permeable for the service itself.
2114 * Specifically: for a service which wants a special directory "foo/" we first create a
2115 * directory "private/" with access mode 0700 owned by root:root. Then we place "foo" inside of
2116 * that directory (i.e. "private/foo/"), and make "foo" a symlink to "private/foo". This way,
2117 * privileged host users can access "foo/" as usual, but unprivileged host users can't look
2118 * into it. Inside of the namespaceof the container "private/" is replaced by a more liberally
2119 * accessible tmpfs, into which the host's "private/foo/" is mounted under the same name, thus
2120 * disabling the access boundary for the service and making sure it only gets access to the
2121 * dirs it needs but no others. Tricky? Yes, absolutely, but it works!
2123 * Note that we don't do this for EXEC_DIRECTORY_CONFIGURATION as that's assumed not to be
2124 * owned by the service itself.
2125 * Also, note that we don't do this for EXEC_DIRECTORY_RUNTIME as that's often used for sharing
2126 * files or sockets with other services. */
2128 private_root
= path_join(params
->prefix
[type
], "private");
2129 if (!private_root
) {
2134 /* First set up private root if it doesn't exist yet, with access mode 0700 and owned by root:root */
2135 r
= mkdir_safe_label(private_root
, 0700, 0, 0, MKDIR_WARN_MODE
);
2139 pp
= path_join(private_root
, *rt
);
2145 /* Create all directories between the configured directory and this private root, and mark them 0755 */
2146 r
= mkdir_parents_label(pp
, 0755);
2150 if (is_dir(p
, false) > 0 &&
2151 (laccess(pp
, F_OK
) < 0 && errno
== ENOENT
)) {
2153 /* Hmm, the private directory doesn't exist yet, but the normal one exists? If so, move
2154 * it over. Most likely the service has been upgraded from one that didn't use
2155 * DynamicUser=1, to one that does. */
2157 if (rename(p
, pp
) < 0) {
2162 /* Otherwise, create the actual directory for the service */
2164 r
= mkdir_label(pp
, context
->directories
[type
].mode
);
2165 if (r
< 0 && r
!= -EEXIST
)
2169 /* And link it up from the original place */
2170 r
= symlink_idempotent(pp
, p
, true);
2175 r
= mkdir_label(p
, context
->directories
[type
].mode
);
2180 if (type
== EXEC_DIRECTORY_CONFIGURATION
) {
2183 /* Don't change the owner/access mode of the configuration directory,
2184 * as in the common case it is not written to by a service, and shall
2185 * not be writable. */
2187 if (stat(p
, &st
) < 0) {
2192 /* Still complain if the access mode doesn't match */
2193 if (((st
.st_mode
^ context
->directories
[type
].mode
) & 07777) != 0)
2194 log_warning("%s \'%s\' already exists but the mode is different. "
2195 "(File system: %o %sMode: %o)",
2196 exec_directory_type_to_string(type
), *rt
,
2197 st
.st_mode
& 07777, exec_directory_type_to_string(type
), context
->directories
[type
].mode
& 07777);
2204 /* Lock down the access mode (we use chmod_and_chown() to make this idempotent. We don't
2205 * specify UID/GID here, so that path_chown_recursive() can optimize things depending on the
2206 * current UID/GID ownership.) */
2207 r
= chmod_and_chown(pp
?: p
, context
->directories
[type
].mode
, UID_INVALID
, GID_INVALID
);
2211 /* Then, change the ownership of the whole tree, if necessary. When dynamic users are used we
2212 * drop the suid/sgid bits, since we really don't want SUID/SGID files for dynamic UID/GID
2213 * assignments to exist.*/
2214 r
= path_chown_recursive(pp
?: p
, uid
, gid
, context
->dynamic_user
? 01777 : 07777);
2222 *exit_status
= exit_status_table
[type
];
2227 static int setup_smack(
2228 const ExecContext
*context
,
2229 const ExecCommand
*command
) {
2236 if (context
->smack_process_label
) {
2237 r
= mac_smack_apply_pid(0, context
->smack_process_label
);
2241 #ifdef SMACK_DEFAULT_PROCESS_LABEL
2243 _cleanup_free_
char *exec_label
= NULL
;
2245 r
= mac_smack_read(command
->path
, SMACK_ATTR_EXEC
, &exec_label
);
2246 if (r
< 0 && !IN_SET(r
, -ENODATA
, -EOPNOTSUPP
))
2249 r
= mac_smack_apply_pid(0, exec_label
? : SMACK_DEFAULT_PROCESS_LABEL
);
2259 static int compile_bind_mounts(
2260 const ExecContext
*context
,
2261 const ExecParameters
*params
,
2262 BindMount
**ret_bind_mounts
,
2263 size_t *ret_n_bind_mounts
,
2264 char ***ret_empty_directories
) {
2266 _cleanup_strv_free_
char **empty_directories
= NULL
;
2267 BindMount
*bind_mounts
;
2269 ExecDirectoryType t
;
2274 assert(ret_bind_mounts
);
2275 assert(ret_n_bind_mounts
);
2276 assert(ret_empty_directories
);
2278 n
= context
->n_bind_mounts
;
2279 for (t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
2280 if (!params
->prefix
[t
])
2283 n
+= strv_length(context
->directories
[t
].paths
);
2287 *ret_bind_mounts
= NULL
;
2288 *ret_n_bind_mounts
= 0;
2289 *ret_empty_directories
= NULL
;
2293 bind_mounts
= new(BindMount
, n
);
2297 for (i
= 0; i
< context
->n_bind_mounts
; i
++) {
2298 BindMount
*item
= context
->bind_mounts
+ i
;
2301 s
= strdup(item
->source
);
2307 d
= strdup(item
->destination
);
2314 bind_mounts
[h
++] = (BindMount
) {
2317 .read_only
= item
->read_only
,
2318 .recursive
= item
->recursive
,
2319 .ignore_enoent
= item
->ignore_enoent
,
2323 for (t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
2326 if (!params
->prefix
[t
])
2329 if (strv_isempty(context
->directories
[t
].paths
))
2332 if (context
->dynamic_user
&&
2333 !IN_SET(t
, EXEC_DIRECTORY_RUNTIME
, EXEC_DIRECTORY_CONFIGURATION
) &&
2334 !(context
->root_directory
|| context
->root_image
)) {
2337 /* So this is for a dynamic user, and we need to make sure the process can access its own
2338 * directory. For that we overmount the usually inaccessible "private" subdirectory with a
2339 * tmpfs that makes it accessible and is empty except for the submounts we do this for. */
2341 private_root
= strjoin(params
->prefix
[t
], "/private");
2342 if (!private_root
) {
2347 r
= strv_consume(&empty_directories
, private_root
);
2352 STRV_FOREACH(suffix
, context
->directories
[t
].paths
) {
2355 if (context
->dynamic_user
&&
2356 !IN_SET(t
, EXEC_DIRECTORY_RUNTIME
, EXEC_DIRECTORY_CONFIGURATION
))
2357 s
= strjoin(params
->prefix
[t
], "/private/", *suffix
);
2359 s
= strjoin(params
->prefix
[t
], "/", *suffix
);
2365 if (context
->dynamic_user
&&
2366 !IN_SET(t
, EXEC_DIRECTORY_RUNTIME
, EXEC_DIRECTORY_CONFIGURATION
) &&
2367 (context
->root_directory
|| context
->root_image
))
2368 /* When RootDirectory= or RootImage= are set, then the symbolic link to the private
2369 * directory is not created on the root directory. So, let's bind-mount the directory
2370 * on the 'non-private' place. */
2371 d
= strjoin(params
->prefix
[t
], "/", *suffix
);
2380 bind_mounts
[h
++] = (BindMount
) {
2384 .nosuid
= context
->dynamic_user
, /* don't allow suid/sgid when DynamicUser= is on */
2386 .ignore_enoent
= false,
2393 *ret_bind_mounts
= bind_mounts
;
2394 *ret_n_bind_mounts
= n
;
2395 *ret_empty_directories
= TAKE_PTR(empty_directories
);
2400 bind_mount_free_many(bind_mounts
, h
);
2404 static int apply_mount_namespace(
2406 const ExecCommand
*command
,
2407 const ExecContext
*context
,
2408 const ExecParameters
*params
,
2409 const ExecRuntime
*runtime
,
2410 char **error_path
) {
2412 _cleanup_strv_free_
char **empty_directories
= NULL
;
2413 char *tmp
= NULL
, *var
= NULL
;
2414 const char *root_dir
= NULL
, *root_image
= NULL
;
2415 NamespaceInfo ns_info
;
2416 bool needs_sandboxing
;
2417 BindMount
*bind_mounts
= NULL
;
2418 size_t n_bind_mounts
= 0;
2423 /* The runtime struct only contains the parent of the private /tmp,
2424 * which is non-accessible to world users. Inside of it there's a /tmp
2425 * that is sticky, and that's the one we want to use here. */
2427 if (context
->private_tmp
&& runtime
) {
2428 if (runtime
->tmp_dir
)
2429 tmp
= strjoina(runtime
->tmp_dir
, "/tmp");
2430 if (runtime
->var_tmp_dir
)
2431 var
= strjoina(runtime
->var_tmp_dir
, "/tmp");
2434 if (params
->flags
& EXEC_APPLY_CHROOT
) {
2435 root_image
= context
->root_image
;
2438 root_dir
= context
->root_directory
;
2441 r
= compile_bind_mounts(context
, params
, &bind_mounts
, &n_bind_mounts
, &empty_directories
);
2445 needs_sandboxing
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && !(command
->flags
& EXEC_COMMAND_FULLY_PRIVILEGED
);
2446 if (needs_sandboxing
)
2447 ns_info
= (NamespaceInfo
) {
2448 .ignore_protect_paths
= false,
2449 .private_dev
= context
->private_devices
,
2450 .protect_control_groups
= context
->protect_control_groups
,
2451 .protect_kernel_tunables
= context
->protect_kernel_tunables
,
2452 .protect_kernel_modules
= context
->protect_kernel_modules
,
2453 .protect_hostname
= context
->protect_hostname
,
2454 .mount_apivfs
= context
->mount_apivfs
,
2455 .private_mounts
= context
->private_mounts
,
2457 else if (!context
->dynamic_user
&& root_dir
)
2459 * If DynamicUser=no and RootDirectory= is set then lets pass a relaxed
2460 * sandbox info, otherwise enforce it, don't ignore protected paths and
2461 * fail if we are enable to apply the sandbox inside the mount namespace.
2463 ns_info
= (NamespaceInfo
) {
2464 .ignore_protect_paths
= true,
2467 ns_info
= (NamespaceInfo
) {};
2469 if (context
->mount_flags
== MS_SHARED
)
2470 log_unit_debug(u
, "shared mount propagation hidden by other fs namespacing unit settings: ignoring");
2472 r
= setup_namespace(root_dir
, root_image
,
2473 &ns_info
, context
->read_write_paths
,
2474 needs_sandboxing
? context
->read_only_paths
: NULL
,
2475 needs_sandboxing
? context
->inaccessible_paths
: NULL
,
2479 context
->temporary_filesystems
,
2480 context
->n_temporary_filesystems
,
2483 needs_sandboxing
? context
->protect_home
: PROTECT_HOME_NO
,
2484 needs_sandboxing
? context
->protect_system
: PROTECT_SYSTEM_NO
,
2485 context
->mount_flags
,
2486 DISSECT_IMAGE_DISCARD_ON_LOOP
,
2489 bind_mount_free_many(bind_mounts
, n_bind_mounts
);
2491 /* If we couldn't set up the namespace this is probably due to a missing capability. setup_namespace() reports
2492 * that with a special, recognizable error ENOANO. In this case, silently proceed, but only if exclusively
2493 * sandboxing options were used, i.e. nothing such as RootDirectory= or BindMount= that would result in a
2494 * completely different execution environment. */
2496 if (n_bind_mounts
== 0 &&
2497 context
->n_temporary_filesystems
== 0 &&
2498 !root_dir
&& !root_image
&&
2499 !context
->dynamic_user
) {
2500 log_unit_debug(u
, "Failed to set up namespace, assuming containerized execution and ignoring.");
2504 log_unit_debug(u
, "Failed to set up namespace, and refusing to continue since the selected namespacing options alter mount environment non-trivially.\n"
2505 "Bind mounts: %zu, temporary filesystems: %zu, root directory: %s, root image: %s, dynamic user: %s",
2506 n_bind_mounts
, context
->n_temporary_filesystems
, yes_no(root_dir
), yes_no(root_image
), yes_no(context
->dynamic_user
));
2514 static int apply_working_directory(
2515 const ExecContext
*context
,
2516 const ExecParameters
*params
,
2518 const bool needs_mount_ns
,
2524 assert(exit_status
);
2526 if (context
->working_directory_home
) {
2529 *exit_status
= EXIT_CHDIR
;
2535 } else if (context
->working_directory
)
2536 wd
= context
->working_directory
;
2540 if (params
->flags
& EXEC_APPLY_CHROOT
) {
2541 if (!needs_mount_ns
&& context
->root_directory
)
2542 if (chroot(context
->root_directory
) < 0) {
2543 *exit_status
= EXIT_CHROOT
;
2549 d
= prefix_roota(context
->root_directory
, wd
);
2551 if (chdir(d
) < 0 && !context
->working_directory_missing_ok
) {
2552 *exit_status
= EXIT_CHDIR
;
2559 static int setup_keyring(
2561 const ExecContext
*context
,
2562 const ExecParameters
*p
,
2563 uid_t uid
, gid_t gid
) {
2565 key_serial_t keyring
;
2574 /* Let's set up a new per-service "session" kernel keyring for each system service. This has the benefit that
2575 * each service runs with its own keyring shared among all processes of the service, but with no hook-up beyond
2576 * that scope, and in particular no link to the per-UID keyring. If we don't do this the keyring will be
2577 * automatically created on-demand and then linked to the per-UID keyring, by the kernel. The kernel's built-in
2578 * on-demand behaviour is very appropriate for login users, but probably not so much for system services, where
2579 * UIDs are not necessarily specific to a service but reused (at least in the case of UID 0). */
2581 if (context
->keyring_mode
== EXEC_KEYRING_INHERIT
)
2584 /* Acquiring a reference to the user keyring is nasty. We briefly change identity in order to get things set up
2585 * properly by the kernel. If we don't do that then we can't create it atomically, and that sucks for parallel
2586 * execution. This mimics what pam_keyinit does, too. Setting up session keyring, to be owned by the right user
2587 * & group is just as nasty as acquiring a reference to the user keyring. */
2589 saved_uid
= getuid();
2590 saved_gid
= getgid();
2592 if (gid_is_valid(gid
) && gid
!= saved_gid
) {
2593 if (setregid(gid
, -1) < 0)
2594 return log_unit_error_errno(u
, errno
, "Failed to change GID for user keyring: %m");
2597 if (uid_is_valid(uid
) && uid
!= saved_uid
) {
2598 if (setreuid(uid
, -1) < 0) {
2599 r
= log_unit_error_errno(u
, errno
, "Failed to change UID for user keyring: %m");
2604 keyring
= keyctl(KEYCTL_JOIN_SESSION_KEYRING
, 0, 0, 0, 0);
2605 if (keyring
== -1) {
2606 if (errno
== ENOSYS
)
2607 log_unit_debug_errno(u
, errno
, "Kernel keyring not supported, ignoring.");
2608 else if (IN_SET(errno
, EACCES
, EPERM
))
2609 log_unit_debug_errno(u
, errno
, "Kernel keyring access prohibited, ignoring.");
2610 else if (errno
== EDQUOT
)
2611 log_unit_debug_errno(u
, errno
, "Out of kernel keyrings to allocate, ignoring.");
2613 r
= log_unit_error_errno(u
, errno
, "Setting up kernel keyring failed: %m");
2618 /* When requested link the user keyring into the session keyring. */
2619 if (context
->keyring_mode
== EXEC_KEYRING_SHARED
) {
2621 if (keyctl(KEYCTL_LINK
,
2622 KEY_SPEC_USER_KEYRING
,
2623 KEY_SPEC_SESSION_KEYRING
, 0, 0) < 0) {
2624 r
= log_unit_error_errno(u
, errno
, "Failed to link user keyring into session keyring: %m");
2629 /* Restore uid/gid back */
2630 if (uid_is_valid(uid
) && uid
!= saved_uid
) {
2631 if (setreuid(saved_uid
, -1) < 0) {
2632 r
= log_unit_error_errno(u
, errno
, "Failed to change UID back for user keyring: %m");
2637 if (gid_is_valid(gid
) && gid
!= saved_gid
) {
2638 if (setregid(saved_gid
, -1) < 0)
2639 return log_unit_error_errno(u
, errno
, "Failed to change GID back for user keyring: %m");
2642 /* Populate they keyring with the invocation ID by default, as original saved_uid. */
2643 if (!sd_id128_is_null(u
->invocation_id
)) {
2646 key
= add_key("user", "invocation_id", &u
->invocation_id
, sizeof(u
->invocation_id
), KEY_SPEC_SESSION_KEYRING
);
2648 log_unit_debug_errno(u
, errno
, "Failed to add invocation ID to keyring, ignoring: %m");
2650 if (keyctl(KEYCTL_SETPERM
, key
,
2651 KEY_POS_VIEW
|KEY_POS_READ
|KEY_POS_SEARCH
|
2652 KEY_USR_VIEW
|KEY_USR_READ
|KEY_USR_SEARCH
, 0, 0) < 0)
2653 r
= log_unit_error_errno(u
, errno
, "Failed to restrict invocation ID permission: %m");
2658 /* Revert back uid & gid for the the last time, and exit */
2659 /* no extra logging, as only the first already reported error matters */
2660 if (getuid() != saved_uid
)
2661 (void) setreuid(saved_uid
, -1);
2663 if (getgid() != saved_gid
)
2664 (void) setregid(saved_gid
, -1);
2669 static void append_socket_pair(int *array
, size_t *n
, const int pair
[static 2]) {
2677 array
[(*n
)++] = pair
[0];
2679 array
[(*n
)++] = pair
[1];
2682 static int close_remaining_fds(
2683 const ExecParameters
*params
,
2684 const ExecRuntime
*runtime
,
2685 const DynamicCreds
*dcreds
,
2689 int *fds
, size_t n_fds
) {
2691 size_t n_dont_close
= 0;
2692 int dont_close
[n_fds
+ 12];
2696 if (params
->stdin_fd
>= 0)
2697 dont_close
[n_dont_close
++] = params
->stdin_fd
;
2698 if (params
->stdout_fd
>= 0)
2699 dont_close
[n_dont_close
++] = params
->stdout_fd
;
2700 if (params
->stderr_fd
>= 0)
2701 dont_close
[n_dont_close
++] = params
->stderr_fd
;
2704 dont_close
[n_dont_close
++] = socket_fd
;
2706 dont_close
[n_dont_close
++] = exec_fd
;
2708 memcpy(dont_close
+ n_dont_close
, fds
, sizeof(int) * n_fds
);
2709 n_dont_close
+= n_fds
;
2713 append_socket_pair(dont_close
, &n_dont_close
, runtime
->netns_storage_socket
);
2717 append_socket_pair(dont_close
, &n_dont_close
, dcreds
->user
->storage_socket
);
2719 append_socket_pair(dont_close
, &n_dont_close
, dcreds
->group
->storage_socket
);
2722 if (user_lookup_fd
>= 0)
2723 dont_close
[n_dont_close
++] = user_lookup_fd
;
2725 return close_all_fds(dont_close
, n_dont_close
);
2728 static int send_user_lookup(
2736 /* Send the resolved UID/GID to PID 1 after we learnt it. We send a single datagram, containing the UID/GID
2737 * data as well as the unit name. Note that we suppress sending this if no user/group to resolve was
2740 if (user_lookup_fd
< 0)
2743 if (!uid_is_valid(uid
) && !gid_is_valid(gid
))
2746 if (writev(user_lookup_fd
,
2748 IOVEC_INIT(&uid
, sizeof(uid
)),
2749 IOVEC_INIT(&gid
, sizeof(gid
)),
2750 IOVEC_INIT_STRING(unit
->id
) }, 3) < 0)
2756 static int acquire_home(const ExecContext
*c
, uid_t uid
, const char** home
, char **buf
) {
2763 /* If WorkingDirectory=~ is set, try to acquire a usable home directory. */
2768 if (!c
->working_directory_home
)
2771 r
= get_home_dir(buf
);
2779 static int compile_suggested_paths(const ExecContext
*c
, const ExecParameters
*p
, char ***ret
) {
2780 _cleanup_strv_free_
char ** list
= NULL
;
2781 ExecDirectoryType t
;
2788 assert(c
->dynamic_user
);
2790 /* Compile a list of paths that it might make sense to read the owning UID from to use as initial candidate for
2791 * dynamic UID allocation, in order to save us from doing costly recursive chown()s of the special
2794 for (t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
2797 if (t
== EXEC_DIRECTORY_CONFIGURATION
)
2803 STRV_FOREACH(i
, c
->directories
[t
].paths
) {
2806 if (t
== EXEC_DIRECTORY_RUNTIME
)
2807 e
= strjoin(p
->prefix
[t
], "/", *i
);
2809 e
= strjoin(p
->prefix
[t
], "/private/", *i
);
2813 r
= strv_consume(&list
, e
);
2819 *ret
= TAKE_PTR(list
);
2824 static char *exec_command_line(char **argv
);
2826 static int exec_parameters_get_cgroup_path(const ExecParameters
*params
, char **ret
) {
2827 bool using_subcgroup
;
2833 if (!params
->cgroup_path
)
2836 /* If we are called for a unit where cgroup delegation is on, and the payload created its own populated
2837 * subcgroup (which we expect it to do, after all it asked for delegation), then we cannot place the control
2838 * processes started after the main unit's process in the unit's main cgroup because it is now an inner one,
2839 * and inner cgroups may not contain processes. Hence, if delegation is on, and this is a control process,
2840 * let's use ".control" as subcgroup instead. Note that we do so only for ExecStartPost=, ExecReload=,
2841 * ExecStop=, ExecStopPost=, i.e. for the commands where the main process is already forked. For ExecStartPre=
2842 * this is not necessary, the cgroup is still empty. We distinguish these cases with the EXEC_CONTROL_CGROUP
2843 * flag, which is only passed for the former statements, not for the latter. */
2845 using_subcgroup
= FLAGS_SET(params
->flags
, EXEC_CONTROL_CGROUP
|EXEC_CGROUP_DELEGATE
|EXEC_IS_CONTROL
);
2846 if (using_subcgroup
)
2847 p
= strjoin(params
->cgroup_path
, "/.control");
2849 p
= strdup(params
->cgroup_path
);
2854 return using_subcgroup
;
2857 static int exec_child(
2859 const ExecCommand
*command
,
2860 const ExecContext
*context
,
2861 const ExecParameters
*params
,
2862 ExecRuntime
*runtime
,
2863 DynamicCreds
*dcreds
,
2867 size_t n_socket_fds
,
2868 size_t n_storage_fds
,
2873 _cleanup_strv_free_
char **our_env
= NULL
, **pass_env
= NULL
, **accum_env
= NULL
, **replaced_argv
= NULL
;
2874 int *fds_with_exec_fd
, n_fds_with_exec_fd
, r
, ngids
= 0, exec_fd
= -1;
2875 _cleanup_free_ gid_t
*supplementary_gids
= NULL
;
2876 const char *username
= NULL
, *groupname
= NULL
;
2877 _cleanup_free_
char *home_buffer
= NULL
;
2878 const char *home
= NULL
, *shell
= NULL
;
2879 char **final_argv
= NULL
;
2880 dev_t journal_stream_dev
= 0;
2881 ino_t journal_stream_ino
= 0;
2882 bool needs_sandboxing
, /* Do we need to set up full sandboxing? (i.e. all namespacing, all MAC stuff, caps, yadda yadda */
2883 needs_setuid
, /* Do we need to do the actual setresuid()/setresgid() calls? */
2884 needs_mount_namespace
, /* Do we need to set up a mount namespace for this kernel? */
2885 needs_ambient_hack
; /* Do we need to apply the ambient capabilities hack? */
2887 _cleanup_free_
char *mac_selinux_context_net
= NULL
;
2888 bool use_selinux
= false;
2891 bool use_smack
= false;
2894 bool use_apparmor
= false;
2896 uid_t uid
= UID_INVALID
;
2897 gid_t gid
= GID_INVALID
;
2899 ExecDirectoryType dt
;
2906 assert(exit_status
);
2908 rename_process_from_path(command
->path
);
2910 /* We reset exactly these signals, since they are the
2911 * only ones we set to SIG_IGN in the main daemon. All
2912 * others we leave untouched because we set them to
2913 * SIG_DFL or a valid handler initially, both of which
2914 * will be demoted to SIG_DFL. */
2915 (void) default_signals(SIGNALS_CRASH_HANDLER
,
2916 SIGNALS_IGNORE
, -1);
2918 if (context
->ignore_sigpipe
)
2919 (void) ignore_signals(SIGPIPE
, -1);
2921 r
= reset_signal_mask();
2923 *exit_status
= EXIT_SIGNAL_MASK
;
2924 return log_unit_error_errno(unit
, r
, "Failed to set process signal mask: %m");
2927 if (params
->idle_pipe
)
2928 do_idle_pipe_dance(params
->idle_pipe
);
2930 /* Close fds we don't need very early to make sure we don't block init reexecution because it cannot bind its
2931 * sockets. Among the fds we close are the logging fds, and we want to keep them closed, so that we don't have
2932 * any fds open we don't really want open during the transition. In order to make logging work, we switch the
2933 * log subsystem into open_when_needed mode, so that it reopens the logs on every single log call. */
2936 log_set_open_when_needed(true);
2938 /* In case anything used libc syslog(), close this here, too */
2941 n_fds
= n_socket_fds
+ n_storage_fds
;
2942 r
= close_remaining_fds(params
, runtime
, dcreds
, user_lookup_fd
, socket_fd
, params
->exec_fd
, fds
, n_fds
);
2944 *exit_status
= EXIT_FDS
;
2945 return log_unit_error_errno(unit
, r
, "Failed to close unwanted file descriptors: %m");
2948 if (!context
->same_pgrp
)
2950 *exit_status
= EXIT_SETSID
;
2951 return log_unit_error_errno(unit
, errno
, "Failed to create new process session: %m");
2954 exec_context_tty_reset(context
, params
);
2956 if (unit_shall_confirm_spawn(unit
)) {
2957 const char *vc
= params
->confirm_spawn
;
2958 _cleanup_free_
char *cmdline
= NULL
;
2960 cmdline
= exec_command_line(command
->argv
);
2962 *exit_status
= EXIT_MEMORY
;
2966 r
= ask_for_confirmation(vc
, unit
, cmdline
);
2967 if (r
!= CONFIRM_EXECUTE
) {
2968 if (r
== CONFIRM_PRETEND_SUCCESS
) {
2969 *exit_status
= EXIT_SUCCESS
;
2972 *exit_status
= EXIT_CONFIRM
;
2973 log_unit_error(unit
, "Execution cancelled by the user");
2978 /* We are about to invoke NSS and PAM modules. Let's tell them what we are doing here, maybe they care. This is
2979 * used by nss-resolve to disable itself when we are about to start systemd-resolved, to avoid deadlocks. Note
2980 * that these env vars do not survive the execve(), which means they really only apply to the PAM and NSS
2981 * invocations themselves. Also note that while we'll only invoke NSS modules involved in user management they
2982 * might internally call into other NSS modules that are involved in hostname resolution, we never know. */
2983 if (setenv("SYSTEMD_ACTIVATION_UNIT", unit
->id
, true) != 0 ||
2984 setenv("SYSTEMD_ACTIVATION_SCOPE", MANAGER_IS_SYSTEM(unit
->manager
) ? "system" : "user", true) != 0) {
2985 *exit_status
= EXIT_MEMORY
;
2986 return log_unit_error_errno(unit
, errno
, "Failed to update environment: %m");
2989 if (context
->dynamic_user
&& dcreds
) {
2990 _cleanup_strv_free_
char **suggested_paths
= NULL
;
2992 /* On top of that, make sure we bypass our own NSS module nss-systemd comprehensively for any NSS
2993 * checks, if DynamicUser=1 is used, as we shouldn't create a feedback loop with ourselves here.*/
2994 if (putenv((char*) "SYSTEMD_NSS_DYNAMIC_BYPASS=1") != 0) {
2995 *exit_status
= EXIT_USER
;
2996 return log_unit_error_errno(unit
, errno
, "Failed to update environment: %m");
2999 r
= compile_suggested_paths(context
, params
, &suggested_paths
);
3001 *exit_status
= EXIT_MEMORY
;
3005 r
= dynamic_creds_realize(dcreds
, suggested_paths
, &uid
, &gid
);
3007 *exit_status
= EXIT_USER
;
3009 log_unit_error(unit
, "Failed to update dynamic user credentials: User or group with specified name already exists.");
3012 return log_unit_error_errno(unit
, r
, "Failed to update dynamic user credentials: %m");
3015 if (!uid_is_valid(uid
)) {
3016 *exit_status
= EXIT_USER
;
3017 log_unit_error(unit
, "UID validation failed for \""UID_FMT
"\"", uid
);
3021 if (!gid_is_valid(gid
)) {
3022 *exit_status
= EXIT_USER
;
3023 log_unit_error(unit
, "GID validation failed for \""GID_FMT
"\"", gid
);
3028 username
= dcreds
->user
->name
;
3031 r
= get_fixed_user(context
, &username
, &uid
, &gid
, &home
, &shell
);
3033 *exit_status
= EXIT_USER
;
3034 return log_unit_error_errno(unit
, r
, "Failed to determine user credentials: %m");
3037 r
= get_fixed_group(context
, &groupname
, &gid
);
3039 *exit_status
= EXIT_GROUP
;
3040 return log_unit_error_errno(unit
, r
, "Failed to determine group credentials: %m");
3044 /* Initialize user supplementary groups and get SupplementaryGroups= ones */
3045 r
= get_supplementary_groups(context
, username
, groupname
, gid
,
3046 &supplementary_gids
, &ngids
);
3048 *exit_status
= EXIT_GROUP
;
3049 return log_unit_error_errno(unit
, r
, "Failed to determine supplementary groups: %m");
3052 r
= send_user_lookup(unit
, user_lookup_fd
, uid
, gid
);
3054 *exit_status
= EXIT_USER
;
3055 return log_unit_error_errno(unit
, r
, "Failed to send user credentials to PID1: %m");
3058 user_lookup_fd
= safe_close(user_lookup_fd
);
3060 r
= acquire_home(context
, uid
, &home
, &home_buffer
);
3062 *exit_status
= EXIT_CHDIR
;
3063 return log_unit_error_errno(unit
, r
, "Failed to determine $HOME for user: %m");
3066 /* If a socket is connected to STDIN/STDOUT/STDERR, we
3067 * must sure to drop O_NONBLOCK */
3069 (void) fd_nonblock(socket_fd
, false);
3071 /* Journald will try to look-up our cgroup in order to populate _SYSTEMD_CGROUP and _SYSTEMD_UNIT fields.
3072 * Hence we need to migrate to the target cgroup from init.scope before connecting to journald */
3073 if (params
->cgroup_path
) {
3074 _cleanup_free_
char *p
= NULL
;
3076 r
= exec_parameters_get_cgroup_path(params
, &p
);
3078 *exit_status
= EXIT_CGROUP
;
3079 return log_unit_error_errno(unit
, r
, "Failed to acquire cgroup path: %m");
3082 r
= cg_attach_everywhere(params
->cgroup_supported
, p
, 0, NULL
, NULL
);
3084 *exit_status
= EXIT_CGROUP
;
3085 return log_unit_error_errno(unit
, r
, "Failed to attach to cgroup %s: %m", p
);
3089 if (context
->network_namespace_path
&& runtime
&& runtime
->netns_storage_socket
[0] >= 0) {
3090 r
= open_netns_path(runtime
->netns_storage_socket
, context
->network_namespace_path
);
3092 *exit_status
= EXIT_NETWORK
;
3093 return log_unit_error_errno(unit
, r
, "Failed to open network namespace path %s: %m", context
->network_namespace_path
);
3097 r
= setup_input(context
, params
, socket_fd
, named_iofds
);
3099 *exit_status
= EXIT_STDIN
;
3100 return log_unit_error_errno(unit
, r
, "Failed to set up standard input: %m");
3103 r
= setup_output(unit
, context
, params
, STDOUT_FILENO
, socket_fd
, named_iofds
, basename(command
->path
), uid
, gid
, &journal_stream_dev
, &journal_stream_ino
);
3105 *exit_status
= EXIT_STDOUT
;
3106 return log_unit_error_errno(unit
, r
, "Failed to set up standard output: %m");
3109 r
= setup_output(unit
, context
, params
, STDERR_FILENO
, socket_fd
, named_iofds
, basename(command
->path
), uid
, gid
, &journal_stream_dev
, &journal_stream_ino
);
3111 *exit_status
= EXIT_STDERR
;
3112 return log_unit_error_errno(unit
, r
, "Failed to set up standard error output: %m");
3115 if (context
->oom_score_adjust_set
) {
3116 /* When we can't make this change due to EPERM, then let's silently skip over it. User namespaces
3117 * prohibit write access to this file, and we shouldn't trip up over that. */
3118 r
= set_oom_score_adjust(context
->oom_score_adjust
);
3119 if (IN_SET(r
, -EPERM
, -EACCES
))
3120 log_unit_debug_errno(unit
, r
, "Failed to adjust OOM setting, assuming containerized execution, ignoring: %m");
3122 *exit_status
= EXIT_OOM_ADJUST
;
3123 return log_unit_error_errno(unit
, r
, "Failed to adjust OOM setting: %m");
3127 if (context
->nice_set
)
3128 if (setpriority(PRIO_PROCESS
, 0, context
->nice
) < 0) {
3129 *exit_status
= EXIT_NICE
;
3130 return log_unit_error_errno(unit
, errno
, "Failed to set up process scheduling priority (nice level): %m");
3133 if (context
->cpu_sched_set
) {
3134 struct sched_param param
= {
3135 .sched_priority
= context
->cpu_sched_priority
,
3138 r
= sched_setscheduler(0,
3139 context
->cpu_sched_policy
|
3140 (context
->cpu_sched_reset_on_fork
?
3141 SCHED_RESET_ON_FORK
: 0),
3144 *exit_status
= EXIT_SETSCHEDULER
;
3145 return log_unit_error_errno(unit
, errno
, "Failed to set up CPU scheduling: %m");
3149 if (context
->cpuset
)
3150 if (sched_setaffinity(0, CPU_ALLOC_SIZE(context
->cpuset_ncpus
), context
->cpuset
) < 0) {
3151 *exit_status
= EXIT_CPUAFFINITY
;
3152 return log_unit_error_errno(unit
, errno
, "Failed to set up CPU affinity: %m");
3155 if (context
->ioprio_set
)
3156 if (ioprio_set(IOPRIO_WHO_PROCESS
, 0, context
->ioprio
) < 0) {
3157 *exit_status
= EXIT_IOPRIO
;
3158 return log_unit_error_errno(unit
, errno
, "Failed to set up IO scheduling priority: %m");
3161 if (context
->timer_slack_nsec
!= NSEC_INFINITY
)
3162 if (prctl(PR_SET_TIMERSLACK
, context
->timer_slack_nsec
) < 0) {
3163 *exit_status
= EXIT_TIMERSLACK
;
3164 return log_unit_error_errno(unit
, errno
, "Failed to set up timer slack: %m");
3167 if (context
->personality
!= PERSONALITY_INVALID
) {
3168 r
= safe_personality(context
->personality
);
3170 *exit_status
= EXIT_PERSONALITY
;
3171 return log_unit_error_errno(unit
, r
, "Failed to set up execution domain (personality): %m");
3175 if (context
->utmp_id
)
3176 utmp_put_init_process(context
->utmp_id
, getpid_cached(), getsid(0),
3178 context
->utmp_mode
== EXEC_UTMP_INIT
? INIT_PROCESS
:
3179 context
->utmp_mode
== EXEC_UTMP_LOGIN
? LOGIN_PROCESS
:
3183 if (uid_is_valid(uid
)) {
3184 r
= chown_terminal(STDIN_FILENO
, uid
);
3186 *exit_status
= EXIT_STDIN
;
3187 return log_unit_error_errno(unit
, r
, "Failed to change ownership of terminal: %m");
3191 /* If delegation is enabled we'll pass ownership of the cgroup to the user of the new process. On cgroup v1
3192 * this is only about systemd's own hierarchy, i.e. not the controller hierarchies, simply because that's not
3193 * safe. On cgroup v2 there's only one hierarchy anyway, and delegation is safe there, hence in that case only
3194 * touch a single hierarchy too. */
3195 if (params
->cgroup_path
&& context
->user
&& (params
->flags
& EXEC_CGROUP_DELEGATE
)) {
3196 r
= cg_set_access(SYSTEMD_CGROUP_CONTROLLER
, params
->cgroup_path
, uid
, gid
);
3198 *exit_status
= EXIT_CGROUP
;
3199 return log_unit_error_errno(unit
, r
, "Failed to adjust control group access: %m");
3203 for (dt
= 0; dt
< _EXEC_DIRECTORY_TYPE_MAX
; dt
++) {
3204 r
= setup_exec_directory(context
, params
, uid
, gid
, dt
, exit_status
);
3206 return log_unit_error_errno(unit
, r
, "Failed to set up special execution directory in %s: %m", params
->prefix
[dt
]);
3209 r
= build_environment(
3221 *exit_status
= EXIT_MEMORY
;
3225 r
= build_pass_environment(context
, &pass_env
);
3227 *exit_status
= EXIT_MEMORY
;
3231 accum_env
= strv_env_merge(5,
3232 params
->environment
,
3235 context
->environment
,
3239 *exit_status
= EXIT_MEMORY
;
3242 accum_env
= strv_env_clean(accum_env
);
3244 (void) umask(context
->umask
);
3246 r
= setup_keyring(unit
, context
, params
, uid
, gid
);
3248 *exit_status
= EXIT_KEYRING
;
3249 return log_unit_error_errno(unit
, r
, "Failed to set up kernel keyring: %m");
3252 /* We need sandboxing if the caller asked us to apply it and the command isn't explicitly excepted from it */
3253 needs_sandboxing
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && !(command
->flags
& EXEC_COMMAND_FULLY_PRIVILEGED
);
3255 /* 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 */
3256 needs_ambient_hack
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && (command
->flags
& EXEC_COMMAND_AMBIENT_MAGIC
) && !ambient_capabilities_supported();
3258 /* 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 */
3259 if (needs_ambient_hack
)
3260 needs_setuid
= false;
3262 needs_setuid
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && !(command
->flags
& (EXEC_COMMAND_FULLY_PRIVILEGED
|EXEC_COMMAND_NO_SETUID
));
3264 if (needs_sandboxing
) {
3265 /* MAC enablement checks need to be done before a new mount ns is created, as they rely on /sys being
3266 * present. The actual MAC context application will happen later, as late as possible, to avoid
3267 * impacting our own code paths. */
3270 use_selinux
= mac_selinux_use();
3273 use_smack
= mac_smack_use();
3276 use_apparmor
= mac_apparmor_use();
3280 if (needs_sandboxing
) {
3283 /* Let's set the resource limits before we call into PAM, so that pam_limits wins over what
3284 * is set here. (See below.) */
3286 r
= setrlimit_closest_all((const struct rlimit
* const *) context
->rlimit
, &which_failed
);
3288 *exit_status
= EXIT_LIMITS
;
3289 return log_unit_error_errno(unit
, r
, "Failed to adjust resource limit RLIMIT_%s: %m", rlimit_to_string(which_failed
));
3295 /* Let's call into PAM after we set up our own idea of resource limits to that pam_limits
3296 * wins here. (See above.) */
3298 if (context
->pam_name
&& username
) {
3299 r
= setup_pam(context
->pam_name
, username
, uid
, gid
, context
->tty_path
, &accum_env
, fds
, n_fds
);
3301 *exit_status
= EXIT_PAM
;
3302 return log_unit_error_errno(unit
, r
, "Failed to set up PAM session: %m");
3307 if ((context
->private_network
|| context
->network_namespace_path
) && runtime
&& runtime
->netns_storage_socket
[0] >= 0) {
3309 if (ns_type_supported(NAMESPACE_NET
)) {
3310 r
= setup_netns(runtime
->netns_storage_socket
);
3312 *exit_status
= EXIT_NETWORK
;
3313 return log_unit_error_errno(unit
, r
, "Failed to set up network namespacing: %m");
3315 } else if (context
->network_namespace_path
) {
3316 *exit_status
= EXIT_NETWORK
;
3317 return log_unit_error_errno(unit
, SYNTHETIC_ERRNO(EOPNOTSUPP
), "NetworkNamespacePath= is not supported, refusing.");
3319 log_unit_warning(unit
, "PrivateNetwork=yes is configured, but the kernel does not support network namespaces, ignoring.");
3322 needs_mount_namespace
= exec_needs_mount_namespace(context
, params
, runtime
);
3323 if (needs_mount_namespace
) {
3324 _cleanup_free_
char *error_path
= NULL
;
3326 r
= apply_mount_namespace(unit
, command
, context
, params
, runtime
, &error_path
);
3328 *exit_status
= EXIT_NAMESPACE
;
3329 return log_unit_error_errno(unit
, r
, "Failed to set up mount namespacing%s%s: %m",
3330 error_path
? ": " : "", strempty(error_path
));
3334 if (context
->protect_hostname
) {
3335 if (ns_type_supported(NAMESPACE_UTS
)) {
3336 if (unshare(CLONE_NEWUTS
) < 0) {
3337 *exit_status
= EXIT_NAMESPACE
;
3338 return log_unit_error_errno(unit
, errno
, "Failed to set up UTS namespacing: %m");
3341 log_unit_warning(unit
, "ProtectHostname=yes is configured, but the kernel does not support UTS namespaces, ignoring namespace setup.");
3343 r
= seccomp_protect_hostname();
3345 *exit_status
= EXIT_SECCOMP
;
3346 return log_unit_error_errno(unit
, r
, "Failed to apply hostname restrictions: %m");
3351 /* Drop groups as early as possbile */
3353 r
= enforce_groups(gid
, supplementary_gids
, ngids
);
3355 *exit_status
= EXIT_GROUP
;
3356 return log_unit_error_errno(unit
, r
, "Changing group credentials failed: %m");
3360 if (needs_sandboxing
) {
3362 if (use_selinux
&& params
->selinux_context_net
&& socket_fd
>= 0) {
3363 r
= mac_selinux_get_child_mls_label(socket_fd
, command
->path
, context
->selinux_context
, &mac_selinux_context_net
);
3365 *exit_status
= EXIT_SELINUX_CONTEXT
;
3366 return log_unit_error_errno(unit
, r
, "Failed to determine SELinux context: %m");
3371 if (context
->private_users
) {
3372 r
= setup_private_users(uid
, gid
);
3374 *exit_status
= EXIT_USER
;
3375 return log_unit_error_errno(unit
, r
, "Failed to set up user namespacing: %m");
3380 /* We repeat the fd closing here, to make sure that nothing is leaked from the PAM modules. Note that we are
3381 * more aggressive this time since socket_fd and the netns fds we don't need anymore. We do keep the exec_fd
3382 * however if we have it as we want to keep it open until the final execve(). */
3384 if (params
->exec_fd
>= 0) {
3385 exec_fd
= params
->exec_fd
;
3387 if (exec_fd
< 3 + (int) n_fds
) {
3390 /* Let's move the exec fd far up, so that it's outside of the fd range we want to pass to the
3391 * process we are about to execute. */
3393 moved_fd
= fcntl(exec_fd
, F_DUPFD_CLOEXEC
, 3 + (int) n_fds
);
3395 *exit_status
= EXIT_FDS
;
3396 return log_unit_error_errno(unit
, errno
, "Couldn't move exec fd up: %m");
3399 safe_close(exec_fd
);
3402 /* This fd should be FD_CLOEXEC already, but let's make sure. */
3403 r
= fd_cloexec(exec_fd
, true);
3405 *exit_status
= EXIT_FDS
;
3406 return log_unit_error_errno(unit
, r
, "Failed to make exec fd FD_CLOEXEC: %m");
3410 fds_with_exec_fd
= newa(int, n_fds
+ 1);
3411 memcpy_safe(fds_with_exec_fd
, fds
, n_fds
* sizeof(int));
3412 fds_with_exec_fd
[n_fds
] = exec_fd
;
3413 n_fds_with_exec_fd
= n_fds
+ 1;
3415 fds_with_exec_fd
= fds
;
3416 n_fds_with_exec_fd
= n_fds
;
3419 r
= close_all_fds(fds_with_exec_fd
, n_fds_with_exec_fd
);
3421 r
= shift_fds(fds
, n_fds
);
3423 r
= flags_fds(fds
, n_socket_fds
, n_storage_fds
, context
->non_blocking
);
3425 *exit_status
= EXIT_FDS
;
3426 return log_unit_error_errno(unit
, r
, "Failed to adjust passed file descriptors: %m");
3429 /* At this point, the fds we want to pass to the program are all ready and set up, with O_CLOEXEC turned off
3430 * and at the right fd numbers. The are no other fds open, with one exception: the exec_fd if it is defined,
3431 * and it has O_CLOEXEC set, after all we want it to be closed by the execve(), so that our parent knows we
3434 secure_bits
= context
->secure_bits
;
3436 if (needs_sandboxing
) {
3439 /* Set the RTPRIO resource limit to 0, but only if nothing else was explicitly
3440 * requested. (Note this is placed after the general resource limit initialization, see
3441 * above, in order to take precedence.) */
3442 if (context
->restrict_realtime
&& !context
->rlimit
[RLIMIT_RTPRIO
]) {
3443 if (setrlimit(RLIMIT_RTPRIO
, &RLIMIT_MAKE_CONST(0)) < 0) {
3444 *exit_status
= EXIT_LIMITS
;
3445 return log_unit_error_errno(unit
, errno
, "Failed to adjust RLIMIT_RTPRIO resource limit: %m");
3450 /* LSM Smack needs the capability CAP_MAC_ADMIN to change the current execution security context of the
3451 * process. This is the latest place before dropping capabilities. Other MAC context are set later. */
3453 r
= setup_smack(context
, command
);
3455 *exit_status
= EXIT_SMACK_PROCESS_LABEL
;
3456 return log_unit_error_errno(unit
, r
, "Failed to set SMACK process label: %m");
3461 bset
= context
->capability_bounding_set
;
3462 /* If the ambient caps hack is enabled (which means the kernel can't do them, and the user asked for
3463 * our magic fallback), then let's add some extra caps, so that the service can drop privs of its own,
3464 * instead of us doing that */
3465 if (needs_ambient_hack
)
3466 bset
|= (UINT64_C(1) << CAP_SETPCAP
) |
3467 (UINT64_C(1) << CAP_SETUID
) |
3468 (UINT64_C(1) << CAP_SETGID
);
3470 if (!cap_test_all(bset
)) {
3471 r
= capability_bounding_set_drop(bset
, false);
3473 *exit_status
= EXIT_CAPABILITIES
;
3474 return log_unit_error_errno(unit
, r
, "Failed to drop capabilities: %m");
3478 /* This is done before enforce_user, but ambient set
3479 * does not survive over setresuid() if keep_caps is not set. */
3480 if (!needs_ambient_hack
&&
3481 context
->capability_ambient_set
!= 0) {
3482 r
= capability_ambient_set_apply(context
->capability_ambient_set
, true);
3484 *exit_status
= EXIT_CAPABILITIES
;
3485 return log_unit_error_errno(unit
, r
, "Failed to apply ambient capabilities (before UID change): %m");
3491 if (uid_is_valid(uid
)) {
3492 r
= enforce_user(context
, uid
);
3494 *exit_status
= EXIT_USER
;
3495 return log_unit_error_errno(unit
, r
, "Failed to change UID to " UID_FMT
": %m", uid
);
3498 if (!needs_ambient_hack
&&
3499 context
->capability_ambient_set
!= 0) {
3501 /* Fix the ambient capabilities after user change. */
3502 r
= capability_ambient_set_apply(context
->capability_ambient_set
, false);
3504 *exit_status
= EXIT_CAPABILITIES
;
3505 return log_unit_error_errno(unit
, r
, "Failed to apply ambient capabilities (after UID change): %m");
3508 /* If we were asked to change user and ambient capabilities
3509 * were requested, we had to add keep-caps to the securebits
3510 * so that we would maintain the inherited capability set
3511 * through the setresuid(). Make sure that the bit is added
3512 * also to the context secure_bits so that we don't try to
3513 * drop the bit away next. */
3515 secure_bits
|= 1<<SECURE_KEEP_CAPS
;
3520 /* Apply working directory here, because the working directory might be on NFS and only the user running
3521 * this service might have the correct privilege to change to the working directory */
3522 r
= apply_working_directory(context
, params
, home
, needs_mount_namespace
, exit_status
);
3524 return log_unit_error_errno(unit
, r
, "Changing to the requested working directory failed: %m");
3526 if (needs_sandboxing
) {
3527 /* Apply other MAC contexts late, but before seccomp syscall filtering, as those should really be last to
3528 * influence our own codepaths as little as possible. Moreover, applying MAC contexts usually requires
3529 * syscalls that are subject to seccomp filtering, hence should probably be applied before the syscalls
3530 * are restricted. */
3534 char *exec_context
= mac_selinux_context_net
?: context
->selinux_context
;
3537 r
= setexeccon(exec_context
);
3539 *exit_status
= EXIT_SELINUX_CONTEXT
;
3540 return log_unit_error_errno(unit
, r
, "Failed to change SELinux context to %s: %m", exec_context
);
3547 if (use_apparmor
&& context
->apparmor_profile
) {
3548 r
= aa_change_onexec(context
->apparmor_profile
);
3549 if (r
< 0 && !context
->apparmor_profile_ignore
) {
3550 *exit_status
= EXIT_APPARMOR_PROFILE
;
3551 return log_unit_error_errno(unit
, errno
, "Failed to prepare AppArmor profile change to %s: %m", context
->apparmor_profile
);
3556 /* PR_GET_SECUREBITS is not privileged, while PR_SET_SECUREBITS is. So to suppress potential EPERMs
3557 * we'll try not to call PR_SET_SECUREBITS unless necessary. */
3558 if (prctl(PR_GET_SECUREBITS
) != secure_bits
)
3559 if (prctl(PR_SET_SECUREBITS
, secure_bits
) < 0) {
3560 *exit_status
= EXIT_SECUREBITS
;
3561 return log_unit_error_errno(unit
, errno
, "Failed to set process secure bits: %m");
3564 if (context_has_no_new_privileges(context
))
3565 if (prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0) < 0) {
3566 *exit_status
= EXIT_NO_NEW_PRIVILEGES
;
3567 return log_unit_error_errno(unit
, errno
, "Failed to disable new privileges: %m");
3571 r
= apply_address_families(unit
, context
);
3573 *exit_status
= EXIT_ADDRESS_FAMILIES
;
3574 return log_unit_error_errno(unit
, r
, "Failed to restrict address families: %m");
3577 r
= apply_memory_deny_write_execute(unit
, context
);
3579 *exit_status
= EXIT_SECCOMP
;
3580 return log_unit_error_errno(unit
, r
, "Failed to disable writing to executable memory: %m");
3583 r
= apply_restrict_realtime(unit
, context
);
3585 *exit_status
= EXIT_SECCOMP
;
3586 return log_unit_error_errno(unit
, r
, "Failed to apply realtime restrictions: %m");
3589 r
= apply_restrict_suid_sgid(unit
, context
);
3591 *exit_status
= EXIT_SECCOMP
;
3592 return log_unit_error_errno(unit
, r
, "Failed to apply SUID/SGID restrictions: %m");
3595 r
= apply_restrict_namespaces(unit
, context
);
3597 *exit_status
= EXIT_SECCOMP
;
3598 return log_unit_error_errno(unit
, r
, "Failed to apply namespace restrictions: %m");
3601 r
= apply_protect_sysctl(unit
, context
);
3603 *exit_status
= EXIT_SECCOMP
;
3604 return log_unit_error_errno(unit
, r
, "Failed to apply sysctl restrictions: %m");
3607 r
= apply_protect_kernel_modules(unit
, context
);
3609 *exit_status
= EXIT_SECCOMP
;
3610 return log_unit_error_errno(unit
, r
, "Failed to apply module loading restrictions: %m");
3613 r
= apply_private_devices(unit
, context
);
3615 *exit_status
= EXIT_SECCOMP
;
3616 return log_unit_error_errno(unit
, r
, "Failed to set up private devices: %m");
3619 r
= apply_syscall_archs(unit
, context
);
3621 *exit_status
= EXIT_SECCOMP
;
3622 return log_unit_error_errno(unit
, r
, "Failed to apply syscall architecture restrictions: %m");
3625 r
= apply_lock_personality(unit
, context
);
3627 *exit_status
= EXIT_SECCOMP
;
3628 return log_unit_error_errno(unit
, r
, "Failed to lock personalities: %m");
3631 /* This really should remain the last step before the execve(), to make sure our own code is unaffected
3632 * by the filter as little as possible. */
3633 r
= apply_syscall_filter(unit
, context
, needs_ambient_hack
);
3635 *exit_status
= EXIT_SECCOMP
;
3636 return log_unit_error_errno(unit
, r
, "Failed to apply system call filters: %m");
3641 if (!strv_isempty(context
->unset_environment
)) {
3644 ee
= strv_env_delete(accum_env
, 1, context
->unset_environment
);
3646 *exit_status
= EXIT_MEMORY
;
3650 strv_free_and_replace(accum_env
, ee
);
3653 if (!FLAGS_SET(command
->flags
, EXEC_COMMAND_NO_ENV_EXPAND
)) {
3654 replaced_argv
= replace_env_argv(command
->argv
, accum_env
);
3655 if (!replaced_argv
) {
3656 *exit_status
= EXIT_MEMORY
;
3659 final_argv
= replaced_argv
;
3661 final_argv
= command
->argv
;
3663 if (DEBUG_LOGGING
) {
3664 _cleanup_free_
char *line
;
3666 line
= exec_command_line(final_argv
);
3668 log_struct(LOG_DEBUG
,
3669 "EXECUTABLE=%s", command
->path
,
3670 LOG_UNIT_MESSAGE(unit
, "Executing: %s", line
),
3672 LOG_UNIT_INVOCATION_ID(unit
));
3678 /* We have finished with all our initializations. Let's now let the manager know that. From this point
3679 * on, if the manager sees POLLHUP on the exec_fd, then execve() was successful. */
3681 if (write(exec_fd
, &hot
, sizeof(hot
)) < 0) {
3682 *exit_status
= EXIT_EXEC
;
3683 return log_unit_error_errno(unit
, errno
, "Failed to enable exec_fd: %m");
3687 execve(command
->path
, final_argv
, accum_env
);
3693 /* The execve() failed. This means the exec_fd is still open. Which means we need to tell the manager
3694 * that POLLHUP on it no longer means execve() succeeded. */
3696 if (write(exec_fd
, &hot
, sizeof(hot
)) < 0) {
3697 *exit_status
= EXIT_EXEC
;
3698 return log_unit_error_errno(unit
, errno
, "Failed to disable exec_fd: %m");
3702 if (r
== -ENOENT
&& (command
->flags
& EXEC_COMMAND_IGNORE_FAILURE
)) {
3703 log_struct_errno(LOG_INFO
, r
,
3704 "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR
,
3706 LOG_UNIT_INVOCATION_ID(unit
),
3707 LOG_UNIT_MESSAGE(unit
, "Executable %s missing, skipping: %m",
3709 "EXECUTABLE=%s", command
->path
);
3713 *exit_status
= EXIT_EXEC
;
3714 return log_unit_error_errno(unit
, r
, "Failed to execute command: %m");
3717 static int exec_context_load_environment(const Unit
*unit
, const ExecContext
*c
, char ***l
);
3718 static int exec_context_named_iofds(const ExecContext
*c
, const ExecParameters
*p
, int named_iofds
[3]);
3720 int exec_spawn(Unit
*unit
,
3721 ExecCommand
*command
,
3722 const ExecContext
*context
,
3723 const ExecParameters
*params
,
3724 ExecRuntime
*runtime
,
3725 DynamicCreds
*dcreds
,
3728 int socket_fd
, r
, named_iofds
[3] = { -1, -1, -1 }, *fds
= NULL
;
3729 _cleanup_free_
char *subcgroup_path
= NULL
;
3730 _cleanup_strv_free_
char **files_env
= NULL
;
3731 size_t n_storage_fds
= 0, n_socket_fds
= 0;
3732 _cleanup_free_
char *line
= NULL
;
3740 assert(params
->fds
|| (params
->n_socket_fds
+ params
->n_storage_fds
<= 0));
3742 if (context
->std_input
== EXEC_INPUT_SOCKET
||
3743 context
->std_output
== EXEC_OUTPUT_SOCKET
||
3744 context
->std_error
== EXEC_OUTPUT_SOCKET
) {
3746 if (params
->n_socket_fds
> 1) {
3747 log_unit_error(unit
, "Got more than one socket.");
3751 if (params
->n_socket_fds
== 0) {
3752 log_unit_error(unit
, "Got no socket.");
3756 socket_fd
= params
->fds
[0];
3760 n_socket_fds
= params
->n_socket_fds
;
3761 n_storage_fds
= params
->n_storage_fds
;
3764 r
= exec_context_named_iofds(context
, params
, named_iofds
);
3766 return log_unit_error_errno(unit
, r
, "Failed to load a named file descriptor: %m");
3768 r
= exec_context_load_environment(unit
, context
, &files_env
);
3770 return log_unit_error_errno(unit
, r
, "Failed to load environment files: %m");
3772 line
= exec_command_line(command
->argv
);
3776 log_struct(LOG_DEBUG
,
3777 LOG_UNIT_MESSAGE(unit
, "About to execute: %s", line
),
3778 "EXECUTABLE=%s", command
->path
,
3780 LOG_UNIT_INVOCATION_ID(unit
));
3782 if (params
->cgroup_path
) {
3783 r
= exec_parameters_get_cgroup_path(params
, &subcgroup_path
);
3785 return log_unit_error_errno(unit
, r
, "Failed to acquire subcgroup path: %m");
3786 if (r
> 0) { /* We are using a child cgroup */
3787 r
= cg_create(SYSTEMD_CGROUP_CONTROLLER
, subcgroup_path
);
3789 return log_unit_error_errno(unit
, r
, "Failed to create control group '%s': %m", subcgroup_path
);
3795 return log_unit_error_errno(unit
, errno
, "Failed to fork: %m");
3798 int exit_status
= EXIT_SUCCESS
;
3800 r
= exec_child(unit
,
3812 unit
->manager
->user_lookup_fds
[1],
3816 log_struct_errno(LOG_ERR
, r
,
3817 "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR
,
3819 LOG_UNIT_INVOCATION_ID(unit
),
3820 LOG_UNIT_MESSAGE(unit
, "Failed at step %s spawning %s: %m",
3821 exit_status_to_string(exit_status
, EXIT_STATUS_SYSTEMD
),
3823 "EXECUTABLE=%s", command
->path
);
3828 log_unit_debug(unit
, "Forked %s as "PID_FMT
, command
->path
, pid
);
3830 /* We add the new process to the cgroup both in the child (so that we can be sure that no user code is ever
3831 * executed outside of the cgroup) and in the parent (so that we can be sure that when we kill the cgroup the
3832 * process will be killed too). */
3834 (void) cg_attach(SYSTEMD_CGROUP_CONTROLLER
, subcgroup_path
, pid
);
3836 exec_status_start(&command
->exec_status
, pid
);
3842 void exec_context_init(ExecContext
*c
) {
3843 ExecDirectoryType i
;
3848 c
->ioprio
= IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE
, 0);
3849 c
->cpu_sched_policy
= SCHED_OTHER
;
3850 c
->syslog_priority
= LOG_DAEMON
|LOG_INFO
;
3851 c
->syslog_level_prefix
= true;
3852 c
->ignore_sigpipe
= true;
3853 c
->timer_slack_nsec
= NSEC_INFINITY
;
3854 c
->personality
= PERSONALITY_INVALID
;
3855 for (i
= 0; i
< _EXEC_DIRECTORY_TYPE_MAX
; i
++)
3856 c
->directories
[i
].mode
= 0755;
3857 c
->capability_bounding_set
= CAP_ALL
;
3858 assert_cc(NAMESPACE_FLAGS_INITIAL
!= NAMESPACE_FLAGS_ALL
);
3859 c
->restrict_namespaces
= NAMESPACE_FLAGS_INITIAL
;
3860 c
->log_level_max
= -1;
3863 void exec_context_done(ExecContext
*c
) {
3864 ExecDirectoryType i
;
3869 c
->environment
= strv_free(c
->environment
);
3870 c
->environment_files
= strv_free(c
->environment_files
);
3871 c
->pass_environment
= strv_free(c
->pass_environment
);
3872 c
->unset_environment
= strv_free(c
->unset_environment
);
3874 rlimit_free_all(c
->rlimit
);
3876 for (l
= 0; l
< 3; l
++) {
3877 c
->stdio_fdname
[l
] = mfree(c
->stdio_fdname
[l
]);
3878 c
->stdio_file
[l
] = mfree(c
->stdio_file
[l
]);
3881 c
->working_directory
= mfree(c
->working_directory
);
3882 c
->root_directory
= mfree(c
->root_directory
);
3883 c
->root_image
= mfree(c
->root_image
);
3884 c
->tty_path
= mfree(c
->tty_path
);
3885 c
->syslog_identifier
= mfree(c
->syslog_identifier
);
3886 c
->user
= mfree(c
->user
);
3887 c
->group
= mfree(c
->group
);
3889 c
->supplementary_groups
= strv_free(c
->supplementary_groups
);
3891 c
->pam_name
= mfree(c
->pam_name
);
3893 c
->read_only_paths
= strv_free(c
->read_only_paths
);
3894 c
->read_write_paths
= strv_free(c
->read_write_paths
);
3895 c
->inaccessible_paths
= strv_free(c
->inaccessible_paths
);
3897 bind_mount_free_many(c
->bind_mounts
, c
->n_bind_mounts
);
3898 c
->bind_mounts
= NULL
;
3899 c
->n_bind_mounts
= 0;
3900 temporary_filesystem_free_many(c
->temporary_filesystems
, c
->n_temporary_filesystems
);
3901 c
->temporary_filesystems
= NULL
;
3902 c
->n_temporary_filesystems
= 0;
3904 c
->cpuset
= cpu_set_mfree(c
->cpuset
);
3906 c
->utmp_id
= mfree(c
->utmp_id
);
3907 c
->selinux_context
= mfree(c
->selinux_context
);
3908 c
->apparmor_profile
= mfree(c
->apparmor_profile
);
3909 c
->smack_process_label
= mfree(c
->smack_process_label
);
3911 c
->syscall_filter
= hashmap_free(c
->syscall_filter
);
3912 c
->syscall_archs
= set_free(c
->syscall_archs
);
3913 c
->address_families
= set_free(c
->address_families
);
3915 for (i
= 0; i
< _EXEC_DIRECTORY_TYPE_MAX
; i
++)
3916 c
->directories
[i
].paths
= strv_free(c
->directories
[i
].paths
);
3918 c
->log_level_max
= -1;
3920 exec_context_free_log_extra_fields(c
);
3922 c
->log_rate_limit_interval_usec
= 0;
3923 c
->log_rate_limit_burst
= 0;
3925 c
->stdin_data
= mfree(c
->stdin_data
);
3926 c
->stdin_data_size
= 0;
3928 c
->network_namespace_path
= mfree(c
->network_namespace_path
);
3931 int exec_context_destroy_runtime_directory(const ExecContext
*c
, const char *runtime_prefix
) {
3936 if (!runtime_prefix
)
3939 STRV_FOREACH(i
, c
->directories
[EXEC_DIRECTORY_RUNTIME
].paths
) {
3940 _cleanup_free_
char *p
;
3942 p
= path_join(runtime_prefix
, *i
);
3946 /* We execute this synchronously, since we need to be sure this is gone when we start the
3948 (void) rm_rf(p
, REMOVE_ROOT
);
3954 static void exec_command_done(ExecCommand
*c
) {
3957 c
->path
= mfree(c
->path
);
3958 c
->argv
= strv_free(c
->argv
);
3961 void exec_command_done_array(ExecCommand
*c
, size_t n
) {
3964 for (i
= 0; i
< n
; i
++)
3965 exec_command_done(c
+i
);
3968 ExecCommand
* exec_command_free_list(ExecCommand
*c
) {
3972 LIST_REMOVE(command
, c
, i
);
3973 exec_command_done(i
);
3980 void exec_command_free_array(ExecCommand
**c
, size_t n
) {
3983 for (i
= 0; i
< n
; i
++)
3984 c
[i
] = exec_command_free_list(c
[i
]);
3987 void exec_command_reset_status_array(ExecCommand
*c
, size_t n
) {
3990 for (i
= 0; i
< n
; i
++)
3991 exec_status_reset(&c
[i
].exec_status
);
3994 void exec_command_reset_status_list_array(ExecCommand
**c
, size_t n
) {
3997 for (i
= 0; i
< n
; i
++) {
4000 LIST_FOREACH(command
, z
, c
[i
])
4001 exec_status_reset(&z
->exec_status
);
4005 typedef struct InvalidEnvInfo
{
4010 static void invalid_env(const char *p
, void *userdata
) {
4011 InvalidEnvInfo
*info
= userdata
;
4013 log_unit_error(info
->unit
, "Ignoring invalid environment assignment '%s': %s", p
, info
->path
);
4016 const char* exec_context_fdname(const ExecContext
*c
, int fd_index
) {
4022 if (c
->std_input
!= EXEC_INPUT_NAMED_FD
)
4025 return c
->stdio_fdname
[STDIN_FILENO
] ?: "stdin";
4028 if (c
->std_output
!= EXEC_OUTPUT_NAMED_FD
)
4031 return c
->stdio_fdname
[STDOUT_FILENO
] ?: "stdout";
4034 if (c
->std_error
!= EXEC_OUTPUT_NAMED_FD
)
4037 return c
->stdio_fdname
[STDERR_FILENO
] ?: "stderr";
4044 static int exec_context_named_iofds(const ExecContext
*c
, const ExecParameters
*p
, int named_iofds
[static 3]) {
4046 const char* stdio_fdname
[3];
4052 targets
= (c
->std_input
== EXEC_INPUT_NAMED_FD
) +
4053 (c
->std_output
== EXEC_OUTPUT_NAMED_FD
) +
4054 (c
->std_error
== EXEC_OUTPUT_NAMED_FD
);
4056 for (i
= 0; i
< 3; i
++)
4057 stdio_fdname
[i
] = exec_context_fdname(c
, i
);
4059 n_fds
= p
->n_storage_fds
+ p
->n_socket_fds
;
4061 for (i
= 0; i
< n_fds
&& targets
> 0; i
++)
4062 if (named_iofds
[STDIN_FILENO
] < 0 &&
4063 c
->std_input
== EXEC_INPUT_NAMED_FD
&&
4064 stdio_fdname
[STDIN_FILENO
] &&
4065 streq(p
->fd_names
[i
], stdio_fdname
[STDIN_FILENO
])) {
4067 named_iofds
[STDIN_FILENO
] = p
->fds
[i
];
4070 } else if (named_iofds
[STDOUT_FILENO
] < 0 &&
4071 c
->std_output
== EXEC_OUTPUT_NAMED_FD
&&
4072 stdio_fdname
[STDOUT_FILENO
] &&
4073 streq(p
->fd_names
[i
], stdio_fdname
[STDOUT_FILENO
])) {
4075 named_iofds
[STDOUT_FILENO
] = p
->fds
[i
];
4078 } else if (named_iofds
[STDERR_FILENO
] < 0 &&
4079 c
->std_error
== EXEC_OUTPUT_NAMED_FD
&&
4080 stdio_fdname
[STDERR_FILENO
] &&
4081 streq(p
->fd_names
[i
], stdio_fdname
[STDERR_FILENO
])) {
4083 named_iofds
[STDERR_FILENO
] = p
->fds
[i
];
4087 return targets
== 0 ? 0 : -ENOENT
;
4090 static int exec_context_load_environment(const Unit
*unit
, const ExecContext
*c
, char ***l
) {
4091 char **i
, **r
= NULL
;
4096 STRV_FOREACH(i
, c
->environment_files
) {
4100 bool ignore
= false;
4102 _cleanup_globfree_ glob_t pglob
= {};
4111 if (!path_is_absolute(fn
)) {
4119 /* Filename supports globbing, take all matching files */
4120 k
= safe_glob(fn
, 0, &pglob
);
4129 /* When we don't match anything, -ENOENT should be returned */
4130 assert(pglob
.gl_pathc
> 0);
4132 for (n
= 0; n
< pglob
.gl_pathc
; n
++) {
4133 k
= load_env_file(NULL
, pglob
.gl_pathv
[n
], &p
);
4141 /* Log invalid environment variables with filename */
4143 InvalidEnvInfo info
= {
4145 .path
= pglob
.gl_pathv
[n
]
4148 p
= strv_env_clean_with_callback(p
, invalid_env
, &info
);
4156 m
= strv_env_merge(2, r
, p
);
4172 static bool tty_may_match_dev_console(const char *tty
) {
4173 _cleanup_free_
char *resolved
= NULL
;
4178 tty
= skip_dev_prefix(tty
);
4180 /* trivial identity? */
4181 if (streq(tty
, "console"))
4184 if (resolve_dev_console(&resolved
) < 0)
4185 return true; /* if we could not resolve, assume it may */
4187 /* "tty0" means the active VC, so it may be the same sometimes */
4188 return path_equal(resolved
, tty
) || (streq(resolved
, "tty0") && tty_is_vc(tty
));
4191 static bool exec_context_may_touch_tty(const ExecContext
*ec
) {
4194 return ec
->tty_reset
||
4196 ec
->tty_vt_disallocate
||
4197 is_terminal_input(ec
->std_input
) ||
4198 is_terminal_output(ec
->std_output
) ||
4199 is_terminal_output(ec
->std_error
);
4202 bool exec_context_may_touch_console(const ExecContext
*ec
) {
4204 return exec_context_may_touch_tty(ec
) &&
4205 tty_may_match_dev_console(exec_context_tty_path(ec
));
4208 static void strv_fprintf(FILE *f
, char **l
) {
4214 fprintf(f
, " %s", *g
);
4217 void exec_context_dump(const ExecContext
*c
, FILE* f
, const char *prefix
) {
4218 ExecDirectoryType dt
;
4226 prefix
= strempty(prefix
);
4230 "%sWorkingDirectory: %s\n"
4231 "%sRootDirectory: %s\n"
4232 "%sNonBlocking: %s\n"
4233 "%sPrivateTmp: %s\n"
4234 "%sPrivateDevices: %s\n"
4235 "%sProtectKernelTunables: %s\n"
4236 "%sProtectKernelModules: %s\n"
4237 "%sProtectControlGroups: %s\n"
4238 "%sPrivateNetwork: %s\n"
4239 "%sPrivateUsers: %s\n"
4240 "%sProtectHome: %s\n"
4241 "%sProtectSystem: %s\n"
4242 "%sMountAPIVFS: %s\n"
4243 "%sIgnoreSIGPIPE: %s\n"
4244 "%sMemoryDenyWriteExecute: %s\n"
4245 "%sRestrictRealtime: %s\n"
4246 "%sRestrictSUIDSGID: %s\n"
4247 "%sKeyringMode: %s\n"
4248 "%sProtectHostname: %s\n",
4250 prefix
, c
->working_directory
? c
->working_directory
: "/",
4251 prefix
, c
->root_directory
? c
->root_directory
: "/",
4252 prefix
, yes_no(c
->non_blocking
),
4253 prefix
, yes_no(c
->private_tmp
),
4254 prefix
, yes_no(c
->private_devices
),
4255 prefix
, yes_no(c
->protect_kernel_tunables
),
4256 prefix
, yes_no(c
->protect_kernel_modules
),
4257 prefix
, yes_no(c
->protect_control_groups
),
4258 prefix
, yes_no(c
->private_network
),
4259 prefix
, yes_no(c
->private_users
),
4260 prefix
, protect_home_to_string(c
->protect_home
),
4261 prefix
, protect_system_to_string(c
->protect_system
),
4262 prefix
, yes_no(c
->mount_apivfs
),
4263 prefix
, yes_no(c
->ignore_sigpipe
),
4264 prefix
, yes_no(c
->memory_deny_write_execute
),
4265 prefix
, yes_no(c
->restrict_realtime
),
4266 prefix
, yes_no(c
->restrict_suid_sgid
),
4267 prefix
, exec_keyring_mode_to_string(c
->keyring_mode
),
4268 prefix
, yes_no(c
->protect_hostname
));
4271 fprintf(f
, "%sRootImage: %s\n", prefix
, c
->root_image
);
4273 STRV_FOREACH(e
, c
->environment
)
4274 fprintf(f
, "%sEnvironment: %s\n", prefix
, *e
);
4276 STRV_FOREACH(e
, c
->environment_files
)
4277 fprintf(f
, "%sEnvironmentFile: %s\n", prefix
, *e
);
4279 STRV_FOREACH(e
, c
->pass_environment
)
4280 fprintf(f
, "%sPassEnvironment: %s\n", prefix
, *e
);
4282 STRV_FOREACH(e
, c
->unset_environment
)
4283 fprintf(f
, "%sUnsetEnvironment: %s\n", prefix
, *e
);
4285 fprintf(f
, "%sRuntimeDirectoryPreserve: %s\n", prefix
, exec_preserve_mode_to_string(c
->runtime_directory_preserve_mode
));
4287 for (dt
= 0; dt
< _EXEC_DIRECTORY_TYPE_MAX
; dt
++) {
4288 fprintf(f
, "%s%sMode: %04o\n", prefix
, exec_directory_type_to_string(dt
), c
->directories
[dt
].mode
);
4290 STRV_FOREACH(d
, c
->directories
[dt
].paths
)
4291 fprintf(f
, "%s%s: %s\n", prefix
, exec_directory_type_to_string(dt
), *d
);
4299 if (c
->oom_score_adjust_set
)
4301 "%sOOMScoreAdjust: %i\n",
4302 prefix
, c
->oom_score_adjust
);
4304 for (i
= 0; i
< RLIM_NLIMITS
; i
++)
4306 fprintf(f
, "%sLimit%s: " RLIM_FMT
"\n",
4307 prefix
, rlimit_to_string(i
), c
->rlimit
[i
]->rlim_max
);
4308 fprintf(f
, "%sLimit%sSoft: " RLIM_FMT
"\n",
4309 prefix
, rlimit_to_string(i
), c
->rlimit
[i
]->rlim_cur
);
4312 if (c
->ioprio_set
) {
4313 _cleanup_free_
char *class_str
= NULL
;
4315 r
= ioprio_class_to_string_alloc(IOPRIO_PRIO_CLASS(c
->ioprio
), &class_str
);
4317 fprintf(f
, "%sIOSchedulingClass: %s\n", prefix
, class_str
);
4319 fprintf(f
, "%sIOPriority: %lu\n", prefix
, IOPRIO_PRIO_DATA(c
->ioprio
));
4322 if (c
->cpu_sched_set
) {
4323 _cleanup_free_
char *policy_str
= NULL
;
4325 r
= sched_policy_to_string_alloc(c
->cpu_sched_policy
, &policy_str
);
4327 fprintf(f
, "%sCPUSchedulingPolicy: %s\n", prefix
, policy_str
);
4330 "%sCPUSchedulingPriority: %i\n"
4331 "%sCPUSchedulingResetOnFork: %s\n",
4332 prefix
, c
->cpu_sched_priority
,
4333 prefix
, yes_no(c
->cpu_sched_reset_on_fork
));
4337 fprintf(f
, "%sCPUAffinity:", prefix
);
4338 for (i
= 0; i
< c
->cpuset_ncpus
; i
++)
4339 if (CPU_ISSET_S(i
, CPU_ALLOC_SIZE(c
->cpuset_ncpus
), c
->cpuset
))
4340 fprintf(f
, " %u", i
);
4344 if (c
->timer_slack_nsec
!= NSEC_INFINITY
)
4345 fprintf(f
, "%sTimerSlackNSec: "NSEC_FMT
"\n", prefix
, c
->timer_slack_nsec
);
4348 "%sStandardInput: %s\n"
4349 "%sStandardOutput: %s\n"
4350 "%sStandardError: %s\n",
4351 prefix
, exec_input_to_string(c
->std_input
),
4352 prefix
, exec_output_to_string(c
->std_output
),
4353 prefix
, exec_output_to_string(c
->std_error
));
4355 if (c
->std_input
== EXEC_INPUT_NAMED_FD
)
4356 fprintf(f
, "%sStandardInputFileDescriptorName: %s\n", prefix
, c
->stdio_fdname
[STDIN_FILENO
]);
4357 if (c
->std_output
== EXEC_OUTPUT_NAMED_FD
)
4358 fprintf(f
, "%sStandardOutputFileDescriptorName: %s\n", prefix
, c
->stdio_fdname
[STDOUT_FILENO
]);
4359 if (c
->std_error
== EXEC_OUTPUT_NAMED_FD
)
4360 fprintf(f
, "%sStandardErrorFileDescriptorName: %s\n", prefix
, c
->stdio_fdname
[STDERR_FILENO
]);
4362 if (c
->std_input
== EXEC_INPUT_FILE
)
4363 fprintf(f
, "%sStandardInputFile: %s\n", prefix
, c
->stdio_file
[STDIN_FILENO
]);
4364 if (c
->std_output
== EXEC_OUTPUT_FILE
)
4365 fprintf(f
, "%sStandardOutputFile: %s\n", prefix
, c
->stdio_file
[STDOUT_FILENO
]);
4366 if (c
->std_output
== EXEC_OUTPUT_FILE_APPEND
)
4367 fprintf(f
, "%sStandardOutputFileToAppend: %s\n", prefix
, c
->stdio_file
[STDOUT_FILENO
]);
4368 if (c
->std_error
== EXEC_OUTPUT_FILE
)
4369 fprintf(f
, "%sStandardErrorFile: %s\n", prefix
, c
->stdio_file
[STDERR_FILENO
]);
4370 if (c
->std_error
== EXEC_OUTPUT_FILE_APPEND
)
4371 fprintf(f
, "%sStandardErrorFileToAppend: %s\n", prefix
, c
->stdio_file
[STDERR_FILENO
]);
4377 "%sTTYVHangup: %s\n"
4378 "%sTTYVTDisallocate: %s\n",
4379 prefix
, c
->tty_path
,
4380 prefix
, yes_no(c
->tty_reset
),
4381 prefix
, yes_no(c
->tty_vhangup
),
4382 prefix
, yes_no(c
->tty_vt_disallocate
));
4384 if (IN_SET(c
->std_output
,
4387 EXEC_OUTPUT_JOURNAL
,
4388 EXEC_OUTPUT_SYSLOG_AND_CONSOLE
,
4389 EXEC_OUTPUT_KMSG_AND_CONSOLE
,
4390 EXEC_OUTPUT_JOURNAL_AND_CONSOLE
) ||
4391 IN_SET(c
->std_error
,
4394 EXEC_OUTPUT_JOURNAL
,
4395 EXEC_OUTPUT_SYSLOG_AND_CONSOLE
,
4396 EXEC_OUTPUT_KMSG_AND_CONSOLE
,
4397 EXEC_OUTPUT_JOURNAL_AND_CONSOLE
)) {
4399 _cleanup_free_
char *fac_str
= NULL
, *lvl_str
= NULL
;
4401 r
= log_facility_unshifted_to_string_alloc(c
->syslog_priority
>> 3, &fac_str
);
4403 fprintf(f
, "%sSyslogFacility: %s\n", prefix
, fac_str
);
4405 r
= log_level_to_string_alloc(LOG_PRI(c
->syslog_priority
), &lvl_str
);
4407 fprintf(f
, "%sSyslogLevel: %s\n", prefix
, lvl_str
);
4410 if (c
->log_level_max
>= 0) {
4411 _cleanup_free_
char *t
= NULL
;
4413 (void) log_level_to_string_alloc(c
->log_level_max
, &t
);
4415 fprintf(f
, "%sLogLevelMax: %s\n", prefix
, strna(t
));
4418 if (c
->log_rate_limit_interval_usec
> 0) {
4419 char buf_timespan
[FORMAT_TIMESPAN_MAX
];
4422 "%sLogRateLimitIntervalSec: %s\n",
4423 prefix
, format_timespan(buf_timespan
, sizeof(buf_timespan
), c
->log_rate_limit_interval_usec
, USEC_PER_SEC
));
4426 if (c
->log_rate_limit_burst
> 0)
4427 fprintf(f
, "%sLogRateLimitBurst: %u\n", prefix
, c
->log_rate_limit_burst
);
4429 if (c
->n_log_extra_fields
> 0) {
4432 for (j
= 0; j
< c
->n_log_extra_fields
; j
++) {
4433 fprintf(f
, "%sLogExtraFields: ", prefix
);
4434 fwrite(c
->log_extra_fields
[j
].iov_base
,
4435 1, c
->log_extra_fields
[j
].iov_len
,
4441 if (c
->secure_bits
) {
4442 _cleanup_free_
char *str
= NULL
;
4444 r
= secure_bits_to_string_alloc(c
->secure_bits
, &str
);
4446 fprintf(f
, "%sSecure Bits: %s\n", prefix
, str
);
4449 if (c
->capability_bounding_set
!= CAP_ALL
) {
4450 _cleanup_free_
char *str
= NULL
;
4452 r
= capability_set_to_string_alloc(c
->capability_bounding_set
, &str
);
4454 fprintf(f
, "%sCapabilityBoundingSet: %s\n", prefix
, str
);
4457 if (c
->capability_ambient_set
!= 0) {
4458 _cleanup_free_
char *str
= NULL
;
4460 r
= capability_set_to_string_alloc(c
->capability_ambient_set
, &str
);
4462 fprintf(f
, "%sAmbientCapabilities: %s\n", prefix
, str
);
4466 fprintf(f
, "%sUser: %s\n", prefix
, c
->user
);
4468 fprintf(f
, "%sGroup: %s\n", prefix
, c
->group
);
4470 fprintf(f
, "%sDynamicUser: %s\n", prefix
, yes_no(c
->dynamic_user
));
4472 if (!strv_isempty(c
->supplementary_groups
)) {
4473 fprintf(f
, "%sSupplementaryGroups:", prefix
);
4474 strv_fprintf(f
, c
->supplementary_groups
);
4479 fprintf(f
, "%sPAMName: %s\n", prefix
, c
->pam_name
);
4481 if (!strv_isempty(c
->read_write_paths
)) {
4482 fprintf(f
, "%sReadWritePaths:", prefix
);
4483 strv_fprintf(f
, c
->read_write_paths
);
4487 if (!strv_isempty(c
->read_only_paths
)) {
4488 fprintf(f
, "%sReadOnlyPaths:", prefix
);
4489 strv_fprintf(f
, c
->read_only_paths
);
4493 if (!strv_isempty(c
->inaccessible_paths
)) {
4494 fprintf(f
, "%sInaccessiblePaths:", prefix
);
4495 strv_fprintf(f
, c
->inaccessible_paths
);
4499 if (c
->n_bind_mounts
> 0)
4500 for (i
= 0; i
< c
->n_bind_mounts
; i
++)
4501 fprintf(f
, "%s%s: %s%s:%s:%s\n", prefix
,
4502 c
->bind_mounts
[i
].read_only
? "BindReadOnlyPaths" : "BindPaths",
4503 c
->bind_mounts
[i
].ignore_enoent
? "-": "",
4504 c
->bind_mounts
[i
].source
,
4505 c
->bind_mounts
[i
].destination
,
4506 c
->bind_mounts
[i
].recursive
? "rbind" : "norbind");
4508 if (c
->n_temporary_filesystems
> 0)
4509 for (i
= 0; i
< c
->n_temporary_filesystems
; i
++) {
4510 TemporaryFileSystem
*t
= c
->temporary_filesystems
+ i
;
4512 fprintf(f
, "%sTemporaryFileSystem: %s%s%s\n", prefix
,
4514 isempty(t
->options
) ? "" : ":",
4515 strempty(t
->options
));
4520 "%sUtmpIdentifier: %s\n",
4521 prefix
, c
->utmp_id
);
4523 if (c
->selinux_context
)
4525 "%sSELinuxContext: %s%s\n",
4526 prefix
, c
->selinux_context_ignore
? "-" : "", c
->selinux_context
);
4528 if (c
->apparmor_profile
)
4530 "%sAppArmorProfile: %s%s\n",
4531 prefix
, c
->apparmor_profile_ignore
? "-" : "", c
->apparmor_profile
);
4533 if (c
->smack_process_label
)
4535 "%sSmackProcessLabel: %s%s\n",
4536 prefix
, c
->smack_process_label_ignore
? "-" : "", c
->smack_process_label
);
4538 if (c
->personality
!= PERSONALITY_INVALID
)
4540 "%sPersonality: %s\n",
4541 prefix
, strna(personality_to_string(c
->personality
)));
4544 "%sLockPersonality: %s\n",
4545 prefix
, yes_no(c
->lock_personality
));
4547 if (c
->syscall_filter
) {
4555 "%sSystemCallFilter: ",
4558 if (!c
->syscall_whitelist
)
4562 HASHMAP_FOREACH_KEY(val
, id
, c
->syscall_filter
, j
) {
4563 _cleanup_free_
char *name
= NULL
;
4564 const char *errno_name
= NULL
;
4565 int num
= PTR_TO_INT(val
);
4572 name
= seccomp_syscall_resolve_num_arch(SCMP_ARCH_NATIVE
, PTR_TO_INT(id
) - 1);
4573 fputs(strna(name
), f
);
4576 errno_name
= errno_to_name(num
);
4578 fprintf(f
, ":%s", errno_name
);
4580 fprintf(f
, ":%d", num
);
4588 if (c
->syscall_archs
) {
4595 "%sSystemCallArchitectures:",
4599 SET_FOREACH(id
, c
->syscall_archs
, j
)
4600 fprintf(f
, " %s", strna(seccomp_arch_to_string(PTR_TO_UINT32(id
) - 1)));
4605 if (exec_context_restrict_namespaces_set(c
)) {
4606 _cleanup_free_
char *s
= NULL
;
4608 r
= namespace_flags_to_string(c
->restrict_namespaces
, &s
);
4610 fprintf(f
, "%sRestrictNamespaces: %s\n",
4614 if (c
->network_namespace_path
)
4616 "%sNetworkNamespacePath: %s\n",
4617 prefix
, c
->network_namespace_path
);
4619 if (c
->syscall_errno
> 0) {
4620 const char *errno_name
;
4622 fprintf(f
, "%sSystemCallErrorNumber: ", prefix
);
4624 errno_name
= errno_to_name(c
->syscall_errno
);
4626 fprintf(f
, "%s\n", errno_name
);
4628 fprintf(f
, "%d\n", c
->syscall_errno
);
4632 bool exec_context_maintains_privileges(const ExecContext
*c
) {
4635 /* Returns true if the process forked off would run under
4636 * an unchanged UID or as root. */
4641 if (streq(c
->user
, "root") || streq(c
->user
, "0"))
4647 int exec_context_get_effective_ioprio(const ExecContext
*c
) {
4655 p
= ioprio_get(IOPRIO_WHO_PROCESS
, 0);
4657 return IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE
, 4);
4662 void exec_context_free_log_extra_fields(ExecContext
*c
) {
4667 for (l
= 0; l
< c
->n_log_extra_fields
; l
++)
4668 free(c
->log_extra_fields
[l
].iov_base
);
4669 c
->log_extra_fields
= mfree(c
->log_extra_fields
);
4670 c
->n_log_extra_fields
= 0;
4673 void exec_context_revert_tty(ExecContext
*c
) {
4678 /* First, reset the TTY (possibly kicking everybody else from the TTY) */
4679 exec_context_tty_reset(c
, NULL
);
4681 /* And then undo what chown_terminal() did earlier. Note that we only do this if we have a path
4682 * configured. If the TTY was passed to us as file descriptor we assume the TTY is opened and managed
4683 * by whoever passed it to us and thus knows better when and how to chmod()/chown() it back. */
4685 if (exec_context_may_touch_tty(c
)) {
4688 path
= exec_context_tty_path(c
);
4690 r
= chmod_and_chown(path
, TTY_MODE
, 0, TTY_GID
);
4691 if (r
< 0 && r
!= -ENOENT
)
4692 log_warning_errno(r
, "Failed to reset TTY ownership/access mode of %s, ignoring: %m", path
);
4697 void exec_status_start(ExecStatus
*s
, pid_t pid
) {
4704 dual_timestamp_get(&s
->start_timestamp
);
4707 void exec_status_exit(ExecStatus
*s
, const ExecContext
*context
, pid_t pid
, int code
, int status
) {
4710 if (s
->pid
!= pid
) {
4716 dual_timestamp_get(&s
->exit_timestamp
);
4721 if (context
&& context
->utmp_id
)
4722 (void) utmp_put_dead_process(context
->utmp_id
, pid
, code
, status
);
4725 void exec_status_reset(ExecStatus
*s
) {
4728 *s
= (ExecStatus
) {};
4731 void exec_status_dump(const ExecStatus
*s
, FILE *f
, const char *prefix
) {
4732 char buf
[FORMAT_TIMESTAMP_MAX
];
4740 prefix
= strempty(prefix
);
4743 "%sPID: "PID_FMT
"\n",
4746 if (dual_timestamp_is_set(&s
->start_timestamp
))
4748 "%sStart Timestamp: %s\n",
4749 prefix
, format_timestamp(buf
, sizeof(buf
), s
->start_timestamp
.realtime
));
4751 if (dual_timestamp_is_set(&s
->exit_timestamp
))
4753 "%sExit Timestamp: %s\n"
4755 "%sExit Status: %i\n",
4756 prefix
, format_timestamp(buf
, sizeof(buf
), s
->exit_timestamp
.realtime
),
4757 prefix
, sigchld_code_to_string(s
->code
),
4761 static char *exec_command_line(char **argv
) {
4769 STRV_FOREACH(a
, argv
)
4777 STRV_FOREACH(a
, argv
) {
4784 if (strpbrk(*a
, WHITESPACE
)) {
4795 /* FIXME: this doesn't really handle arguments that have
4796 * spaces and ticks in them */
4801 static void exec_command_dump(ExecCommand
*c
, FILE *f
, const char *prefix
) {
4802 _cleanup_free_
char *cmd
= NULL
;
4803 const char *prefix2
;
4808 prefix
= strempty(prefix
);
4809 prefix2
= strjoina(prefix
, "\t");
4811 cmd
= exec_command_line(c
->argv
);
4813 "%sCommand Line: %s\n",
4814 prefix
, cmd
? cmd
: strerror(ENOMEM
));
4816 exec_status_dump(&c
->exec_status
, f
, prefix2
);
4819 void exec_command_dump_list(ExecCommand
*c
, FILE *f
, const char *prefix
) {
4822 prefix
= strempty(prefix
);
4824 LIST_FOREACH(command
, c
, c
)
4825 exec_command_dump(c
, f
, prefix
);
4828 void exec_command_append_list(ExecCommand
**l
, ExecCommand
*e
) {
4835 /* It's kind of important, that we keep the order here */
4836 LIST_FIND_TAIL(command
, *l
, end
);
4837 LIST_INSERT_AFTER(command
, *l
, end
, e
);
4842 int exec_command_set(ExecCommand
*c
, const char *path
, ...) {
4850 l
= strv_new_ap(path
, ap
);
4862 free_and_replace(c
->path
, p
);
4864 return strv_free_and_replace(c
->argv
, l
);
4867 int exec_command_append(ExecCommand
*c
, const char *path
, ...) {
4868 _cleanup_strv_free_
char **l
= NULL
;
4876 l
= strv_new_ap(path
, ap
);
4882 r
= strv_extend_strv(&c
->argv
, l
, false);
4889 static void *remove_tmpdir_thread(void *p
) {
4890 _cleanup_free_
char *path
= p
;
4892 (void) rm_rf(path
, REMOVE_ROOT
|REMOVE_PHYSICAL
);
4896 static ExecRuntime
* exec_runtime_free(ExecRuntime
*rt
, bool destroy
) {
4903 (void) hashmap_remove(rt
->manager
->exec_runtime_by_id
, rt
->id
);
4905 /* When destroy is true, then rm_rf tmp_dir and var_tmp_dir. */
4906 if (destroy
&& rt
->tmp_dir
) {
4907 log_debug("Spawning thread to nuke %s", rt
->tmp_dir
);
4909 r
= asynchronous_job(remove_tmpdir_thread
, rt
->tmp_dir
);
4911 log_warning_errno(r
, "Failed to nuke %s: %m", rt
->tmp_dir
);
4918 if (destroy
&& rt
->var_tmp_dir
) {
4919 log_debug("Spawning thread to nuke %s", rt
->var_tmp_dir
);
4921 r
= asynchronous_job(remove_tmpdir_thread
, rt
->var_tmp_dir
);
4923 log_warning_errno(r
, "Failed to nuke %s: %m", rt
->var_tmp_dir
);
4924 free(rt
->var_tmp_dir
);
4927 rt
->var_tmp_dir
= NULL
;
4930 rt
->id
= mfree(rt
->id
);
4931 rt
->tmp_dir
= mfree(rt
->tmp_dir
);
4932 rt
->var_tmp_dir
= mfree(rt
->var_tmp_dir
);
4933 safe_close_pair(rt
->netns_storage_socket
);
4937 static void exec_runtime_freep(ExecRuntime
**rt
) {
4938 (void) exec_runtime_free(*rt
, false);
4941 static int exec_runtime_allocate(ExecRuntime
**ret
) {
4946 n
= new(ExecRuntime
, 1);
4950 *n
= (ExecRuntime
) {
4951 .netns_storage_socket
= { -1, -1 },
4958 static int exec_runtime_add(
4961 const char *tmp_dir
,
4962 const char *var_tmp_dir
,
4963 const int netns_storage_socket
[2],
4964 ExecRuntime
**ret
) {
4966 _cleanup_(exec_runtime_freep
) ExecRuntime
*rt
= NULL
;
4972 r
= hashmap_ensure_allocated(&m
->exec_runtime_by_id
, &string_hash_ops
);
4976 r
= exec_runtime_allocate(&rt
);
4980 rt
->id
= strdup(id
);
4985 rt
->tmp_dir
= strdup(tmp_dir
);
4989 /* When tmp_dir is set, then we require var_tmp_dir is also set. */
4990 assert(var_tmp_dir
);
4991 rt
->var_tmp_dir
= strdup(var_tmp_dir
);
4992 if (!rt
->var_tmp_dir
)
4996 if (netns_storage_socket
) {
4997 rt
->netns_storage_socket
[0] = netns_storage_socket
[0];
4998 rt
->netns_storage_socket
[1] = netns_storage_socket
[1];
5001 r
= hashmap_put(m
->exec_runtime_by_id
, rt
->id
, rt
);
5010 /* do not remove created ExecRuntime object when the operation succeeds. */
5015 static int exec_runtime_make(Manager
*m
, const ExecContext
*c
, const char *id
, ExecRuntime
**ret
) {
5016 _cleanup_free_
char *tmp_dir
= NULL
, *var_tmp_dir
= NULL
;
5017 _cleanup_close_pair_
int netns_storage_socket
[2] = { -1, -1 };
5024 /* It is not necessary to create ExecRuntime object. */
5025 if (!c
->private_network
&& !c
->private_tmp
&& !c
->network_namespace_path
)
5028 if (c
->private_tmp
) {
5029 r
= setup_tmp_dirs(id
, &tmp_dir
, &var_tmp_dir
);
5034 if (c
->private_network
|| c
->network_namespace_path
) {
5035 if (socketpair(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
, 0, netns_storage_socket
) < 0)
5039 r
= exec_runtime_add(m
, id
, tmp_dir
, var_tmp_dir
, netns_storage_socket
, ret
);
5044 netns_storage_socket
[0] = netns_storage_socket
[1] = -1;
5048 int exec_runtime_acquire(Manager
*m
, const ExecContext
*c
, const char *id
, bool create
, ExecRuntime
**ret
) {
5056 rt
= hashmap_get(m
->exec_runtime_by_id
, id
);
5058 /* We already have a ExecRuntime object, let's increase the ref count and reuse it */
5064 /* If not found, then create a new object. */
5065 r
= exec_runtime_make(m
, c
, id
, &rt
);
5067 /* When r == 0, it is not necessary to create ExecRuntime object. */
5071 /* increment reference counter. */
5077 ExecRuntime
*exec_runtime_unref(ExecRuntime
*rt
, bool destroy
) {
5081 assert(rt
->n_ref
> 0);
5087 return exec_runtime_free(rt
, destroy
);
5090 int exec_runtime_serialize(const Manager
*m
, FILE *f
, FDSet
*fds
) {
5098 HASHMAP_FOREACH(rt
, m
->exec_runtime_by_id
, i
) {
5099 fprintf(f
, "exec-runtime=%s", rt
->id
);
5102 fprintf(f
, " tmp-dir=%s", rt
->tmp_dir
);
5104 if (rt
->var_tmp_dir
)
5105 fprintf(f
, " var-tmp-dir=%s", rt
->var_tmp_dir
);
5107 if (rt
->netns_storage_socket
[0] >= 0) {
5110 copy
= fdset_put_dup(fds
, rt
->netns_storage_socket
[0]);
5114 fprintf(f
, " netns-socket-0=%i", copy
);
5117 if (rt
->netns_storage_socket
[1] >= 0) {
5120 copy
= fdset_put_dup(fds
, rt
->netns_storage_socket
[1]);
5124 fprintf(f
, " netns-socket-1=%i", copy
);
5133 int exec_runtime_deserialize_compat(Unit
*u
, const char *key
, const char *value
, FDSet
*fds
) {
5134 _cleanup_(exec_runtime_freep
) ExecRuntime
*rt_create
= NULL
;
5138 /* This is for the migration from old (v237 or earlier) deserialization text.
5139 * Due to the bug #7790, this may not work with the units that use JoinsNamespaceOf=.
5140 * Even if the ExecRuntime object originally created by the other unit, we cannot judge
5141 * so or not from the serialized text, then we always creates a new object owned by this. */
5147 /* Manager manages ExecRuntime objects by the unit id.
5148 * So, we omit the serialized text when the unit does not have id (yet?)... */
5149 if (isempty(u
->id
)) {
5150 log_unit_debug(u
, "Invocation ID not found. Dropping runtime parameter.");
5154 r
= hashmap_ensure_allocated(&u
->manager
->exec_runtime_by_id
, &string_hash_ops
);
5156 log_unit_debug_errno(u
, r
, "Failed to allocate storage for runtime parameter: %m");
5160 rt
= hashmap_get(u
->manager
->exec_runtime_by_id
, u
->id
);
5162 r
= exec_runtime_allocate(&rt_create
);
5166 rt_create
->id
= strdup(u
->id
);
5173 if (streq(key
, "tmp-dir")) {
5176 copy
= strdup(value
);
5180 free_and_replace(rt
->tmp_dir
, copy
);
5182 } else if (streq(key
, "var-tmp-dir")) {
5185 copy
= strdup(value
);
5189 free_and_replace(rt
->var_tmp_dir
, copy
);
5191 } else if (streq(key
, "netns-socket-0")) {
5194 if (safe_atoi(value
, &fd
) < 0 || !fdset_contains(fds
, fd
)) {
5195 log_unit_debug(u
, "Failed to parse netns socket value: %s", value
);
5199 safe_close(rt
->netns_storage_socket
[0]);
5200 rt
->netns_storage_socket
[0] = fdset_remove(fds
, fd
);
5202 } else if (streq(key
, "netns-socket-1")) {
5205 if (safe_atoi(value
, &fd
) < 0 || !fdset_contains(fds
, fd
)) {
5206 log_unit_debug(u
, "Failed to parse netns socket value: %s", value
);
5210 safe_close(rt
->netns_storage_socket
[1]);
5211 rt
->netns_storage_socket
[1] = fdset_remove(fds
, fd
);
5215 /* If the object is newly created, then put it to the hashmap which manages ExecRuntime objects. */
5217 r
= hashmap_put(u
->manager
->exec_runtime_by_id
, rt_create
->id
, rt_create
);
5219 log_unit_debug_errno(u
, r
, "Failed to put runtime parameter to manager's storage: %m");
5223 rt_create
->manager
= u
->manager
;
5232 void exec_runtime_deserialize_one(Manager
*m
, const char *value
, FDSet
*fds
) {
5233 char *id
= NULL
, *tmp_dir
= NULL
, *var_tmp_dir
= NULL
;
5234 int r
, fd0
= -1, fd1
= -1;
5235 const char *p
, *v
= value
;
5242 n
= strcspn(v
, " ");
5243 id
= strndupa(v
, n
);
5248 v
= startswith(p
, "tmp-dir=");
5250 n
= strcspn(v
, " ");
5251 tmp_dir
= strndupa(v
, n
);
5257 v
= startswith(p
, "var-tmp-dir=");
5259 n
= strcspn(v
, " ");
5260 var_tmp_dir
= strndupa(v
, n
);
5266 v
= startswith(p
, "netns-socket-0=");
5270 n
= strcspn(v
, " ");
5271 buf
= strndupa(v
, n
);
5272 if (safe_atoi(buf
, &fd0
) < 0 || !fdset_contains(fds
, fd0
)) {
5273 log_debug("Unable to process exec-runtime netns fd specification.");
5276 fd0
= fdset_remove(fds
, fd0
);
5282 v
= startswith(p
, "netns-socket-1=");
5286 n
= strcspn(v
, " ");
5287 buf
= strndupa(v
, n
);
5288 if (safe_atoi(buf
, &fd1
) < 0 || !fdset_contains(fds
, fd1
)) {
5289 log_debug("Unable to process exec-runtime netns fd specification.");
5292 fd1
= fdset_remove(fds
, fd1
);
5297 r
= exec_runtime_add(m
, id
, tmp_dir
, var_tmp_dir
, (int[]) { fd0
, fd1
}, NULL
);
5299 log_debug_errno(r
, "Failed to add exec-runtime: %m");
5302 void exec_runtime_vacuum(Manager
*m
) {
5308 /* Free unreferenced ExecRuntime objects. This is used after manager deserialization process. */
5310 HASHMAP_FOREACH(rt
, m
->exec_runtime_by_id
, i
) {
5314 (void) exec_runtime_free(rt
, false);
5318 void exec_params_clear(ExecParameters
*p
) {
5322 strv_free(p
->environment
);
5325 static const char* const exec_input_table
[_EXEC_INPUT_MAX
] = {
5326 [EXEC_INPUT_NULL
] = "null",
5327 [EXEC_INPUT_TTY
] = "tty",
5328 [EXEC_INPUT_TTY_FORCE
] = "tty-force",
5329 [EXEC_INPUT_TTY_FAIL
] = "tty-fail",
5330 [EXEC_INPUT_SOCKET
] = "socket",
5331 [EXEC_INPUT_NAMED_FD
] = "fd",
5332 [EXEC_INPUT_DATA
] = "data",
5333 [EXEC_INPUT_FILE
] = "file",
5336 DEFINE_STRING_TABLE_LOOKUP(exec_input
, ExecInput
);
5338 static const char* const exec_output_table
[_EXEC_OUTPUT_MAX
] = {
5339 [EXEC_OUTPUT_INHERIT
] = "inherit",
5340 [EXEC_OUTPUT_NULL
] = "null",
5341 [EXEC_OUTPUT_TTY
] = "tty",
5342 [EXEC_OUTPUT_SYSLOG
] = "syslog",
5343 [EXEC_OUTPUT_SYSLOG_AND_CONSOLE
] = "syslog+console",
5344 [EXEC_OUTPUT_KMSG
] = "kmsg",
5345 [EXEC_OUTPUT_KMSG_AND_CONSOLE
] = "kmsg+console",
5346 [EXEC_OUTPUT_JOURNAL
] = "journal",
5347 [EXEC_OUTPUT_JOURNAL_AND_CONSOLE
] = "journal+console",
5348 [EXEC_OUTPUT_SOCKET
] = "socket",
5349 [EXEC_OUTPUT_NAMED_FD
] = "fd",
5350 [EXEC_OUTPUT_FILE
] = "file",
5351 [EXEC_OUTPUT_FILE_APPEND
] = "append",
5354 DEFINE_STRING_TABLE_LOOKUP(exec_output
, ExecOutput
);
5356 static const char* const exec_utmp_mode_table
[_EXEC_UTMP_MODE_MAX
] = {
5357 [EXEC_UTMP_INIT
] = "init",
5358 [EXEC_UTMP_LOGIN
] = "login",
5359 [EXEC_UTMP_USER
] = "user",
5362 DEFINE_STRING_TABLE_LOOKUP(exec_utmp_mode
, ExecUtmpMode
);
5364 static const char* const exec_preserve_mode_table
[_EXEC_PRESERVE_MODE_MAX
] = {
5365 [EXEC_PRESERVE_NO
] = "no",
5366 [EXEC_PRESERVE_YES
] = "yes",
5367 [EXEC_PRESERVE_RESTART
] = "restart",
5370 DEFINE_STRING_TABLE_LOOKUP_WITH_BOOLEAN(exec_preserve_mode
, ExecPreserveMode
, EXEC_PRESERVE_YES
);
5372 static const char* const exec_directory_type_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
5373 [EXEC_DIRECTORY_RUNTIME
] = "RuntimeDirectory",
5374 [EXEC_DIRECTORY_STATE
] = "StateDirectory",
5375 [EXEC_DIRECTORY_CACHE
] = "CacheDirectory",
5376 [EXEC_DIRECTORY_LOGS
] = "LogsDirectory",
5377 [EXEC_DIRECTORY_CONFIGURATION
] = "ConfigurationDirectory",
5380 DEFINE_STRING_TABLE_LOOKUP(exec_directory_type
, ExecDirectoryType
);
5382 static const char* const exec_directory_env_name_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
5383 [EXEC_DIRECTORY_RUNTIME
] = "RUNTIME_DIRECTORY",
5384 [EXEC_DIRECTORY_STATE
] = "STATE_DIRECTORY",
5385 [EXEC_DIRECTORY_CACHE
] = "CACHE_DIRECTORY",
5386 [EXEC_DIRECTORY_LOGS
] = "LOGS_DIRECTORY",
5387 [EXEC_DIRECTORY_CONFIGURATION
] = "CONFIGURATION_DIRECTORY",
5390 DEFINE_PRIVATE_STRING_TABLE_LOOKUP_TO_STRING(exec_directory_env_name
, ExecDirectoryType
);
5392 static const char* const exec_keyring_mode_table
[_EXEC_KEYRING_MODE_MAX
] = {
5393 [EXEC_KEYRING_INHERIT
] = "inherit",
5394 [EXEC_KEYRING_PRIVATE
] = "private",
5395 [EXEC_KEYRING_SHARED
] = "shared",
5398 DEFINE_STRING_TABLE_LOOKUP(exec_keyring_mode
, ExecKeyringMode
);