1 /* SPDX-License-Identifier: LGPL-2.1-or-later */
6 #include <sys/eventfd.h>
10 #include <sys/personality.h>
11 #include <sys/prctl.h>
13 #include <sys/types.h>
19 #include <security/pam_appl.h>
23 #include <selinux/selinux.h>
31 #include <sys/apparmor.h>
34 #include "sd-messages.h"
38 #include "alloc-util.h"
40 #include "apparmor-util.h"
45 #include "capability-util.h"
46 #include "cgroup-setup.h"
47 #include "chown-recursive.h"
48 #include "cpu-set-util.h"
49 #include "creds-util.h"
50 #include "data-fd-util.h"
54 #include "errno-list.h"
57 #include "exit-status.h"
60 #include "format-util.h"
62 #include "glob-util.h"
63 #include "hexdecoct.h"
70 #include "manager-dump.h"
71 #include "memory-util.h"
72 #include "missing_fs.h"
74 #include "mount-util.h"
75 #include "mountpoint-util.h"
76 #include "namespace.h"
77 #include "parse-util.h"
78 #include "path-util.h"
79 #include "process-util.h"
80 #include "random-util.h"
81 #include "rlimit-util.h"
84 #include "seccomp-util.h"
86 #include "securebits-util.h"
87 #include "selinux-util.h"
88 #include "signal-util.h"
89 #include "smack-util.h"
90 #include "socket-util.h"
92 #include "stat-util.h"
93 #include "string-table.h"
94 #include "string-util.h"
96 #include "syslog-util.h"
97 #include "terminal-util.h"
98 #include "tmpfile-util.h"
99 #include "umask-util.h"
100 #include "unit-serialize.h"
101 #include "user-util.h"
102 #include "utmp-wtmp.h"
104 #define IDLE_TIMEOUT_USEC (5*USEC_PER_SEC)
105 #define IDLE_TIMEOUT2_USEC (1*USEC_PER_SEC)
107 #define SNDBUF_SIZE (8*1024*1024)
109 static int shift_fds(int fds
[], size_t n_fds
) {
113 /* Modifies the fds array! (sorts it) */
117 for (int start
= 0;;) {
118 int restart_from
= -1;
120 for (int i
= start
; i
< (int) n_fds
; i
++) {
123 /* Already at right index? */
127 nfd
= fcntl(fds
[i
], F_DUPFD
, i
+ 3);
134 /* Hmm, the fd we wanted isn't free? Then
135 * let's remember that and try again from here */
136 if (nfd
!= i
+3 && restart_from
< 0)
140 if (restart_from
< 0)
143 start
= restart_from
;
149 static int flags_fds(const int fds
[], size_t n_socket_fds
, size_t n_storage_fds
, bool nonblock
) {
153 n_fds
= n_socket_fds
+ n_storage_fds
;
159 /* Drops/Sets O_NONBLOCK and FD_CLOEXEC from the file flags.
160 * O_NONBLOCK only applies to socket activation though. */
162 for (size_t i
= 0; i
< n_fds
; i
++) {
164 if (i
< n_socket_fds
) {
165 r
= fd_nonblock(fds
[i
], nonblock
);
170 /* We unconditionally drop FD_CLOEXEC from the fds,
171 * since after all we want to pass these fds to our
174 r
= fd_cloexec(fds
[i
], false);
182 static const char *exec_context_tty_path(const ExecContext
*context
) {
185 if (context
->stdio_as_fds
)
188 if (context
->tty_path
)
189 return context
->tty_path
;
191 return "/dev/console";
194 static void exec_context_tty_reset(const ExecContext
*context
, const ExecParameters
*p
) {
199 path
= exec_context_tty_path(context
);
201 if (context
->tty_vhangup
) {
202 if (p
&& p
->stdin_fd
>= 0)
203 (void) terminal_vhangup_fd(p
->stdin_fd
);
205 (void) terminal_vhangup(path
);
208 if (context
->tty_reset
) {
209 if (p
&& p
->stdin_fd
>= 0)
210 (void) reset_terminal_fd(p
->stdin_fd
, true);
212 (void) reset_terminal(path
);
215 if (context
->tty_vt_disallocate
&& path
)
216 (void) vt_disallocate(path
);
219 static bool is_terminal_input(ExecInput i
) {
222 EXEC_INPUT_TTY_FORCE
,
223 EXEC_INPUT_TTY_FAIL
);
226 static bool is_terminal_output(ExecOutput o
) {
229 EXEC_OUTPUT_KMSG_AND_CONSOLE
,
230 EXEC_OUTPUT_JOURNAL_AND_CONSOLE
);
233 static bool is_kmsg_output(ExecOutput o
) {
236 EXEC_OUTPUT_KMSG_AND_CONSOLE
);
239 static bool exec_context_needs_term(const ExecContext
*c
) {
242 /* Return true if the execution context suggests we should set $TERM to something useful. */
244 if (is_terminal_input(c
->std_input
))
247 if (is_terminal_output(c
->std_output
))
250 if (is_terminal_output(c
->std_error
))
253 return !!c
->tty_path
;
256 static int open_null_as(int flags
, int nfd
) {
261 fd
= open("/dev/null", flags
|O_NOCTTY
);
265 return move_fd(fd
, nfd
, false);
268 static int connect_journal_socket(
270 const char *log_namespace
,
274 union sockaddr_union sa
;
276 uid_t olduid
= UID_INVALID
;
277 gid_t oldgid
= GID_INVALID
;
282 strjoina("/run/systemd/journal.", log_namespace
, "/stdout") :
283 "/run/systemd/journal/stdout";
284 r
= sockaddr_un_set_path(&sa
.un
, j
);
289 if (gid_is_valid(gid
)) {
292 if (setegid(gid
) < 0)
296 if (uid_is_valid(uid
)) {
299 if (seteuid(uid
) < 0) {
305 r
= connect(fd
, &sa
.sa
, sa_len
) < 0 ? -errno
: 0;
307 /* If we fail to restore the uid or gid, things will likely
308 fail later on. This should only happen if an LSM interferes. */
310 if (uid_is_valid(uid
))
311 (void) seteuid(olduid
);
314 if (gid_is_valid(gid
))
315 (void) setegid(oldgid
);
320 static int connect_logger_as(
322 const ExecContext
*context
,
323 const ExecParameters
*params
,
330 _cleanup_close_
int fd
= -1;
335 assert(output
< _EXEC_OUTPUT_MAX
);
339 fd
= socket(AF_UNIX
, SOCK_STREAM
, 0);
343 r
= connect_journal_socket(fd
, context
->log_namespace
, uid
, gid
);
347 if (shutdown(fd
, SHUT_RD
) < 0)
350 (void) fd_inc_sndbuf(fd
, SNDBUF_SIZE
);
360 context
->syslog_identifier
?: ident
,
361 params
->flags
& EXEC_PASS_LOG_UNIT
? unit
->id
: "",
362 context
->syslog_priority
,
363 !!context
->syslog_level_prefix
,
365 is_kmsg_output(output
),
366 is_terminal_output(output
)) < 0)
369 return move_fd(TAKE_FD(fd
), nfd
, false);
372 static int open_terminal_as(const char *path
, int flags
, int nfd
) {
378 fd
= open_terminal(path
, flags
| O_NOCTTY
);
382 return move_fd(fd
, nfd
, false);
385 static int acquire_path(const char *path
, int flags
, mode_t mode
) {
386 union sockaddr_union sa
;
388 _cleanup_close_
int fd
= -1;
393 if (IN_SET(flags
& O_ACCMODE
, O_WRONLY
, O_RDWR
))
396 fd
= open(path
, flags
|O_NOCTTY
, mode
);
400 if (errno
!= ENXIO
) /* ENXIO is returned when we try to open() an AF_UNIX file system socket on Linux */
403 /* So, it appears the specified path could be an AF_UNIX socket. Let's see if we can connect to it. */
405 r
= sockaddr_un_set_path(&sa
.un
, path
);
407 return r
== -EINVAL
? -ENXIO
: r
;
410 fd
= socket(AF_UNIX
, SOCK_STREAM
, 0);
414 if (connect(fd
, &sa
.sa
, sa_len
) < 0)
415 return errno
== EINVAL
? -ENXIO
: -errno
; /* Propagate initial error if we get EINVAL, i.e. we have
416 * indication that this wasn't an AF_UNIX socket after all */
418 if ((flags
& O_ACCMODE
) == O_RDONLY
)
419 r
= shutdown(fd
, SHUT_WR
);
420 else if ((flags
& O_ACCMODE
) == O_WRONLY
)
421 r
= shutdown(fd
, SHUT_RD
);
430 static int fixup_input(
431 const ExecContext
*context
,
433 bool apply_tty_stdin
) {
439 std_input
= context
->std_input
;
441 if (is_terminal_input(std_input
) && !apply_tty_stdin
)
442 return EXEC_INPUT_NULL
;
444 if (std_input
== EXEC_INPUT_SOCKET
&& socket_fd
< 0)
445 return EXEC_INPUT_NULL
;
447 if (std_input
== EXEC_INPUT_DATA
&& context
->stdin_data_size
== 0)
448 return EXEC_INPUT_NULL
;
453 static int fixup_output(ExecOutput output
, int socket_fd
) {
455 if (output
== EXEC_OUTPUT_SOCKET
&& socket_fd
< 0)
456 return EXEC_OUTPUT_INHERIT
;
461 static int setup_input(
462 const ExecContext
*context
,
463 const ExecParameters
*params
,
465 const int named_iofds
[static 3]) {
473 if (params
->stdin_fd
>= 0) {
474 if (dup2(params
->stdin_fd
, STDIN_FILENO
) < 0)
477 /* Try to make this the controlling tty, if it is a tty, and reset it */
478 if (isatty(STDIN_FILENO
)) {
479 (void) ioctl(STDIN_FILENO
, TIOCSCTTY
, context
->std_input
== EXEC_INPUT_TTY_FORCE
);
480 (void) reset_terminal_fd(STDIN_FILENO
, true);
486 i
= fixup_input(context
, socket_fd
, params
->flags
& EXEC_APPLY_TTY_STDIN
);
490 case EXEC_INPUT_NULL
:
491 return open_null_as(O_RDONLY
, STDIN_FILENO
);
494 case EXEC_INPUT_TTY_FORCE
:
495 case EXEC_INPUT_TTY_FAIL
: {
498 fd
= acquire_terminal(exec_context_tty_path(context
),
499 i
== EXEC_INPUT_TTY_FAIL
? ACQUIRE_TERMINAL_TRY
:
500 i
== EXEC_INPUT_TTY_FORCE
? ACQUIRE_TERMINAL_FORCE
:
501 ACQUIRE_TERMINAL_WAIT
,
506 return move_fd(fd
, STDIN_FILENO
, false);
509 case EXEC_INPUT_SOCKET
:
510 assert(socket_fd
>= 0);
512 return dup2(socket_fd
, STDIN_FILENO
) < 0 ? -errno
: STDIN_FILENO
;
514 case EXEC_INPUT_NAMED_FD
:
515 assert(named_iofds
[STDIN_FILENO
] >= 0);
517 (void) fd_nonblock(named_iofds
[STDIN_FILENO
], false);
518 return dup2(named_iofds
[STDIN_FILENO
], STDIN_FILENO
) < 0 ? -errno
: STDIN_FILENO
;
520 case EXEC_INPUT_DATA
: {
523 fd
= acquire_data_fd(context
->stdin_data
, context
->stdin_data_size
, 0);
527 return move_fd(fd
, STDIN_FILENO
, false);
530 case EXEC_INPUT_FILE
: {
534 assert(context
->stdio_file
[STDIN_FILENO
]);
536 rw
= (context
->std_output
== EXEC_OUTPUT_FILE
&& streq_ptr(context
->stdio_file
[STDIN_FILENO
], context
->stdio_file
[STDOUT_FILENO
])) ||
537 (context
->std_error
== EXEC_OUTPUT_FILE
&& streq_ptr(context
->stdio_file
[STDIN_FILENO
], context
->stdio_file
[STDERR_FILENO
]));
539 fd
= acquire_path(context
->stdio_file
[STDIN_FILENO
], rw
? O_RDWR
: O_RDONLY
, 0666 & ~context
->umask
);
543 return move_fd(fd
, STDIN_FILENO
, false);
547 assert_not_reached();
551 static bool can_inherit_stderr_from_stdout(
552 const ExecContext
*context
,
558 /* Returns true, if given the specified STDERR and STDOUT output we can directly dup() the stdout fd to the
561 if (e
== EXEC_OUTPUT_INHERIT
)
566 if (e
== EXEC_OUTPUT_NAMED_FD
)
567 return streq_ptr(context
->stdio_fdname
[STDOUT_FILENO
], context
->stdio_fdname
[STDERR_FILENO
]);
569 if (IN_SET(e
, EXEC_OUTPUT_FILE
, EXEC_OUTPUT_FILE_APPEND
, EXEC_OUTPUT_FILE_TRUNCATE
))
570 return streq_ptr(context
->stdio_file
[STDOUT_FILENO
], context
->stdio_file
[STDERR_FILENO
]);
575 static int setup_output(
577 const ExecContext
*context
,
578 const ExecParameters
*params
,
581 const int named_iofds
[static 3],
585 dev_t
*journal_stream_dev
,
586 ino_t
*journal_stream_ino
) {
596 assert(journal_stream_dev
);
597 assert(journal_stream_ino
);
599 if (fileno
== STDOUT_FILENO
&& params
->stdout_fd
>= 0) {
601 if (dup2(params
->stdout_fd
, STDOUT_FILENO
) < 0)
604 return STDOUT_FILENO
;
607 if (fileno
== STDERR_FILENO
&& params
->stderr_fd
>= 0) {
608 if (dup2(params
->stderr_fd
, STDERR_FILENO
) < 0)
611 return STDERR_FILENO
;
614 i
= fixup_input(context
, socket_fd
, params
->flags
& EXEC_APPLY_TTY_STDIN
);
615 o
= fixup_output(context
->std_output
, socket_fd
);
617 if (fileno
== STDERR_FILENO
) {
619 e
= fixup_output(context
->std_error
, socket_fd
);
621 /* This expects the input and output are already set up */
623 /* Don't change the stderr file descriptor if we inherit all
624 * the way and are not on a tty */
625 if (e
== EXEC_OUTPUT_INHERIT
&&
626 o
== EXEC_OUTPUT_INHERIT
&&
627 i
== EXEC_INPUT_NULL
&&
628 !is_terminal_input(context
->std_input
) &&
632 /* Duplicate from stdout if possible */
633 if (can_inherit_stderr_from_stdout(context
, o
, e
))
634 return dup2(STDOUT_FILENO
, fileno
) < 0 ? -errno
: fileno
;
638 } else if (o
== EXEC_OUTPUT_INHERIT
) {
639 /* If input got downgraded, inherit the original value */
640 if (i
== EXEC_INPUT_NULL
&& is_terminal_input(context
->std_input
))
641 return open_terminal_as(exec_context_tty_path(context
), O_WRONLY
, fileno
);
643 /* If the input is connected to anything that's not a /dev/null or a data fd, inherit that... */
644 if (!IN_SET(i
, EXEC_INPUT_NULL
, EXEC_INPUT_DATA
))
645 return dup2(STDIN_FILENO
, fileno
) < 0 ? -errno
: fileno
;
647 /* If we are not started from PID 1 we just inherit STDOUT from our parent process. */
651 /* We need to open /dev/null here anew, to get the right access mode. */
652 return open_null_as(O_WRONLY
, fileno
);
657 case EXEC_OUTPUT_NULL
:
658 return open_null_as(O_WRONLY
, fileno
);
660 case EXEC_OUTPUT_TTY
:
661 if (is_terminal_input(i
))
662 return dup2(STDIN_FILENO
, fileno
) < 0 ? -errno
: fileno
;
664 /* We don't reset the terminal if this is just about output */
665 return open_terminal_as(exec_context_tty_path(context
), O_WRONLY
, fileno
);
667 case EXEC_OUTPUT_KMSG
:
668 case EXEC_OUTPUT_KMSG_AND_CONSOLE
:
669 case EXEC_OUTPUT_JOURNAL
:
670 case EXEC_OUTPUT_JOURNAL_AND_CONSOLE
:
671 r
= connect_logger_as(unit
, context
, params
, o
, ident
, fileno
, uid
, gid
);
673 log_unit_warning_errno(unit
, r
, "Failed to connect %s to the journal socket, ignoring: %m",
674 fileno
== STDOUT_FILENO
? "stdout" : "stderr");
675 r
= open_null_as(O_WRONLY
, fileno
);
679 /* If we connected this fd to the journal via a stream, patch the device/inode into the passed
680 * parameters, but only then. This is useful so that we can set $JOURNAL_STREAM that permits
681 * services to detect whether they are connected to the journal or not.
683 * If both stdout and stderr are connected to a stream then let's make sure to store the data
684 * about STDERR as that's usually the best way to do logging. */
686 if (fstat(fileno
, &st
) >= 0 &&
687 (*journal_stream_ino
== 0 || fileno
== STDERR_FILENO
)) {
688 *journal_stream_dev
= st
.st_dev
;
689 *journal_stream_ino
= st
.st_ino
;
694 case EXEC_OUTPUT_SOCKET
:
695 assert(socket_fd
>= 0);
697 return dup2(socket_fd
, fileno
) < 0 ? -errno
: fileno
;
699 case EXEC_OUTPUT_NAMED_FD
:
700 assert(named_iofds
[fileno
] >= 0);
702 (void) fd_nonblock(named_iofds
[fileno
], false);
703 return dup2(named_iofds
[fileno
], fileno
) < 0 ? -errno
: fileno
;
705 case EXEC_OUTPUT_FILE
:
706 case EXEC_OUTPUT_FILE_APPEND
:
707 case EXEC_OUTPUT_FILE_TRUNCATE
: {
711 assert(context
->stdio_file
[fileno
]);
713 rw
= context
->std_input
== EXEC_INPUT_FILE
&&
714 streq_ptr(context
->stdio_file
[fileno
], context
->stdio_file
[STDIN_FILENO
]);
717 return dup2(STDIN_FILENO
, fileno
) < 0 ? -errno
: fileno
;
720 if (o
== EXEC_OUTPUT_FILE_APPEND
)
722 else if (o
== EXEC_OUTPUT_FILE_TRUNCATE
)
725 fd
= acquire_path(context
->stdio_file
[fileno
], flags
, 0666 & ~context
->umask
);
729 return move_fd(fd
, fileno
, 0);
733 assert_not_reached();
737 static int chown_terminal(int fd
, uid_t uid
) {
742 /* Before we chown/chmod the TTY, let's ensure this is actually a tty */
743 if (isatty(fd
) < 1) {
744 if (IN_SET(errno
, EINVAL
, ENOTTY
))
745 return 0; /* not a tty */
750 /* This might fail. What matters are the results. */
751 r
= fchmod_and_chown(fd
, TTY_MODE
, uid
, GID_INVALID
);
758 static int setup_confirm_stdio(const char *vc
, int *_saved_stdin
, int *_saved_stdout
) {
759 _cleanup_close_
int fd
= -1, saved_stdin
= -1, saved_stdout
= -1;
762 assert(_saved_stdin
);
763 assert(_saved_stdout
);
765 saved_stdin
= fcntl(STDIN_FILENO
, F_DUPFD
, 3);
769 saved_stdout
= fcntl(STDOUT_FILENO
, F_DUPFD
, 3);
770 if (saved_stdout
< 0)
773 fd
= acquire_terminal(vc
, ACQUIRE_TERMINAL_WAIT
, DEFAULT_CONFIRM_USEC
);
777 r
= chown_terminal(fd
, getuid());
781 r
= reset_terminal_fd(fd
, true);
785 r
= rearrange_stdio(fd
, fd
, STDERR_FILENO
);
790 *_saved_stdin
= saved_stdin
;
791 *_saved_stdout
= saved_stdout
;
793 saved_stdin
= saved_stdout
= -1;
798 static void write_confirm_error_fd(int err
, int fd
, const Unit
*u
) {
801 if (err
== -ETIMEDOUT
)
802 dprintf(fd
, "Confirmation question timed out for %s, assuming positive response.\n", u
->id
);
805 dprintf(fd
, "Couldn't ask confirmation for %s: %m, assuming positive response.\n", u
->id
);
809 static void write_confirm_error(int err
, const char *vc
, const Unit
*u
) {
810 _cleanup_close_
int fd
= -1;
814 fd
= open_terminal(vc
, O_WRONLY
|O_NOCTTY
|O_CLOEXEC
);
818 write_confirm_error_fd(err
, fd
, u
);
821 static int restore_confirm_stdio(int *saved_stdin
, int *saved_stdout
) {
825 assert(saved_stdout
);
829 if (*saved_stdin
>= 0)
830 if (dup2(*saved_stdin
, STDIN_FILENO
) < 0)
833 if (*saved_stdout
>= 0)
834 if (dup2(*saved_stdout
, STDOUT_FILENO
) < 0)
837 *saved_stdin
= safe_close(*saved_stdin
);
838 *saved_stdout
= safe_close(*saved_stdout
);
844 CONFIRM_PRETEND_FAILURE
= -1,
845 CONFIRM_PRETEND_SUCCESS
= 0,
849 static int ask_for_confirmation(const char *vc
, Unit
*u
, const char *cmdline
) {
850 int saved_stdout
= -1, saved_stdin
= -1, r
;
851 _cleanup_free_
char *e
= NULL
;
854 /* For any internal errors, assume a positive response. */
855 r
= setup_confirm_stdio(vc
, &saved_stdin
, &saved_stdout
);
857 write_confirm_error(r
, vc
, u
);
858 return CONFIRM_EXECUTE
;
861 /* confirm_spawn might have been disabled while we were sleeping. */
862 if (manager_is_confirm_spawn_disabled(u
->manager
)) {
867 e
= ellipsize(cmdline
, 60, 100);
875 r
= ask_char(&c
, "yfshiDjcn", "Execute %s? [y, f, s – h for help] ", e
);
877 write_confirm_error_fd(r
, STDOUT_FILENO
, u
);
884 printf("Resuming normal execution.\n");
885 manager_disable_confirm_spawn();
889 unit_dump(u
, stdout
, " ");
890 continue; /* ask again */
892 printf("Failing execution.\n");
893 r
= CONFIRM_PRETEND_FAILURE
;
896 printf(" c - continue, proceed without asking anymore\n"
897 " D - dump, show the state of the unit\n"
898 " f - fail, don't execute the command and pretend it failed\n"
900 " i - info, show a short summary of the unit\n"
901 " j - jobs, show jobs that are in progress\n"
902 " s - skip, don't execute the command and pretend it succeeded\n"
903 " y - yes, execute the command\n");
904 continue; /* ask again */
906 printf(" Description: %s\n"
909 u
->id
, u
->description
, cmdline
);
910 continue; /* ask again */
912 manager_dump_jobs(u
->manager
, stdout
, " ");
913 continue; /* ask again */
915 /* 'n' was removed in favor of 'f'. */
916 printf("Didn't understand 'n', did you mean 'f'?\n");
917 continue; /* ask again */
919 printf("Skipping execution.\n");
920 r
= CONFIRM_PRETEND_SUCCESS
;
926 assert_not_reached();
932 restore_confirm_stdio(&saved_stdin
, &saved_stdout
);
936 static int get_fixed_user(const ExecContext
*c
, const char **user
,
937 uid_t
*uid
, gid_t
*gid
,
938 const char **home
, const char **shell
) {
947 /* Note that we don't set $HOME or $SHELL if they are not particularly enlightening anyway
948 * (i.e. are "/" or "/bin/nologin"). */
951 r
= get_user_creds(&name
, uid
, gid
, home
, shell
, USER_CREDS_CLEAN
);
959 static int get_fixed_group(const ExecContext
*c
, const char **group
, gid_t
*gid
) {
969 r
= get_group_creds(&name
, gid
, 0);
977 static int get_supplementary_groups(const ExecContext
*c
, const char *user
,
978 const char *group
, gid_t gid
,
979 gid_t
**supplementary_gids
, int *ngids
) {
983 bool keep_groups
= false;
984 gid_t
*groups
= NULL
;
985 _cleanup_free_ gid_t
*l_gids
= NULL
;
990 * If user is given, then lookup GID and supplementary groups list.
991 * We avoid NSS lookups for gid=0. Also we have to initialize groups
992 * here and as early as possible so we keep the list of supplementary
993 * groups of the caller.
995 if (user
&& gid_is_valid(gid
) && gid
!= 0) {
996 /* First step, initialize groups from /etc/groups */
997 if (initgroups(user
, gid
) < 0)
1003 if (strv_isempty(c
->supplementary_groups
))
1007 * If SupplementaryGroups= was passed then NGROUPS_MAX has to
1008 * be positive, otherwise fail.
1011 ngroups_max
= (int) sysconf(_SC_NGROUPS_MAX
);
1012 if (ngroups_max
<= 0)
1013 return errno_or_else(EOPNOTSUPP
);
1015 l_gids
= new(gid_t
, ngroups_max
);
1021 * Lookup the list of groups that the user belongs to, we
1022 * avoid NSS lookups here too for gid=0.
1025 if (getgrouplist(user
, gid
, l_gids
, &k
) < 0)
1030 STRV_FOREACH(i
, c
->supplementary_groups
) {
1033 if (k
>= ngroups_max
)
1037 r
= get_group_creds(&g
, l_gids
+k
, 0);
1045 * Sets ngids to zero to drop all supplementary groups, happens
1046 * when we are under root and SupplementaryGroups= is empty.
1053 /* Otherwise get the final list of supplementary groups */
1054 groups
= memdup(l_gids
, sizeof(gid_t
) * k
);
1058 *supplementary_gids
= groups
;
1066 static int enforce_groups(gid_t gid
, const gid_t
*supplementary_gids
, int ngids
) {
1069 /* Handle SupplementaryGroups= if it is not empty */
1071 r
= maybe_setgroups(ngids
, supplementary_gids
);
1076 if (gid_is_valid(gid
)) {
1077 /* Then set our gids */
1078 if (setresgid(gid
, gid
, gid
) < 0)
1085 static int set_securebits(int bits
, int mask
) {
1086 int current
, applied
;
1087 current
= prctl(PR_GET_SECUREBITS
);
1090 /* Clear all securebits defined in mask and set bits */
1091 applied
= (current
& ~mask
) | bits
;
1092 if (current
== applied
)
1094 if (prctl(PR_SET_SECUREBITS
, applied
) < 0)
1099 static int enforce_user(const ExecContext
*context
, uid_t uid
) {
1103 if (!uid_is_valid(uid
))
1106 /* Sets (but doesn't look up) the uid and make sure we keep the
1107 * capabilities while doing so. For setting secure bits the capability CAP_SETPCAP is
1108 * required, so we also need keep-caps in this case.
1111 if (context
->capability_ambient_set
!= 0 || context
->secure_bits
!= 0) {
1113 /* First step: If we need to keep capabilities but
1114 * drop privileges we need to make sure we keep our
1115 * caps, while we drop privileges. */
1117 /* Add KEEP_CAPS to the securebits */
1118 r
= set_securebits(1<<SECURE_KEEP_CAPS
, 0);
1124 /* Second step: actually set the uids */
1125 if (setresuid(uid
, uid
, uid
) < 0)
1128 /* At this point we should have all necessary capabilities but
1129 are otherwise a normal user. However, the caps might got
1130 corrupted due to the setresuid() so we need clean them up
1131 later. This is done outside of this call. */
1138 static int null_conv(
1140 const struct pam_message
**msg
,
1141 struct pam_response
**resp
,
1142 void *appdata_ptr
) {
1144 /* We don't support conversations */
1146 return PAM_CONV_ERR
;
1151 static int setup_pam(
1158 const int fds
[], size_t n_fds
) {
1162 static const struct pam_conv conv
= {
1167 _cleanup_(barrier_destroy
) Barrier barrier
= BARRIER_NULL
;
1168 pam_handle_t
*handle
= NULL
;
1170 int pam_code
= PAM_SUCCESS
, r
;
1171 char **nv
, **e
= NULL
;
1172 bool close_session
= false;
1173 pid_t pam_pid
= 0, parent_pid
;
1180 /* We set up PAM in the parent process, then fork. The child
1181 * will then stay around until killed via PR_GET_PDEATHSIG or
1182 * systemd via the cgroup logic. It will then remove the PAM
1183 * session again. The parent process will exec() the actual
1184 * daemon. We do things this way to ensure that the main PID
1185 * of the daemon is the one we initially fork()ed. */
1187 r
= barrier_create(&barrier
);
1191 if (log_get_max_level() < LOG_DEBUG
)
1192 flags
|= PAM_SILENT
;
1194 pam_code
= pam_start(name
, user
, &conv
, &handle
);
1195 if (pam_code
!= PAM_SUCCESS
) {
1201 _cleanup_free_
char *q
= NULL
;
1203 /* Hmm, so no TTY was explicitly passed, but an fd passed to us directly might be a TTY. Let's figure
1204 * out if that's the case, and read the TTY off it. */
1206 if (getttyname_malloc(STDIN_FILENO
, &q
) >= 0)
1207 tty
= strjoina("/dev/", q
);
1211 pam_code
= pam_set_item(handle
, PAM_TTY
, tty
);
1212 if (pam_code
!= PAM_SUCCESS
)
1216 STRV_FOREACH(nv
, *env
) {
1217 pam_code
= pam_putenv(handle
, *nv
);
1218 if (pam_code
!= PAM_SUCCESS
)
1222 pam_code
= pam_acct_mgmt(handle
, flags
);
1223 if (pam_code
!= PAM_SUCCESS
)
1226 pam_code
= pam_setcred(handle
, PAM_ESTABLISH_CRED
| flags
);
1227 if (pam_code
!= PAM_SUCCESS
)
1228 log_debug("pam_setcred() failed, ignoring: %s", pam_strerror(handle
, pam_code
));
1230 pam_code
= pam_open_session(handle
, flags
);
1231 if (pam_code
!= PAM_SUCCESS
)
1234 close_session
= true;
1236 e
= pam_getenvlist(handle
);
1238 pam_code
= PAM_BUF_ERR
;
1242 /* Block SIGTERM, so that we know that it won't get lost in
1245 assert_se(sigprocmask_many(SIG_BLOCK
, &old_ss
, SIGTERM
, -1) >= 0);
1247 parent_pid
= getpid_cached();
1249 r
= safe_fork("(sd-pam)", 0, &pam_pid
);
1253 int sig
, ret
= EXIT_PAM
;
1255 /* The child's job is to reset the PAM session on
1257 barrier_set_role(&barrier
, BARRIER_CHILD
);
1259 /* Make sure we don't keep open the passed fds in this child. We assume that otherwise only
1260 * those fds are open here that have been opened by PAM. */
1261 (void) close_many(fds
, n_fds
);
1263 /* Drop privileges - we don't need any to pam_close_session
1264 * and this will make PR_SET_PDEATHSIG work in most cases.
1265 * If this fails, ignore the error - but expect sd-pam threads
1266 * to fail to exit normally */
1268 r
= maybe_setgroups(0, NULL
);
1270 log_warning_errno(r
, "Failed to setgroups() in sd-pam: %m");
1271 if (setresgid(gid
, gid
, gid
) < 0)
1272 log_warning_errno(errno
, "Failed to setresgid() in sd-pam: %m");
1273 if (setresuid(uid
, uid
, uid
) < 0)
1274 log_warning_errno(errno
, "Failed to setresuid() in sd-pam: %m");
1276 (void) ignore_signals(SIGPIPE
);
1278 /* Wait until our parent died. This will only work if
1279 * the above setresuid() succeeds, otherwise the kernel
1280 * will not allow unprivileged parents kill their privileged
1281 * children this way. We rely on the control groups kill logic
1282 * to do the rest for us. */
1283 if (prctl(PR_SET_PDEATHSIG
, SIGTERM
) < 0)
1286 /* Tell the parent that our setup is done. This is especially
1287 * important regarding dropping privileges. Otherwise, unit
1288 * setup might race against our setresuid(2) call.
1290 * If the parent aborted, we'll detect this below, hence ignore
1291 * return failure here. */
1292 (void) barrier_place(&barrier
);
1294 /* Check if our parent process might already have died? */
1295 if (getppid() == parent_pid
) {
1298 assert_se(sigemptyset(&ss
) >= 0);
1299 assert_se(sigaddset(&ss
, SIGTERM
) >= 0);
1302 if (sigwait(&ss
, &sig
) < 0) {
1309 assert(sig
== SIGTERM
);
1314 pam_code
= pam_setcred(handle
, PAM_DELETE_CRED
| flags
);
1315 if (pam_code
!= PAM_SUCCESS
)
1318 /* If our parent died we'll end the session */
1319 if (getppid() != parent_pid
) {
1320 pam_code
= pam_close_session(handle
, flags
);
1321 if (pam_code
!= PAM_SUCCESS
)
1328 pam_end(handle
, pam_code
| flags
);
1332 barrier_set_role(&barrier
, BARRIER_PARENT
);
1334 /* If the child was forked off successfully it will do all the
1335 * cleanups, so forget about the handle here. */
1338 /* Unblock SIGTERM again in the parent */
1339 assert_se(sigprocmask(SIG_SETMASK
, &old_ss
, NULL
) >= 0);
1341 /* We close the log explicitly here, since the PAM modules
1342 * might have opened it, but we don't want this fd around. */
1345 /* Synchronously wait for the child to initialize. We don't care for
1346 * errors as we cannot recover. However, warn loudly if it happens. */
1347 if (!barrier_place_and_sync(&barrier
))
1348 log_error("PAM initialization failed");
1350 return strv_free_and_replace(*env
, e
);
1353 if (pam_code
!= PAM_SUCCESS
) {
1354 log_error("PAM failed: %s", pam_strerror(handle
, pam_code
));
1355 r
= -EPERM
; /* PAM errors do not map to errno */
1357 log_error_errno(r
, "PAM failed: %m");
1361 pam_code
= pam_close_session(handle
, flags
);
1363 pam_end(handle
, pam_code
| flags
);
1375 static void rename_process_from_path(const char *path
) {
1376 char process_name
[11];
1380 /* This resulting string must fit in 10 chars (i.e. the length
1381 * of "/sbin/init") to look pretty in /bin/ps */
1385 rename_process("(...)");
1391 /* The end of the process name is usually more
1392 * interesting, since the first bit might just be
1398 process_name
[0] = '(';
1399 memcpy(process_name
+1, p
, l
);
1400 process_name
[1+l
] = ')';
1401 process_name
[1+l
+1] = 0;
1403 rename_process(process_name
);
1406 static bool context_has_address_families(const ExecContext
*c
) {
1409 return c
->address_families_allow_list
||
1410 !set_isempty(c
->address_families
);
1413 static bool context_has_syscall_filters(const ExecContext
*c
) {
1416 return c
->syscall_allow_list
||
1417 !hashmap_isempty(c
->syscall_filter
);
1420 static bool context_has_syscall_logs(const ExecContext
*c
) {
1423 return c
->syscall_log_allow_list
||
1424 !hashmap_isempty(c
->syscall_log
);
1427 static bool context_has_no_new_privileges(const ExecContext
*c
) {
1430 if (c
->no_new_privileges
)
1433 if (have_effective_cap(CAP_SYS_ADMIN
)) /* if we are privileged, we don't need NNP */
1436 /* We need NNP if we have any form of seccomp and are unprivileged */
1437 return c
->lock_personality
||
1438 c
->memory_deny_write_execute
||
1439 c
->private_devices
||
1441 c
->protect_hostname
||
1442 c
->protect_kernel_tunables
||
1443 c
->protect_kernel_modules
||
1444 c
->protect_kernel_logs
||
1445 context_has_address_families(c
) ||
1446 exec_context_restrict_namespaces_set(c
) ||
1447 c
->restrict_realtime
||
1448 c
->restrict_suid_sgid
||
1449 !set_isempty(c
->syscall_archs
) ||
1450 context_has_syscall_filters(c
) ||
1451 context_has_syscall_logs(c
);
1454 static bool exec_context_has_credentials(const ExecContext
*context
) {
1458 return !hashmap_isempty(context
->set_credentials
) ||
1459 !hashmap_isempty(context
->load_credentials
);
1464 static bool skip_seccomp_unavailable(const Unit
* u
, const char* msg
) {
1466 if (is_seccomp_available())
1469 log_unit_debug(u
, "SECCOMP features not detected in the kernel, skipping %s", msg
);
1473 static int apply_syscall_filter(const Unit
* u
, const ExecContext
*c
, bool needs_ambient_hack
) {
1474 uint32_t negative_action
, default_action
, action
;
1480 if (!context_has_syscall_filters(c
))
1483 if (skip_seccomp_unavailable(u
, "SystemCallFilter="))
1486 negative_action
= c
->syscall_errno
== SECCOMP_ERROR_NUMBER_KILL
? scmp_act_kill_process() : SCMP_ACT_ERRNO(c
->syscall_errno
);
1488 if (c
->syscall_allow_list
) {
1489 default_action
= negative_action
;
1490 action
= SCMP_ACT_ALLOW
;
1492 default_action
= SCMP_ACT_ALLOW
;
1493 action
= negative_action
;
1496 if (needs_ambient_hack
) {
1497 r
= seccomp_filter_set_add(c
->syscall_filter
, c
->syscall_allow_list
, syscall_filter_sets
+ SYSCALL_FILTER_SET_SETUID
);
1502 return seccomp_load_syscall_filter_set_raw(default_action
, c
->syscall_filter
, action
, false);
1505 static int apply_syscall_log(const Unit
* u
, const ExecContext
*c
) {
1507 uint32_t default_action
, action
;
1513 if (!context_has_syscall_logs(c
))
1517 if (skip_seccomp_unavailable(u
, "SystemCallLog="))
1520 if (c
->syscall_log_allow_list
) {
1521 /* Log nothing but the ones listed */
1522 default_action
= SCMP_ACT_ALLOW
;
1523 action
= SCMP_ACT_LOG
;
1525 /* Log everything but the ones listed */
1526 default_action
= SCMP_ACT_LOG
;
1527 action
= SCMP_ACT_ALLOW
;
1530 return seccomp_load_syscall_filter_set_raw(default_action
, c
->syscall_log
, action
, false);
1532 /* old libseccomp */
1533 log_unit_debug(u
, "SECCOMP feature SCMP_ACT_LOG not available, skipping SystemCallLog=");
1538 static int apply_syscall_archs(const Unit
*u
, const ExecContext
*c
) {
1542 if (set_isempty(c
->syscall_archs
))
1545 if (skip_seccomp_unavailable(u
, "SystemCallArchitectures="))
1548 return seccomp_restrict_archs(c
->syscall_archs
);
1551 static int apply_address_families(const Unit
* u
, const ExecContext
*c
) {
1555 if (!context_has_address_families(c
))
1558 if (skip_seccomp_unavailable(u
, "RestrictAddressFamilies="))
1561 return seccomp_restrict_address_families(c
->address_families
, c
->address_families_allow_list
);
1564 static int apply_memory_deny_write_execute(const Unit
* u
, const ExecContext
*c
) {
1568 if (!c
->memory_deny_write_execute
)
1571 if (skip_seccomp_unavailable(u
, "MemoryDenyWriteExecute="))
1574 return seccomp_memory_deny_write_execute();
1577 static int apply_restrict_realtime(const Unit
* u
, const ExecContext
*c
) {
1581 if (!c
->restrict_realtime
)
1584 if (skip_seccomp_unavailable(u
, "RestrictRealtime="))
1587 return seccomp_restrict_realtime();
1590 static int apply_restrict_suid_sgid(const Unit
* u
, const ExecContext
*c
) {
1594 if (!c
->restrict_suid_sgid
)
1597 if (skip_seccomp_unavailable(u
, "RestrictSUIDSGID="))
1600 return seccomp_restrict_suid_sgid();
1603 static int apply_protect_sysctl(const Unit
*u
, const ExecContext
*c
) {
1607 /* Turn off the legacy sysctl() system call. Many distributions turn this off while building the kernel, but
1608 * let's protect even those systems where this is left on in the kernel. */
1610 if (!c
->protect_kernel_tunables
)
1613 if (skip_seccomp_unavailable(u
, "ProtectKernelTunables="))
1616 return seccomp_protect_sysctl();
1619 static int apply_protect_kernel_modules(const Unit
*u
, const ExecContext
*c
) {
1623 /* Turn off module syscalls on ProtectKernelModules=yes */
1625 if (!c
->protect_kernel_modules
)
1628 if (skip_seccomp_unavailable(u
, "ProtectKernelModules="))
1631 return seccomp_load_syscall_filter_set(SCMP_ACT_ALLOW
, syscall_filter_sets
+ SYSCALL_FILTER_SET_MODULE
, SCMP_ACT_ERRNO(EPERM
), false);
1634 static int apply_protect_kernel_logs(const Unit
*u
, const ExecContext
*c
) {
1638 if (!c
->protect_kernel_logs
)
1641 if (skip_seccomp_unavailable(u
, "ProtectKernelLogs="))
1644 return seccomp_protect_syslog();
1647 static int apply_protect_clock(const Unit
*u
, const ExecContext
*c
) {
1651 if (!c
->protect_clock
)
1654 if (skip_seccomp_unavailable(u
, "ProtectClock="))
1657 return seccomp_load_syscall_filter_set(SCMP_ACT_ALLOW
, syscall_filter_sets
+ SYSCALL_FILTER_SET_CLOCK
, SCMP_ACT_ERRNO(EPERM
), false);
1660 static int apply_private_devices(const Unit
*u
, const ExecContext
*c
) {
1664 /* If PrivateDevices= is set, also turn off iopl and all @raw-io syscalls. */
1666 if (!c
->private_devices
)
1669 if (skip_seccomp_unavailable(u
, "PrivateDevices="))
1672 return seccomp_load_syscall_filter_set(SCMP_ACT_ALLOW
, syscall_filter_sets
+ SYSCALL_FILTER_SET_RAW_IO
, SCMP_ACT_ERRNO(EPERM
), false);
1675 static int apply_restrict_namespaces(const Unit
*u
, const ExecContext
*c
) {
1679 if (!exec_context_restrict_namespaces_set(c
))
1682 if (skip_seccomp_unavailable(u
, "RestrictNamespaces="))
1685 return seccomp_restrict_namespaces(c
->restrict_namespaces
);
1688 static int apply_lock_personality(const Unit
* u
, const ExecContext
*c
) {
1689 unsigned long personality
;
1695 if (!c
->lock_personality
)
1698 if (skip_seccomp_unavailable(u
, "LockPersonality="))
1701 personality
= c
->personality
;
1703 /* If personality is not specified, use either PER_LINUX or PER_LINUX32 depending on what is currently set. */
1704 if (personality
== PERSONALITY_INVALID
) {
1706 r
= opinionated_personality(&personality
);
1711 return seccomp_lock_personality(personality
);
1716 static int apply_protect_hostname(const Unit
*u
, const ExecContext
*c
, int *ret_exit_status
) {
1720 if (!c
->protect_hostname
)
1723 if (ns_type_supported(NAMESPACE_UTS
)) {
1724 if (unshare(CLONE_NEWUTS
) < 0) {
1725 if (!ERRNO_IS_NOT_SUPPORTED(errno
) && !ERRNO_IS_PRIVILEGE(errno
)) {
1726 *ret_exit_status
= EXIT_NAMESPACE
;
1727 return log_unit_error_errno(u
, errno
, "Failed to set up UTS namespacing: %m");
1730 log_unit_warning(u
, "ProtectHostname=yes is configured, but UTS namespace setup is prohibited (container manager?), ignoring namespace setup.");
1733 log_unit_warning(u
, "ProtectHostname=yes is configured, but the kernel does not support UTS namespaces, ignoring namespace setup.");
1738 if (skip_seccomp_unavailable(u
, "ProtectHostname="))
1741 r
= seccomp_protect_hostname();
1743 *ret_exit_status
= EXIT_SECCOMP
;
1744 return log_unit_error_errno(u
, r
, "Failed to apply hostname restrictions: %m");
1751 static void do_idle_pipe_dance(int idle_pipe
[static 4]) {
1754 idle_pipe
[1] = safe_close(idle_pipe
[1]);
1755 idle_pipe
[2] = safe_close(idle_pipe
[2]);
1757 if (idle_pipe
[0] >= 0) {
1760 r
= fd_wait_for_event(idle_pipe
[0], POLLHUP
, IDLE_TIMEOUT_USEC
);
1762 if (idle_pipe
[3] >= 0 && r
== 0 /* timeout */) {
1765 /* Signal systemd that we are bored and want to continue. */
1766 n
= write(idle_pipe
[3], "x", 1);
1768 /* Wait for systemd to react to the signal above. */
1769 (void) fd_wait_for_event(idle_pipe
[0], POLLHUP
, IDLE_TIMEOUT2_USEC
);
1772 idle_pipe
[0] = safe_close(idle_pipe
[0]);
1776 idle_pipe
[3] = safe_close(idle_pipe
[3]);
1779 static const char *exec_directory_env_name_to_string(ExecDirectoryType t
);
1781 static int build_environment(
1783 const ExecContext
*c
,
1784 const ExecParameters
*p
,
1787 const char *username
,
1789 dev_t journal_stream_dev
,
1790 ino_t journal_stream_ino
,
1793 _cleanup_strv_free_
char **our_env
= NULL
;
1802 #define N_ENV_VARS 17
1803 our_env
= new0(char*, N_ENV_VARS
+ _EXEC_DIRECTORY_TYPE_MAX
);
1808 _cleanup_free_
char *joined
= NULL
;
1810 if (asprintf(&x
, "LISTEN_PID="PID_FMT
, getpid_cached()) < 0)
1812 our_env
[n_env
++] = x
;
1814 if (asprintf(&x
, "LISTEN_FDS=%zu", n_fds
) < 0)
1816 our_env
[n_env
++] = x
;
1818 joined
= strv_join(p
->fd_names
, ":");
1822 x
= strjoin("LISTEN_FDNAMES=", joined
);
1825 our_env
[n_env
++] = x
;
1828 if ((p
->flags
& EXEC_SET_WATCHDOG
) && p
->watchdog_usec
> 0) {
1829 if (asprintf(&x
, "WATCHDOG_PID="PID_FMT
, getpid_cached()) < 0)
1831 our_env
[n_env
++] = x
;
1833 if (asprintf(&x
, "WATCHDOG_USEC="USEC_FMT
, p
->watchdog_usec
) < 0)
1835 our_env
[n_env
++] = x
;
1838 /* If this is D-Bus, tell the nss-systemd module, since it relies on being able to use D-Bus look up dynamic
1839 * users via PID 1, possibly dead-locking the dbus daemon. This way it will not use D-Bus to resolve names, but
1840 * check the database directly. */
1841 if (p
->flags
& EXEC_NSS_BYPASS_BUS
) {
1842 x
= strdup("SYSTEMD_NSS_BYPASS_BUS=1");
1845 our_env
[n_env
++] = x
;
1849 x
= strjoin("HOME=", home
);
1853 path_simplify(x
+ 5);
1854 our_env
[n_env
++] = x
;
1858 x
= strjoin("LOGNAME=", username
);
1861 our_env
[n_env
++] = x
;
1863 x
= strjoin("USER=", username
);
1866 our_env
[n_env
++] = x
;
1870 x
= strjoin("SHELL=", shell
);
1874 path_simplify(x
+ 6);
1875 our_env
[n_env
++] = x
;
1878 if (!sd_id128_is_null(u
->invocation_id
)) {
1879 if (asprintf(&x
, "INVOCATION_ID=" SD_ID128_FORMAT_STR
, SD_ID128_FORMAT_VAL(u
->invocation_id
)) < 0)
1882 our_env
[n_env
++] = x
;
1885 if (exec_context_needs_term(c
)) {
1886 const char *tty_path
, *term
= NULL
;
1888 tty_path
= exec_context_tty_path(c
);
1890 /* If we are forked off PID 1 and we are supposed to operate on /dev/console, then let's try
1891 * to inherit the $TERM set for PID 1. This is useful for containers so that the $TERM the
1892 * container manager passes to PID 1 ends up all the way in the console login shown. */
1894 if (path_equal_ptr(tty_path
, "/dev/console") && getppid() == 1)
1895 term
= getenv("TERM");
1898 term
= default_term_for_tty(tty_path
);
1900 x
= strjoin("TERM=", term
);
1903 our_env
[n_env
++] = x
;
1906 if (journal_stream_dev
!= 0 && journal_stream_ino
!= 0) {
1907 if (asprintf(&x
, "JOURNAL_STREAM=" DEV_FMT
":" INO_FMT
, journal_stream_dev
, journal_stream_ino
) < 0)
1910 our_env
[n_env
++] = x
;
1913 if (c
->log_namespace
) {
1914 x
= strjoin("LOG_NAMESPACE=", c
->log_namespace
);
1918 our_env
[n_env
++] = x
;
1921 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
1922 _cleanup_free_
char *pre
= NULL
, *joined
= NULL
;
1928 if (strv_isempty(c
->directories
[t
].paths
))
1931 n
= exec_directory_env_name_to_string(t
);
1935 pre
= strjoin(p
->prefix
[t
], "/");
1939 joined
= strv_join_full(c
->directories
[t
].paths
, ":", pre
, true);
1943 x
= strjoin(n
, "=", joined
);
1947 our_env
[n_env
++] = x
;
1950 if (exec_context_has_credentials(c
) && p
->prefix
[EXEC_DIRECTORY_RUNTIME
]) {
1951 x
= strjoin("CREDENTIALS_DIRECTORY=", p
->prefix
[EXEC_DIRECTORY_RUNTIME
], "/credentials/", u
->id
);
1955 our_env
[n_env
++] = x
;
1958 if (asprintf(&x
, "SYSTEMD_EXEC_PID=" PID_FMT
, getpid_cached()) < 0)
1961 our_env
[n_env
++] = x
;
1963 our_env
[n_env
++] = NULL
;
1964 assert(n_env
<= N_ENV_VARS
+ _EXEC_DIRECTORY_TYPE_MAX
);
1967 *ret
= TAKE_PTR(our_env
);
1972 static int build_pass_environment(const ExecContext
*c
, char ***ret
) {
1973 _cleanup_strv_free_
char **pass_env
= NULL
;
1977 STRV_FOREACH(i
, c
->pass_environment
) {
1978 _cleanup_free_
char *x
= NULL
;
1984 x
= strjoin(*i
, "=", v
);
1988 if (!GREEDY_REALLOC(pass_env
, n_env
+ 2))
1991 pass_env
[n_env
++] = TAKE_PTR(x
);
1992 pass_env
[n_env
] = NULL
;
1995 *ret
= TAKE_PTR(pass_env
);
2000 bool exec_needs_mount_namespace(
2001 const ExecContext
*context
,
2002 const ExecParameters
*params
,
2003 const ExecRuntime
*runtime
) {
2007 if (context
->root_image
)
2010 if (!strv_isempty(context
->read_write_paths
) ||
2011 !strv_isempty(context
->read_only_paths
) ||
2012 !strv_isempty(context
->inaccessible_paths
) ||
2013 !strv_isempty(context
->exec_paths
) ||
2014 !strv_isempty(context
->no_exec_paths
))
2017 if (context
->n_bind_mounts
> 0)
2020 if (context
->n_temporary_filesystems
> 0)
2023 if (context
->n_mount_images
> 0)
2026 if (context
->n_extension_images
> 0)
2029 if (!IN_SET(context
->mount_flags
, 0, MS_SHARED
))
2032 if (context
->private_tmp
&& runtime
&& (runtime
->tmp_dir
|| runtime
->var_tmp_dir
))
2035 if (context
->private_devices
||
2036 context
->private_mounts
||
2037 context
->protect_system
!= PROTECT_SYSTEM_NO
||
2038 context
->protect_home
!= PROTECT_HOME_NO
||
2039 context
->protect_kernel_tunables
||
2040 context
->protect_kernel_modules
||
2041 context
->protect_kernel_logs
||
2042 context
->protect_control_groups
||
2043 context
->protect_proc
!= PROTECT_PROC_DEFAULT
||
2044 context
->proc_subset
!= PROC_SUBSET_ALL
||
2045 context
->private_ipc
||
2046 context
->ipc_namespace_path
)
2049 if (context
->root_directory
) {
2050 if (exec_context_get_effective_mount_apivfs(context
))
2053 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
2054 if (params
&& !params
->prefix
[t
])
2057 if (!strv_isempty(context
->directories
[t
].paths
))
2062 if (context
->dynamic_user
&&
2063 (!strv_isempty(context
->directories
[EXEC_DIRECTORY_STATE
].paths
) ||
2064 !strv_isempty(context
->directories
[EXEC_DIRECTORY_CACHE
].paths
) ||
2065 !strv_isempty(context
->directories
[EXEC_DIRECTORY_LOGS
].paths
)))
2068 if (context
->log_namespace
)
2074 static int setup_private_users(uid_t ouid
, gid_t ogid
, uid_t uid
, gid_t gid
) {
2075 _cleanup_free_
char *uid_map
= NULL
, *gid_map
= NULL
;
2076 _cleanup_close_pair_
int errno_pipe
[2] = { -1, -1 };
2077 _cleanup_close_
int unshare_ready_fd
= -1;
2078 _cleanup_(sigkill_waitp
) pid_t pid
= 0;
2083 /* Set up a user namespace and map the original UID/GID (IDs from before any user or group changes, i.e.
2084 * the IDs from the user or system manager(s)) to itself, the selected UID/GID to itself, and everything else to
2085 * nobody. In order to be able to write this mapping we need CAP_SETUID in the original user namespace, which
2086 * we however lack after opening the user namespace. To work around this we fork() a temporary child process,
2087 * which waits for the parent to create the new user namespace while staying in the original namespace. The
2088 * child then writes the UID mapping, under full privileges. The parent waits for the child to finish and
2089 * continues execution normally.
2090 * For unprivileged users (i.e. without capabilities), the root to root mapping is excluded. As such, it
2091 * does not need CAP_SETUID to write the single line mapping to itself. */
2093 /* Can only set up multiple mappings with CAP_SETUID. */
2094 if (have_effective_cap(CAP_SETUID
) && uid
!= ouid
&& uid_is_valid(uid
))
2095 r
= asprintf(&uid_map
,
2096 UID_FMT
" " UID_FMT
" 1\n" /* Map $OUID → $OUID */
2097 UID_FMT
" " UID_FMT
" 1\n", /* Map $UID → $UID */
2098 ouid
, ouid
, uid
, uid
);
2100 r
= asprintf(&uid_map
,
2101 UID_FMT
" " UID_FMT
" 1\n", /* Map $OUID → $OUID */
2107 /* Can only set up multiple mappings with CAP_SETGID. */
2108 if (have_effective_cap(CAP_SETGID
) && gid
!= ogid
&& gid_is_valid(gid
))
2109 r
= asprintf(&gid_map
,
2110 GID_FMT
" " GID_FMT
" 1\n" /* Map $OGID → $OGID */
2111 GID_FMT
" " GID_FMT
" 1\n", /* Map $GID → $GID */
2112 ogid
, ogid
, gid
, gid
);
2114 r
= asprintf(&gid_map
,
2115 GID_FMT
" " GID_FMT
" 1\n", /* Map $OGID -> $OGID */
2121 /* Create a communication channel so that the parent can tell the child when it finished creating the user
2123 unshare_ready_fd
= eventfd(0, EFD_CLOEXEC
);
2124 if (unshare_ready_fd
< 0)
2127 /* Create a communication channel so that the child can tell the parent a proper error code in case it
2129 if (pipe2(errno_pipe
, O_CLOEXEC
) < 0)
2132 r
= safe_fork("(sd-userns)", FORK_RESET_SIGNALS
|FORK_DEATHSIG
, &pid
);
2136 _cleanup_close_
int fd
= -1;
2140 /* Child process, running in the original user namespace. Let's update the parent's UID/GID map from
2141 * here, after the parent opened its own user namespace. */
2144 errno_pipe
[0] = safe_close(errno_pipe
[0]);
2146 /* Wait until the parent unshared the user namespace */
2147 if (read(unshare_ready_fd
, &c
, sizeof(c
)) < 0) {
2152 /* Disable the setgroups() system call in the child user namespace, for good. */
2153 a
= procfs_file_alloca(ppid
, "setgroups");
2154 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
2156 if (errno
!= ENOENT
) {
2161 /* If the file is missing the kernel is too old, let's continue anyway. */
2163 if (write(fd
, "deny\n", 5) < 0) {
2168 fd
= safe_close(fd
);
2171 /* First write the GID map */
2172 a
= procfs_file_alloca(ppid
, "gid_map");
2173 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
2178 if (write(fd
, gid_map
, strlen(gid_map
)) < 0) {
2182 fd
= safe_close(fd
);
2184 /* The write the UID map */
2185 a
= procfs_file_alloca(ppid
, "uid_map");
2186 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
2191 if (write(fd
, uid_map
, strlen(uid_map
)) < 0) {
2196 _exit(EXIT_SUCCESS
);
2199 (void) write(errno_pipe
[1], &r
, sizeof(r
));
2200 _exit(EXIT_FAILURE
);
2203 errno_pipe
[1] = safe_close(errno_pipe
[1]);
2205 if (unshare(CLONE_NEWUSER
) < 0)
2208 /* Let the child know that the namespace is ready now */
2209 if (write(unshare_ready_fd
, &c
, sizeof(c
)) < 0)
2212 /* Try to read an error code from the child */
2213 n
= read(errno_pipe
[0], &r
, sizeof(r
));
2216 if (n
== sizeof(r
)) { /* an error code was sent to us */
2221 if (n
!= 0) /* on success we should have read 0 bytes */
2224 r
= wait_for_terminate_and_check("(sd-userns)", pid
, 0);
2228 if (r
!= EXIT_SUCCESS
) /* If something strange happened with the child, let's consider this fatal, too */
2234 static bool exec_directory_is_private(const ExecContext
*context
, ExecDirectoryType type
) {
2235 if (!context
->dynamic_user
)
2238 if (type
== EXEC_DIRECTORY_CONFIGURATION
)
2241 if (type
== EXEC_DIRECTORY_RUNTIME
&& context
->runtime_directory_preserve_mode
== EXEC_PRESERVE_NO
)
2247 static int setup_exec_directory(
2248 const ExecContext
*context
,
2249 const ExecParameters
*params
,
2252 ExecDirectoryType type
,
2255 static const int exit_status_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
2256 [EXEC_DIRECTORY_RUNTIME
] = EXIT_RUNTIME_DIRECTORY
,
2257 [EXEC_DIRECTORY_STATE
] = EXIT_STATE_DIRECTORY
,
2258 [EXEC_DIRECTORY_CACHE
] = EXIT_CACHE_DIRECTORY
,
2259 [EXEC_DIRECTORY_LOGS
] = EXIT_LOGS_DIRECTORY
,
2260 [EXEC_DIRECTORY_CONFIGURATION
] = EXIT_CONFIGURATION_DIRECTORY
,
2267 assert(type
>= 0 && type
< _EXEC_DIRECTORY_TYPE_MAX
);
2268 assert(exit_status
);
2270 if (!params
->prefix
[type
])
2273 if (params
->flags
& EXEC_CHOWN_DIRECTORIES
) {
2274 if (!uid_is_valid(uid
))
2276 if (!gid_is_valid(gid
))
2280 STRV_FOREACH(rt
, context
->directories
[type
].paths
) {
2281 _cleanup_free_
char *p
= NULL
, *pp
= NULL
;
2283 p
= path_join(params
->prefix
[type
], *rt
);
2289 r
= mkdir_parents_label(p
, 0755);
2293 if (exec_directory_is_private(context
, type
)) {
2294 /* So, here's one extra complication when dealing with DynamicUser=1 units. In that
2295 * case we want to avoid leaving a directory around fully accessible that is owned by
2296 * a dynamic user whose UID is later on reused. To lock this down we use the same
2297 * trick used by container managers to prohibit host users to get access to files of
2298 * the same UID in containers: we place everything inside a directory that has an
2299 * access mode of 0700 and is owned root:root, so that it acts as security boundary
2300 * for unprivileged host code. We then use fs namespacing to make this directory
2301 * permeable for the service itself.
2303 * Specifically: for a service which wants a special directory "foo/" we first create
2304 * a directory "private/" with access mode 0700 owned by root:root. Then we place
2305 * "foo" inside of that directory (i.e. "private/foo/"), and make "foo" a symlink to
2306 * "private/foo". This way, privileged host users can access "foo/" as usual, but
2307 * unprivileged host users can't look into it. Inside of the namespace of the unit
2308 * "private/" is replaced by a more liberally accessible tmpfs, into which the host's
2309 * "private/foo/" is mounted under the same name, thus disabling the access boundary
2310 * for the service and making sure it only gets access to the dirs it needs but no
2311 * others. Tricky? Yes, absolutely, but it works!
2313 * Note that we don't do this for EXEC_DIRECTORY_CONFIGURATION as that's assumed not
2314 * to be owned by the service itself.
2316 * Also, note that we don't do this for EXEC_DIRECTORY_RUNTIME as that's often used
2317 * for sharing files or sockets with other services. */
2319 pp
= path_join(params
->prefix
[type
], "private");
2325 /* First set up private root if it doesn't exist yet, with access mode 0700 and owned by root:root */
2326 r
= mkdir_safe_label(pp
, 0700, 0, 0, MKDIR_WARN_MODE
);
2330 if (!path_extend(&pp
, *rt
)) {
2335 /* Create all directories between the configured directory and this private root, and mark them 0755 */
2336 r
= mkdir_parents_label(pp
, 0755);
2340 if (is_dir(p
, false) > 0 &&
2341 (laccess(pp
, F_OK
) < 0 && errno
== ENOENT
)) {
2343 /* Hmm, the private directory doesn't exist yet, but the normal one exists? If so, move
2344 * it over. Most likely the service has been upgraded from one that didn't use
2345 * DynamicUser=1, to one that does. */
2347 log_info("Found pre-existing public %s= directory %s, migrating to %s.\n"
2348 "Apparently, service previously had DynamicUser= turned off, and has now turned it on.",
2349 exec_directory_type_to_string(type
), p
, pp
);
2351 if (rename(p
, pp
) < 0) {
2356 /* Otherwise, create the actual directory for the service */
2358 r
= mkdir_label(pp
, context
->directories
[type
].mode
);
2359 if (r
< 0 && r
!= -EEXIST
)
2363 /* And link it up from the original place */
2364 r
= symlink_idempotent(pp
, p
, true);
2369 _cleanup_free_
char *target
= NULL
;
2371 if (type
!= EXEC_DIRECTORY_CONFIGURATION
&&
2372 readlink_and_make_absolute(p
, &target
) >= 0) {
2373 _cleanup_free_
char *q
= NULL
, *q_resolved
= NULL
, *target_resolved
= NULL
;
2375 /* This already exists and is a symlink? Interesting. Maybe it's one created
2376 * by DynamicUser=1 (see above)?
2378 * We do this for all directory types except for ConfigurationDirectory=,
2379 * since they all support the private/ symlink logic at least in some
2380 * configurations, see above. */
2382 r
= chase_symlinks(target
, NULL
, 0, &target_resolved
, NULL
);
2386 q
= path_join(params
->prefix
[type
], "private", *rt
);
2392 /* /var/lib or friends may be symlinks. So, let's chase them also. */
2393 r
= chase_symlinks(q
, NULL
, CHASE_NONEXISTENT
, &q_resolved
, NULL
);
2397 if (path_equal(q_resolved
, target_resolved
)) {
2399 /* Hmm, apparently DynamicUser= was once turned on for this service,
2400 * but is no longer. Let's move the directory back up. */
2402 log_info("Found pre-existing private %s= directory %s, migrating to %s.\n"
2403 "Apparently, service previously had DynamicUser= turned on, and has now turned it off.",
2404 exec_directory_type_to_string(type
), q
, p
);
2406 if (unlink(p
) < 0) {
2411 if (rename(q
, p
) < 0) {
2418 r
= mkdir_label(p
, context
->directories
[type
].mode
);
2423 if (type
== EXEC_DIRECTORY_CONFIGURATION
) {
2426 /* Don't change the owner/access mode of the configuration directory,
2427 * as in the common case it is not written to by a service, and shall
2428 * not be writable. */
2430 if (stat(p
, &st
) < 0) {
2435 /* Still complain if the access mode doesn't match */
2436 if (((st
.st_mode
^ context
->directories
[type
].mode
) & 07777) != 0)
2437 log_warning("%s \'%s\' already exists but the mode is different. "
2438 "(File system: %o %sMode: %o)",
2439 exec_directory_type_to_string(type
), *rt
,
2440 st
.st_mode
& 07777, exec_directory_type_to_string(type
), context
->directories
[type
].mode
& 07777);
2447 /* Lock down the access mode (we use chmod_and_chown() to make this idempotent. We don't
2448 * specify UID/GID here, so that path_chown_recursive() can optimize things depending on the
2449 * current UID/GID ownership.) */
2450 r
= chmod_and_chown(pp
?: p
, context
->directories
[type
].mode
, UID_INVALID
, GID_INVALID
);
2454 /* Then, change the ownership of the whole tree, if necessary. When dynamic users are used we
2455 * drop the suid/sgid bits, since we really don't want SUID/SGID files for dynamic UID/GID
2456 * assignments to exist. */
2457 r
= path_chown_recursive(pp
?: p
, uid
, gid
, context
->dynamic_user
? 01777 : 07777);
2465 *exit_status
= exit_status_table
[type
];
2469 static int write_credential(
2475 bool ownership_ok
) {
2477 _cleanup_(unlink_and_freep
) char *tmp
= NULL
;
2478 _cleanup_close_
int fd
= -1;
2481 r
= tempfn_random_child("", "cred", &tmp
);
2485 fd
= openat(dfd
, tmp
, O_CREAT
|O_RDWR
|O_CLOEXEC
|O_EXCL
|O_NOFOLLOW
|O_NOCTTY
, 0600);
2491 r
= loop_write(fd
, data
, size
, /* do_poll = */ false);
2495 if (fchmod(fd
, 0400) < 0) /* Take away "w" bit */
2498 if (uid_is_valid(uid
) && uid
!= getuid()) {
2499 r
= fd_add_uid_acl_permission(fd
, uid
, ACL_READ
);
2501 if (!ERRNO_IS_NOT_SUPPORTED(r
) && !ERRNO_IS_PRIVILEGE(r
))
2504 if (!ownership_ok
) /* Ideally we use ACLs, since we can neatly express what we want
2505 * to express: that the user gets read access and nothing
2506 * else. But if the backing fs can't support that (e.g. ramfs)
2507 * then we can use file ownership instead. But that's only safe if
2508 * we can then re-mount the whole thing read-only, so that the
2509 * user can no longer chmod() the file to gain write access. */
2512 if (fchown(fd
, uid
, GID_INVALID
) < 0)
2517 if (renameat(dfd
, tmp
, dfd
, id
) < 0)
2524 static int acquire_credentials(
2525 const ExecContext
*context
,
2526 const ExecParameters
*params
,
2530 bool ownership_ok
) {
2532 uint64_t left
= CREDENTIALS_TOTAL_SIZE_MAX
;
2533 _cleanup_close_
int dfd
= -1;
2534 ExecLoadCredential
*lc
;
2535 ExecSetCredential
*sc
;
2541 dfd
= open(p
, O_DIRECTORY
|O_CLOEXEC
);
2545 /* First, load credentials off disk (or acquire via AF_UNIX socket) */
2546 HASHMAP_FOREACH(lc
, context
->load_credentials
) {
2547 ReadFullFileFlags flags
= READ_FULL_FILE_SECURE
|READ_FULL_FILE_FAIL_WHEN_LARGER
;
2548 _cleanup_(erase_and_freep
) char *data
= NULL
;
2549 _cleanup_free_
char *j
= NULL
, *bindname
= NULL
;
2550 bool missing_ok
= true;
2554 if (path_is_absolute(lc
->path
)) {
2555 /* If this is an absolute path, read the data directly from it, and support AF_UNIX sockets */
2557 flags
|= READ_FULL_FILE_CONNECT_SOCKET
;
2559 /* Pass some minimal info about the unit and the credential name we are looking to acquire
2560 * via the source socket address in case we read off an AF_UNIX socket. */
2561 if (asprintf(&bindname
, "@%" PRIx64
"/unit/%s/%s", random_u64(), unit
, lc
->id
) < 0)
2566 } else if (params
->received_credentials
) {
2567 /* If this is a relative path, take it relative to the credentials we received
2568 * ourselves. We don't support the AF_UNIX stuff in this mode, since we are operating
2569 * on a credential store, i.e. this is guaranteed to be regular files. */
2570 j
= path_join(params
->received_credentials
, lc
->path
);
2579 r
= read_full_file_full(
2582 lc
->encrypted
? CREDENTIAL_ENCRYPTED_SIZE_MAX
: CREDENTIAL_SIZE_MAX
,
2583 flags
| (lc
->encrypted
? READ_FULL_FILE_UNBASE64
: 0),
2588 if (r
== -ENOENT
&& (missing_ok
|| hashmap_contains(context
->set_credentials
, lc
->id
))) {
2589 /* Make a missing inherited credential non-fatal, let's just continue. After all apps
2590 * will get clear errors if we don't pass such a missing credential on as they
2591 * themselves will get ENOENT when trying to read them, which should not be much
2592 * worse than when we handle the error here and make it fatal.
2594 * Also, if the source file doesn't exist, but a fallback is set via SetCredentials=
2595 * we are fine, too. */
2596 log_debug_errno(r
, "Couldn't read inherited credential '%s', skipping: %m", lc
->path
);
2600 return log_debug_errno(r
, "Failed to read credential '%s': %m", lc
->path
);
2602 if (lc
->encrypted
) {
2603 _cleanup_free_
void *plaintext
= NULL
;
2604 size_t plaintext_size
= 0;
2606 r
= decrypt_credential_and_warn(lc
->id
, now(CLOCK_REALTIME
), NULL
, data
, size
, &plaintext
, &plaintext_size
);
2610 free_and_replace(data
, plaintext
);
2611 size
= plaintext_size
;
2614 add
= strlen(lc
->id
) + size
;
2618 r
= write_credential(dfd
, lc
->id
, data
, size
, uid
, ownership_ok
);
2625 /* First we use the literally specified credentials. Note that they might be overridden again below,
2626 * and thus act as a "default" if the same credential is specified multiple times */
2627 HASHMAP_FOREACH(sc
, context
->set_credentials
) {
2628 _cleanup_(erase_and_freep
) void *plaintext
= NULL
;
2632 if (faccessat(dfd
, sc
->id
, F_OK
, AT_SYMLINK_NOFOLLOW
) >= 0)
2634 if (errno
!= ENOENT
)
2635 return log_debug_errno(errno
, "Failed to test if credential %s exists: %m", sc
->id
);
2637 if (sc
->encrypted
) {
2638 r
= decrypt_credential_and_warn(sc
->id
, now(CLOCK_REALTIME
), NULL
, sc
->data
, sc
->size
, &plaintext
, &size
);
2648 add
= strlen(sc
->id
) + size
;
2652 r
= write_credential(dfd
, sc
->id
, data
, size
, uid
, ownership_ok
);
2660 if (fchmod(dfd
, 0500) < 0) /* Now take away the "w" bit */
2663 /* After we created all keys with the right perms, also make sure the credential store as a whole is
2666 if (uid_is_valid(uid
) && uid
!= getuid()) {
2667 r
= fd_add_uid_acl_permission(dfd
, uid
, ACL_READ
| ACL_EXECUTE
);
2669 if (!ERRNO_IS_NOT_SUPPORTED(r
) && !ERRNO_IS_PRIVILEGE(r
))
2675 if (fchown(dfd
, uid
, GID_INVALID
) < 0)
2683 static int setup_credentials_internal(
2684 const ExecContext
*context
,
2685 const ExecParameters
*params
,
2687 const char *final
, /* This is where the credential store shall eventually end up at */
2688 const char *workspace
, /* This is where we can prepare it before moving it to the final place */
2689 bool reuse_workspace
, /* Whether to reuse any existing workspace mount if it already is a mount */
2690 bool must_mount
, /* Whether to require that we mount something, it's not OK to use the plain directory fall back */
2693 int r
, workspace_mounted
; /* negative if we don't know yet whether we have/can mount something; true
2694 * if we mounted something; false if we definitely can't mount anything */
2702 if (reuse_workspace
) {
2703 r
= path_is_mount_point(workspace
, NULL
, 0);
2707 workspace_mounted
= true; /* If this is already a mount, and we are supposed to reuse it, let's keep this in mind */
2709 workspace_mounted
= -1; /* We need to figure out if we can mount something to the workspace */
2711 workspace_mounted
= -1; /* ditto */
2713 r
= path_is_mount_point(final
, NULL
, 0);
2717 /* If the final place already has something mounted, we use that. If the workspace also has
2718 * something mounted we assume it's actually the same mount (but with MS_RDONLY
2720 final_mounted
= true;
2722 if (workspace_mounted
< 0) {
2723 /* If the final place is mounted, but the workspace we isn't, then let's bind mount
2724 * the final version to the workspace, and make it writable, so that we can make
2727 r
= mount_nofollow_verbose(LOG_DEBUG
, final
, workspace
, NULL
, MS_BIND
|MS_REC
, NULL
);
2731 r
= mount_nofollow_verbose(LOG_DEBUG
, NULL
, workspace
, NULL
, MS_BIND
|MS_REMOUNT
|MS_NODEV
|MS_NOEXEC
|MS_NOSUID
, NULL
);
2735 workspace_mounted
= true;
2738 final_mounted
= false;
2740 if (workspace_mounted
< 0) {
2741 /* Nothing is mounted on the workspace yet, let's try to mount something now */
2742 for (int try = 0;; try++) {
2745 /* Try "ramfs" first, since it's not swap backed */
2746 r
= mount_nofollow_verbose(LOG_DEBUG
, "ramfs", workspace
, "ramfs", MS_NODEV
|MS_NOEXEC
|MS_NOSUID
, "mode=0700");
2748 workspace_mounted
= true;
2752 } else if (try == 1) {
2753 _cleanup_free_
char *opts
= NULL
;
2755 if (asprintf(&opts
, "mode=0700,nr_inodes=1024,size=%zu", (size_t) CREDENTIALS_TOTAL_SIZE_MAX
) < 0)
2758 /* Fall back to "tmpfs" otherwise */
2759 r
= mount_nofollow_verbose(LOG_DEBUG
, "tmpfs", workspace
, "tmpfs", MS_NODEV
|MS_NOEXEC
|MS_NOSUID
, opts
);
2761 workspace_mounted
= true;
2766 /* If that didn't work, try to make a bind mount from the final to the workspace, so that we can make it writable there. */
2767 r
= mount_nofollow_verbose(LOG_DEBUG
, final
, workspace
, NULL
, MS_BIND
|MS_REC
, NULL
);
2769 if (!ERRNO_IS_PRIVILEGE(r
)) /* Propagate anything that isn't a permission problem */
2772 if (must_mount
) /* If we it's not OK to use the plain directory
2773 * fallback, propagate all errors too */
2776 /* If we lack privileges to bind mount stuff, then let's gracefully
2777 * proceed for compat with container envs, and just use the final dir
2780 workspace_mounted
= false;
2784 /* Make the new bind mount writable (i.e. drop MS_RDONLY) */
2785 r
= mount_nofollow_verbose(LOG_DEBUG
, NULL
, workspace
, NULL
, MS_BIND
|MS_REMOUNT
|MS_NODEV
|MS_NOEXEC
|MS_NOSUID
, NULL
);
2789 workspace_mounted
= true;
2795 assert(!must_mount
|| workspace_mounted
> 0);
2796 where
= workspace_mounted
? workspace
: final
;
2798 r
= acquire_credentials(context
, params
, unit
, where
, uid
, workspace_mounted
);
2802 if (workspace_mounted
) {
2803 /* Make workspace read-only now, so that any bind mount we make from it defaults to read-only too */
2804 r
= mount_nofollow_verbose(LOG_DEBUG
, NULL
, workspace
, NULL
, MS_BIND
|MS_REMOUNT
|MS_RDONLY
|MS_NODEV
|MS_NOEXEC
|MS_NOSUID
, NULL
);
2808 /* And mount it to the final place, read-only */
2810 r
= umount_verbose(LOG_DEBUG
, workspace
, MNT_DETACH
|UMOUNT_NOFOLLOW
);
2812 r
= mount_nofollow_verbose(LOG_DEBUG
, workspace
, final
, NULL
, MS_MOVE
, NULL
);
2816 _cleanup_free_
char *parent
= NULL
;
2818 /* If we do not have our own mount put used the plain directory fallback, then we need to
2819 * open access to the top-level credential directory and the per-service directory now */
2821 parent
= dirname_malloc(final
);
2824 if (chmod(parent
, 0755) < 0)
2831 static int setup_credentials(
2832 const ExecContext
*context
,
2833 const ExecParameters
*params
,
2837 _cleanup_free_
char *p
= NULL
, *q
= NULL
;
2844 if (!exec_context_has_credentials(context
))
2847 if (!params
->prefix
[EXEC_DIRECTORY_RUNTIME
])
2850 /* This where we'll place stuff when we are done; this main credentials directory is world-readable,
2851 * and the subdir we mount over with a read-only file system readable by the service's user */
2852 q
= path_join(params
->prefix
[EXEC_DIRECTORY_RUNTIME
], "credentials");
2856 r
= mkdir_label(q
, 0755); /* top-level dir: world readable/searchable */
2857 if (r
< 0 && r
!= -EEXIST
)
2860 p
= path_join(q
, unit
);
2864 r
= mkdir_label(p
, 0700); /* per-unit dir: private to user */
2865 if (r
< 0 && r
!= -EEXIST
)
2868 r
= safe_fork("(sd-mkdcreds)", FORK_DEATHSIG
|FORK_WAIT
|FORK_NEW_MOUNTNS
, NULL
);
2870 _cleanup_free_
char *t
= NULL
, *u
= NULL
;
2872 /* If this is not a privilege or support issue then propagate the error */
2873 if (!ERRNO_IS_NOT_SUPPORTED(r
) && !ERRNO_IS_PRIVILEGE(r
))
2876 /* Temporary workspace, that remains inaccessible all the time. We prepare stuff there before moving
2877 * it into place, so that users can't access half-initialized credential stores. */
2878 t
= path_join(params
->prefix
[EXEC_DIRECTORY_RUNTIME
], "systemd/temporary-credentials");
2882 /* We can't set up a mount namespace. In that case operate on a fixed, inaccessible per-unit
2883 * directory outside of /run/credentials/ first, and then move it over to /run/credentials/
2884 * after it is fully set up */
2885 u
= path_join(t
, unit
);
2889 FOREACH_STRING(i
, t
, u
) {
2890 r
= mkdir_label(i
, 0700);
2891 if (r
< 0 && r
!= -EEXIST
)
2895 r
= setup_credentials_internal(
2899 p
, /* final mount point */
2900 u
, /* temporary workspace to overmount */
2901 true, /* reuse the workspace if it is already a mount */
2902 false, /* it's OK to fall back to a plain directory if we can't mount anything */
2905 (void) rmdir(u
); /* remove the workspace again if we can. */
2910 } else if (r
== 0) {
2912 /* We managed to set up a mount namespace, and are now in a child. That's great. In this case
2913 * we can use the same directory for all cases, after turning off propagation. Question
2914 * though is: where do we turn off propagation exactly, and where do we place the workspace
2915 * directory? We need some place that is guaranteed to be a mount point in the host, and
2916 * which is guaranteed to have a subdir we can mount over. /run/ is not suitable for this,
2917 * since we ultimately want to move the resulting file system there, i.e. we need propagation
2918 * for /run/ eventually. We could use our own /run/systemd/bind mount on itself, but that
2919 * would be visible in the host mount table all the time, which we want to avoid. Hence, what
2920 * we do here instead we use /dev/ and /dev/shm/ for our purposes. We know for sure that
2921 * /dev/ is a mount point and we now for sure that /dev/shm/ exists. Hence we can turn off
2922 * propagation on the former, and then overmount the latter.
2924 * Yes it's nasty playing games with /dev/ and /dev/shm/ like this, since it does not exist
2925 * for this purpose, but there are few other candidates that work equally well for us, and
2926 * given that the we do this in a privately namespaced short-lived single-threaded process
2927 * that no one else sees this should be OK to do. */
2929 r
= mount_nofollow_verbose(LOG_DEBUG
, NULL
, "/dev", NULL
, MS_SLAVE
|MS_REC
, NULL
); /* Turn off propagation from our namespace to host */
2933 r
= setup_credentials_internal(
2937 p
, /* final mount point */
2938 "/dev/shm", /* temporary workspace to overmount */
2939 false, /* do not reuse /dev/shm if it is already a mount, under no circumstances */
2940 true, /* insist that something is mounted, do not allow fallback to plain directory */
2945 _exit(EXIT_SUCCESS
);
2948 _exit(EXIT_FAILURE
);
2955 static int setup_smack(
2956 const ExecContext
*context
,
2957 int executable_fd
) {
2961 assert(executable_fd
>= 0);
2963 if (context
->smack_process_label
) {
2964 r
= mac_smack_apply_pid(0, context
->smack_process_label
);
2968 #ifdef SMACK_DEFAULT_PROCESS_LABEL
2970 _cleanup_free_
char *exec_label
= NULL
;
2972 r
= mac_smack_read_fd(executable_fd
, SMACK_ATTR_EXEC
, &exec_label
);
2973 if (r
< 0 && !IN_SET(r
, -ENODATA
, -EOPNOTSUPP
))
2976 r
= mac_smack_apply_pid(0, exec_label
? : SMACK_DEFAULT_PROCESS_LABEL
);
2986 static int compile_bind_mounts(
2987 const ExecContext
*context
,
2988 const ExecParameters
*params
,
2989 BindMount
**ret_bind_mounts
,
2990 size_t *ret_n_bind_mounts
,
2991 char ***ret_empty_directories
) {
2993 _cleanup_strv_free_
char **empty_directories
= NULL
;
2994 BindMount
*bind_mounts
;
3000 assert(ret_bind_mounts
);
3001 assert(ret_n_bind_mounts
);
3002 assert(ret_empty_directories
);
3004 n
= context
->n_bind_mounts
;
3005 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
3006 if (!params
->prefix
[t
])
3009 n
+= strv_length(context
->directories
[t
].paths
);
3013 *ret_bind_mounts
= NULL
;
3014 *ret_n_bind_mounts
= 0;
3015 *ret_empty_directories
= NULL
;
3019 bind_mounts
= new(BindMount
, n
);
3023 for (size_t i
= 0; i
< context
->n_bind_mounts
; i
++) {
3024 BindMount
*item
= context
->bind_mounts
+ i
;
3027 s
= strdup(item
->source
);
3033 d
= strdup(item
->destination
);
3040 bind_mounts
[h
++] = (BindMount
) {
3043 .read_only
= item
->read_only
,
3044 .recursive
= item
->recursive
,
3045 .ignore_enoent
= item
->ignore_enoent
,
3049 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
3052 if (!params
->prefix
[t
])
3055 if (strv_isempty(context
->directories
[t
].paths
))
3058 if (exec_directory_is_private(context
, t
) &&
3059 !exec_context_with_rootfs(context
)) {
3062 /* So this is for a dynamic user, and we need to make sure the process can access its own
3063 * directory. For that we overmount the usually inaccessible "private" subdirectory with a
3064 * tmpfs that makes it accessible and is empty except for the submounts we do this for. */
3066 private_root
= path_join(params
->prefix
[t
], "private");
3067 if (!private_root
) {
3072 r
= strv_consume(&empty_directories
, private_root
);
3077 STRV_FOREACH(suffix
, context
->directories
[t
].paths
) {
3080 if (exec_directory_is_private(context
, t
))
3081 s
= path_join(params
->prefix
[t
], "private", *suffix
);
3083 s
= path_join(params
->prefix
[t
], *suffix
);
3089 if (exec_directory_is_private(context
, t
) &&
3090 exec_context_with_rootfs(context
))
3091 /* When RootDirectory= or RootImage= are set, then the symbolic link to the private
3092 * directory is not created on the root directory. So, let's bind-mount the directory
3093 * on the 'non-private' place. */
3094 d
= path_join(params
->prefix
[t
], *suffix
);
3103 bind_mounts
[h
++] = (BindMount
) {
3107 .nosuid
= context
->dynamic_user
, /* don't allow suid/sgid when DynamicUser= is on */
3109 .ignore_enoent
= false,
3116 *ret_bind_mounts
= bind_mounts
;
3117 *ret_n_bind_mounts
= n
;
3118 *ret_empty_directories
= TAKE_PTR(empty_directories
);
3123 bind_mount_free_many(bind_mounts
, h
);
3127 static bool insist_on_sandboxing(
3128 const ExecContext
*context
,
3129 const char *root_dir
,
3130 const char *root_image
,
3131 const BindMount
*bind_mounts
,
3132 size_t n_bind_mounts
) {
3135 assert(n_bind_mounts
== 0 || bind_mounts
);
3137 /* Checks whether we need to insist on fs namespacing. i.e. whether we have settings configured that
3138 * would alter the view on the file system beyond making things read-only or invisible, i.e. would
3139 * rearrange stuff in a way we cannot ignore gracefully. */
3141 if (context
->n_temporary_filesystems
> 0)
3144 if (root_dir
|| root_image
)
3147 if (context
->n_mount_images
> 0)
3150 if (context
->dynamic_user
)
3153 /* If there are any bind mounts set that don't map back onto themselves, fs namespacing becomes
3155 for (size_t i
= 0; i
< n_bind_mounts
; i
++)
3156 if (!path_equal(bind_mounts
[i
].source
, bind_mounts
[i
].destination
))
3159 if (context
->log_namespace
)
3165 static int apply_mount_namespace(
3167 ExecCommandFlags command_flags
,
3168 const ExecContext
*context
,
3169 const ExecParameters
*params
,
3170 const ExecRuntime
*runtime
,
3171 char **error_path
) {
3173 _cleanup_strv_free_
char **empty_directories
= NULL
;
3174 const char *tmp_dir
= NULL
, *var_tmp_dir
= NULL
;
3175 const char *root_dir
= NULL
, *root_image
= NULL
;
3176 _cleanup_free_
char *creds_path
= NULL
, *incoming_dir
= NULL
, *propagate_dir
= NULL
;
3177 NamespaceInfo ns_info
;
3178 bool needs_sandboxing
;
3179 BindMount
*bind_mounts
= NULL
;
3180 size_t n_bind_mounts
= 0;
3185 if (params
->flags
& EXEC_APPLY_CHROOT
) {
3186 root_image
= context
->root_image
;
3189 root_dir
= context
->root_directory
;
3192 r
= compile_bind_mounts(context
, params
, &bind_mounts
, &n_bind_mounts
, &empty_directories
);
3196 needs_sandboxing
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && !(command_flags
& EXEC_COMMAND_FULLY_PRIVILEGED
);
3197 if (needs_sandboxing
) {
3198 /* The runtime struct only contains the parent of the private /tmp,
3199 * which is non-accessible to world users. Inside of it there's a /tmp
3200 * that is sticky, and that's the one we want to use here.
3201 * This does not apply when we are using /run/systemd/empty as fallback. */
3203 if (context
->private_tmp
&& runtime
) {
3204 if (streq_ptr(runtime
->tmp_dir
, RUN_SYSTEMD_EMPTY
))
3205 tmp_dir
= runtime
->tmp_dir
;
3206 else if (runtime
->tmp_dir
)
3207 tmp_dir
= strjoina(runtime
->tmp_dir
, "/tmp");
3209 if (streq_ptr(runtime
->var_tmp_dir
, RUN_SYSTEMD_EMPTY
))
3210 var_tmp_dir
= runtime
->var_tmp_dir
;
3211 else if (runtime
->var_tmp_dir
)
3212 var_tmp_dir
= strjoina(runtime
->var_tmp_dir
, "/tmp");
3215 ns_info
= (NamespaceInfo
) {
3216 .ignore_protect_paths
= false,
3217 .private_dev
= context
->private_devices
,
3218 .protect_control_groups
= context
->protect_control_groups
,
3219 .protect_kernel_tunables
= context
->protect_kernel_tunables
,
3220 .protect_kernel_modules
= context
->protect_kernel_modules
,
3221 .protect_kernel_logs
= context
->protect_kernel_logs
,
3222 .protect_hostname
= context
->protect_hostname
,
3223 .mount_apivfs
= exec_context_get_effective_mount_apivfs(context
),
3224 .private_mounts
= context
->private_mounts
,
3225 .protect_home
= context
->protect_home
,
3226 .protect_system
= context
->protect_system
,
3227 .protect_proc
= context
->protect_proc
,
3228 .proc_subset
= context
->proc_subset
,
3229 .private_ipc
= context
->private_ipc
|| context
->ipc_namespace_path
,
3230 /* If NNP is on, we can turn on MS_NOSUID, since it won't have any effect anymore. */
3231 .mount_nosuid
= context
->no_new_privileges
&& !mac_selinux_use(),
3233 } else if (!context
->dynamic_user
&& root_dir
)
3235 * If DynamicUser=no and RootDirectory= is set then lets pass a relaxed
3236 * sandbox info, otherwise enforce it, don't ignore protected paths and
3237 * fail if we are enable to apply the sandbox inside the mount namespace.
3239 ns_info
= (NamespaceInfo
) {
3240 .ignore_protect_paths
= true,
3243 ns_info
= (NamespaceInfo
) {};
3245 if (context
->mount_flags
== MS_SHARED
)
3246 log_unit_debug(u
, "shared mount propagation hidden by other fs namespacing unit settings: ignoring");
3248 if (exec_context_has_credentials(context
) &&
3249 params
->prefix
[EXEC_DIRECTORY_RUNTIME
] &&
3250 FLAGS_SET(params
->flags
, EXEC_WRITE_CREDENTIALS
)) {
3251 creds_path
= path_join(params
->prefix
[EXEC_DIRECTORY_RUNTIME
], "credentials", u
->id
);
3258 if (MANAGER_IS_SYSTEM(u
->manager
)) {
3259 propagate_dir
= path_join("/run/systemd/propagate/", u
->id
);
3260 if (!propagate_dir
) {
3265 incoming_dir
= strdup("/run/systemd/incoming");
3266 if (!incoming_dir
) {
3272 r
= setup_namespace(root_dir
, root_image
, context
->root_image_options
,
3273 &ns_info
, context
->read_write_paths
,
3274 needs_sandboxing
? context
->read_only_paths
: NULL
,
3275 needs_sandboxing
? context
->inaccessible_paths
: NULL
,
3276 needs_sandboxing
? context
->exec_paths
: NULL
,
3277 needs_sandboxing
? context
->no_exec_paths
: NULL
,
3281 context
->temporary_filesystems
,
3282 context
->n_temporary_filesystems
,
3283 context
->mount_images
,
3284 context
->n_mount_images
,
3288 context
->log_namespace
,
3289 context
->mount_flags
,
3290 context
->root_hash
, context
->root_hash_size
, context
->root_hash_path
,
3291 context
->root_hash_sig
, context
->root_hash_sig_size
, context
->root_hash_sig_path
,
3292 context
->root_verity
,
3293 context
->extension_images
,
3294 context
->n_extension_images
,
3297 root_dir
|| root_image
? params
->notify_socket
: NULL
,
3300 /* If we couldn't set up the namespace this is probably due to a missing capability. setup_namespace() reports
3301 * that with a special, recognizable error ENOANO. In this case, silently proceed, but only if exclusively
3302 * sandboxing options were used, i.e. nothing such as RootDirectory= or BindMount= that would result in a
3303 * completely different execution environment. */
3305 if (insist_on_sandboxing(
3307 root_dir
, root_image
,
3310 log_unit_debug(u
, "Failed to set up namespace, and refusing to continue since the selected namespacing options alter mount environment non-trivially.\n"
3311 "Bind mounts: %zu, temporary filesystems: %zu, root directory: %s, root image: %s, dynamic user: %s",
3312 n_bind_mounts
, context
->n_temporary_filesystems
, yes_no(root_dir
), yes_no(root_image
), yes_no(context
->dynamic_user
));
3316 log_unit_debug(u
, "Failed to set up namespace, assuming containerized execution and ignoring.");
3322 bind_mount_free_many(bind_mounts
, n_bind_mounts
);
3326 static int apply_working_directory(
3327 const ExecContext
*context
,
3328 const ExecParameters
*params
,
3335 assert(exit_status
);
3337 if (context
->working_directory_home
) {
3340 *exit_status
= EXIT_CHDIR
;
3347 wd
= empty_to_root(context
->working_directory
);
3349 if (params
->flags
& EXEC_APPLY_CHROOT
)
3352 d
= prefix_roota(context
->root_directory
, wd
);
3354 if (chdir(d
) < 0 && !context
->working_directory_missing_ok
) {
3355 *exit_status
= EXIT_CHDIR
;
3362 static int apply_root_directory(
3363 const ExecContext
*context
,
3364 const ExecParameters
*params
,
3365 const bool needs_mount_ns
,
3369 assert(exit_status
);
3371 if (params
->flags
& EXEC_APPLY_CHROOT
)
3372 if (!needs_mount_ns
&& context
->root_directory
)
3373 if (chroot(context
->root_directory
) < 0) {
3374 *exit_status
= EXIT_CHROOT
;
3381 static int setup_keyring(
3383 const ExecContext
*context
,
3384 const ExecParameters
*p
,
3385 uid_t uid
, gid_t gid
) {
3387 key_serial_t keyring
;
3396 /* Let's set up a new per-service "session" kernel keyring for each system service. This has the benefit that
3397 * each service runs with its own keyring shared among all processes of the service, but with no hook-up beyond
3398 * that scope, and in particular no link to the per-UID keyring. If we don't do this the keyring will be
3399 * automatically created on-demand and then linked to the per-UID keyring, by the kernel. The kernel's built-in
3400 * on-demand behaviour is very appropriate for login users, but probably not so much for system services, where
3401 * UIDs are not necessarily specific to a service but reused (at least in the case of UID 0). */
3403 if (context
->keyring_mode
== EXEC_KEYRING_INHERIT
)
3406 /* Acquiring a reference to the user keyring is nasty. We briefly change identity in order to get things set up
3407 * properly by the kernel. If we don't do that then we can't create it atomically, and that sucks for parallel
3408 * execution. This mimics what pam_keyinit does, too. Setting up session keyring, to be owned by the right user
3409 * & group is just as nasty as acquiring a reference to the user keyring. */
3411 saved_uid
= getuid();
3412 saved_gid
= getgid();
3414 if (gid_is_valid(gid
) && gid
!= saved_gid
) {
3415 if (setregid(gid
, -1) < 0)
3416 return log_unit_error_errno(u
, errno
, "Failed to change GID for user keyring: %m");
3419 if (uid_is_valid(uid
) && uid
!= saved_uid
) {
3420 if (setreuid(uid
, -1) < 0) {
3421 r
= log_unit_error_errno(u
, errno
, "Failed to change UID for user keyring: %m");
3426 keyring
= keyctl(KEYCTL_JOIN_SESSION_KEYRING
, 0, 0, 0, 0);
3427 if (keyring
== -1) {
3428 if (errno
== ENOSYS
)
3429 log_unit_debug_errno(u
, errno
, "Kernel keyring not supported, ignoring.");
3430 else if (ERRNO_IS_PRIVILEGE(errno
))
3431 log_unit_debug_errno(u
, errno
, "Kernel keyring access prohibited, ignoring.");
3432 else if (errno
== EDQUOT
)
3433 log_unit_debug_errno(u
, errno
, "Out of kernel keyrings to allocate, ignoring.");
3435 r
= log_unit_error_errno(u
, errno
, "Setting up kernel keyring failed: %m");
3440 /* When requested link the user keyring into the session keyring. */
3441 if (context
->keyring_mode
== EXEC_KEYRING_SHARED
) {
3443 if (keyctl(KEYCTL_LINK
,
3444 KEY_SPEC_USER_KEYRING
,
3445 KEY_SPEC_SESSION_KEYRING
, 0, 0) < 0) {
3446 r
= log_unit_error_errno(u
, errno
, "Failed to link user keyring into session keyring: %m");
3451 /* Restore uid/gid back */
3452 if (uid_is_valid(uid
) && uid
!= saved_uid
) {
3453 if (setreuid(saved_uid
, -1) < 0) {
3454 r
= log_unit_error_errno(u
, errno
, "Failed to change UID back for user keyring: %m");
3459 if (gid_is_valid(gid
) && gid
!= saved_gid
) {
3460 if (setregid(saved_gid
, -1) < 0)
3461 return log_unit_error_errno(u
, errno
, "Failed to change GID back for user keyring: %m");
3464 /* Populate they keyring with the invocation ID by default, as original saved_uid. */
3465 if (!sd_id128_is_null(u
->invocation_id
)) {
3468 key
= add_key("user", "invocation_id", &u
->invocation_id
, sizeof(u
->invocation_id
), KEY_SPEC_SESSION_KEYRING
);
3470 log_unit_debug_errno(u
, errno
, "Failed to add invocation ID to keyring, ignoring: %m");
3472 if (keyctl(KEYCTL_SETPERM
, key
,
3473 KEY_POS_VIEW
|KEY_POS_READ
|KEY_POS_SEARCH
|
3474 KEY_USR_VIEW
|KEY_USR_READ
|KEY_USR_SEARCH
, 0, 0) < 0)
3475 r
= log_unit_error_errno(u
, errno
, "Failed to restrict invocation ID permission: %m");
3480 /* Revert back uid & gid for the last time, and exit */
3481 /* no extra logging, as only the first already reported error matters */
3482 if (getuid() != saved_uid
)
3483 (void) setreuid(saved_uid
, -1);
3485 if (getgid() != saved_gid
)
3486 (void) setregid(saved_gid
, -1);
3491 static void append_socket_pair(int *array
, size_t *n
, const int pair
[static 2]) {
3497 array
[(*n
)++] = pair
[0];
3499 array
[(*n
)++] = pair
[1];
3502 static int close_remaining_fds(
3503 const ExecParameters
*params
,
3504 const ExecRuntime
*runtime
,
3505 const DynamicCreds
*dcreds
,
3508 const int *fds
, size_t n_fds
) {
3510 size_t n_dont_close
= 0;
3511 int dont_close
[n_fds
+ 12];
3515 if (params
->stdin_fd
>= 0)
3516 dont_close
[n_dont_close
++] = params
->stdin_fd
;
3517 if (params
->stdout_fd
>= 0)
3518 dont_close
[n_dont_close
++] = params
->stdout_fd
;
3519 if (params
->stderr_fd
>= 0)
3520 dont_close
[n_dont_close
++] = params
->stderr_fd
;
3523 dont_close
[n_dont_close
++] = socket_fd
;
3525 memcpy(dont_close
+ n_dont_close
, fds
, sizeof(int) * n_fds
);
3526 n_dont_close
+= n_fds
;
3530 append_socket_pair(dont_close
, &n_dont_close
, runtime
->netns_storage_socket
);
3531 append_socket_pair(dont_close
, &n_dont_close
, runtime
->ipcns_storage_socket
);
3536 append_socket_pair(dont_close
, &n_dont_close
, dcreds
->user
->storage_socket
);
3538 append_socket_pair(dont_close
, &n_dont_close
, dcreds
->group
->storage_socket
);
3541 if (user_lookup_fd
>= 0)
3542 dont_close
[n_dont_close
++] = user_lookup_fd
;
3544 return close_all_fds(dont_close
, n_dont_close
);
3547 static int send_user_lookup(
3555 /* Send the resolved UID/GID to PID 1 after we learnt it. We send a single datagram, containing the UID/GID
3556 * data as well as the unit name. Note that we suppress sending this if no user/group to resolve was
3559 if (user_lookup_fd
< 0)
3562 if (!uid_is_valid(uid
) && !gid_is_valid(gid
))
3565 if (writev(user_lookup_fd
,
3567 IOVEC_INIT(&uid
, sizeof(uid
)),
3568 IOVEC_INIT(&gid
, sizeof(gid
)),
3569 IOVEC_INIT_STRING(unit
->id
) }, 3) < 0)
3575 static int acquire_home(const ExecContext
*c
, uid_t uid
, const char** home
, char **buf
) {
3582 /* If WorkingDirectory=~ is set, try to acquire a usable home directory. */
3587 if (!c
->working_directory_home
)
3590 r
= get_home_dir(buf
);
3598 static int compile_suggested_paths(const ExecContext
*c
, const ExecParameters
*p
, char ***ret
) {
3599 _cleanup_strv_free_
char ** list
= NULL
;
3606 assert(c
->dynamic_user
);
3608 /* Compile a list of paths that it might make sense to read the owning UID from to use as initial candidate for
3609 * dynamic UID allocation, in order to save us from doing costly recursive chown()s of the special
3612 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
3615 if (t
== EXEC_DIRECTORY_CONFIGURATION
)
3621 STRV_FOREACH(i
, c
->directories
[t
].paths
) {
3624 if (exec_directory_is_private(c
, t
))
3625 e
= path_join(p
->prefix
[t
], "private", *i
);
3627 e
= path_join(p
->prefix
[t
], *i
);
3631 r
= strv_consume(&list
, e
);
3637 *ret
= TAKE_PTR(list
);
3642 static int exec_parameters_get_cgroup_path(const ExecParameters
*params
, char **ret
) {
3643 bool using_subcgroup
;
3649 if (!params
->cgroup_path
)
3652 /* If we are called for a unit where cgroup delegation is on, and the payload created its own populated
3653 * subcgroup (which we expect it to do, after all it asked for delegation), then we cannot place the control
3654 * processes started after the main unit's process in the unit's main cgroup because it is now an inner one,
3655 * and inner cgroups may not contain processes. Hence, if delegation is on, and this is a control process,
3656 * let's use ".control" as subcgroup instead. Note that we do so only for ExecStartPost=, ExecReload=,
3657 * ExecStop=, ExecStopPost=, i.e. for the commands where the main process is already forked. For ExecStartPre=
3658 * this is not necessary, the cgroup is still empty. We distinguish these cases with the EXEC_CONTROL_CGROUP
3659 * flag, which is only passed for the former statements, not for the latter. */
3661 using_subcgroup
= FLAGS_SET(params
->flags
, EXEC_CONTROL_CGROUP
|EXEC_CGROUP_DELEGATE
|EXEC_IS_CONTROL
);
3662 if (using_subcgroup
)
3663 p
= path_join(params
->cgroup_path
, ".control");
3665 p
= strdup(params
->cgroup_path
);
3670 return using_subcgroup
;
3673 static int exec_context_cpu_affinity_from_numa(const ExecContext
*c
, CPUSet
*ret
) {
3674 _cleanup_(cpu_set_reset
) CPUSet s
= {};
3680 if (!c
->numa_policy
.nodes
.set
) {
3681 log_debug("Can't derive CPU affinity mask from NUMA mask because NUMA mask is not set, ignoring");
3685 r
= numa_to_cpu_set(&c
->numa_policy
, &s
);
3691 return cpu_set_add_all(ret
, &s
);
3694 bool exec_context_get_cpu_affinity_from_numa(const ExecContext
*c
) {
3697 return c
->cpu_affinity_from_numa
;
3700 static int add_shifted_fd(int *fds
, size_t fds_size
, size_t *n_fds
, int fd
, int *ret_fd
) {
3705 assert(*n_fds
< fds_size
);
3713 if (fd
< 3 + (int) *n_fds
) {
3714 /* Let's move the fd up, so that it's outside of the fd range we will use to store
3715 * the fds we pass to the process (or which are closed only during execve). */
3717 r
= fcntl(fd
, F_DUPFD_CLOEXEC
, 3 + (int) *n_fds
);
3721 CLOSE_AND_REPLACE(fd
, r
);
3724 *ret_fd
= fds
[*n_fds
] = fd
;
3729 static int exec_child(
3731 const ExecCommand
*command
,
3732 const ExecContext
*context
,
3733 const ExecParameters
*params
,
3734 ExecRuntime
*runtime
,
3735 DynamicCreds
*dcreds
,
3737 const int named_iofds
[static 3],
3739 size_t n_socket_fds
,
3740 size_t n_storage_fds
,
3745 _cleanup_strv_free_
char **our_env
= NULL
, **pass_env
= NULL
, **accum_env
= NULL
, **replaced_argv
= NULL
;
3746 int r
, ngids
= 0, exec_fd
;
3747 _cleanup_free_ gid_t
*supplementary_gids
= NULL
;
3748 const char *username
= NULL
, *groupname
= NULL
;
3749 _cleanup_free_
char *home_buffer
= NULL
;
3750 const char *home
= NULL
, *shell
= NULL
;
3751 char **final_argv
= NULL
;
3752 dev_t journal_stream_dev
= 0;
3753 ino_t journal_stream_ino
= 0;
3754 bool userns_set_up
= false;
3755 bool needs_sandboxing
, /* Do we need to set up full sandboxing? (i.e. all namespacing, all MAC stuff, caps, yadda yadda */
3756 needs_setuid
, /* Do we need to do the actual setresuid()/setresgid() calls? */
3757 needs_mount_namespace
, /* Do we need to set up a mount namespace for this kernel? */
3758 needs_ambient_hack
; /* Do we need to apply the ambient capabilities hack? */
3760 _cleanup_free_
char *mac_selinux_context_net
= NULL
;
3761 bool use_selinux
= false;
3764 bool use_smack
= false;
3767 bool use_apparmor
= false;
3769 uid_t saved_uid
= getuid();
3770 gid_t saved_gid
= getgid();
3771 uid_t uid
= UID_INVALID
;
3772 gid_t gid
= GID_INVALID
;
3773 size_t n_fds
= n_socket_fds
+ n_storage_fds
, /* fds to pass to the child */
3774 n_keep_fds
; /* total number of fds not to close */
3776 _cleanup_free_ gid_t
*gids_after_pam
= NULL
;
3777 int ngids_after_pam
= 0;
3783 assert(exit_status
);
3785 rename_process_from_path(command
->path
);
3787 /* We reset exactly these signals, since they are the only ones we set to SIG_IGN in the main
3788 * daemon. All others we leave untouched because we set them to SIG_DFL or a valid handler initially,
3789 * both of which will be demoted to SIG_DFL. */
3790 (void) default_signals(SIGNALS_CRASH_HANDLER
,
3793 if (context
->ignore_sigpipe
)
3794 (void) ignore_signals(SIGPIPE
);
3796 r
= reset_signal_mask();
3798 *exit_status
= EXIT_SIGNAL_MASK
;
3799 return log_unit_error_errno(unit
, r
, "Failed to set process signal mask: %m");
3802 if (params
->idle_pipe
)
3803 do_idle_pipe_dance(params
->idle_pipe
);
3805 /* Close fds we don't need very early to make sure we don't block init reexecution because it cannot bind its
3806 * sockets. Among the fds we close are the logging fds, and we want to keep them closed, so that we don't have
3807 * any fds open we don't really want open during the transition. In order to make logging work, we switch the
3808 * log subsystem into open_when_needed mode, so that it reopens the logs on every single log call. */
3811 log_set_open_when_needed(true);
3813 /* In case anything used libc syslog(), close this here, too */
3816 int keep_fds
[n_fds
+ 2];
3817 memcpy_safe(keep_fds
, fds
, n_fds
* sizeof(int));
3820 r
= add_shifted_fd(keep_fds
, ELEMENTSOF(keep_fds
), &n_keep_fds
, params
->exec_fd
, &exec_fd
);
3822 *exit_status
= EXIT_FDS
;
3823 return log_unit_error_errno(unit
, r
, "Failed to shift fd and set FD_CLOEXEC: %m");
3826 r
= close_remaining_fds(params
, runtime
, dcreds
, user_lookup_fd
, socket_fd
, keep_fds
, n_keep_fds
);
3828 *exit_status
= EXIT_FDS
;
3829 return log_unit_error_errno(unit
, r
, "Failed to close unwanted file descriptors: %m");
3832 if (!context
->same_pgrp
&&
3834 *exit_status
= EXIT_SETSID
;
3835 return log_unit_error_errno(unit
, errno
, "Failed to create new process session: %m");
3838 exec_context_tty_reset(context
, params
);
3840 if (unit_shall_confirm_spawn(unit
)) {
3841 const char *vc
= params
->confirm_spawn
;
3842 _cleanup_free_
char *cmdline
= NULL
;
3844 cmdline
= quote_command_line(command
->argv
);
3846 *exit_status
= EXIT_MEMORY
;
3850 r
= ask_for_confirmation(vc
, unit
, cmdline
);
3851 if (r
!= CONFIRM_EXECUTE
) {
3852 if (r
== CONFIRM_PRETEND_SUCCESS
) {
3853 *exit_status
= EXIT_SUCCESS
;
3856 *exit_status
= EXIT_CONFIRM
;
3857 return log_unit_error_errno(unit
, SYNTHETIC_ERRNO(ECANCELED
),
3858 "Execution cancelled by the user");
3862 /* We are about to invoke NSS and PAM modules. Let's tell them what we are doing here, maybe they care. This is
3863 * used by nss-resolve to disable itself when we are about to start systemd-resolved, to avoid deadlocks. Note
3864 * that these env vars do not survive the execve(), which means they really only apply to the PAM and NSS
3865 * invocations themselves. Also note that while we'll only invoke NSS modules involved in user management they
3866 * might internally call into other NSS modules that are involved in hostname resolution, we never know. */
3867 if (setenv("SYSTEMD_ACTIVATION_UNIT", unit
->id
, true) != 0 ||
3868 setenv("SYSTEMD_ACTIVATION_SCOPE", MANAGER_IS_SYSTEM(unit
->manager
) ? "system" : "user", true) != 0) {
3869 *exit_status
= EXIT_MEMORY
;
3870 return log_unit_error_errno(unit
, errno
, "Failed to update environment: %m");
3873 if (context
->dynamic_user
&& dcreds
) {
3874 _cleanup_strv_free_
char **suggested_paths
= NULL
;
3876 /* On top of that, make sure we bypass our own NSS module nss-systemd comprehensively for any NSS
3877 * checks, if DynamicUser=1 is used, as we shouldn't create a feedback loop with ourselves here. */
3878 if (putenv((char*) "SYSTEMD_NSS_DYNAMIC_BYPASS=1") != 0) {
3879 *exit_status
= EXIT_USER
;
3880 return log_unit_error_errno(unit
, errno
, "Failed to update environment: %m");
3883 r
= compile_suggested_paths(context
, params
, &suggested_paths
);
3885 *exit_status
= EXIT_MEMORY
;
3889 r
= dynamic_creds_realize(dcreds
, suggested_paths
, &uid
, &gid
);
3891 *exit_status
= EXIT_USER
;
3893 return log_unit_error_errno(unit
, SYNTHETIC_ERRNO(EOPNOTSUPP
),
3894 "Failed to update dynamic user credentials: User or group with specified name already exists.");
3895 return log_unit_error_errno(unit
, r
, "Failed to update dynamic user credentials: %m");
3898 if (!uid_is_valid(uid
)) {
3899 *exit_status
= EXIT_USER
;
3900 return log_unit_error_errno(unit
, SYNTHETIC_ERRNO(ESRCH
), "UID validation failed for \""UID_FMT
"\"", uid
);
3903 if (!gid_is_valid(gid
)) {
3904 *exit_status
= EXIT_USER
;
3905 return log_unit_error_errno(unit
, SYNTHETIC_ERRNO(ESRCH
), "GID validation failed for \""GID_FMT
"\"", gid
);
3909 username
= dcreds
->user
->name
;
3912 r
= get_fixed_user(context
, &username
, &uid
, &gid
, &home
, &shell
);
3914 *exit_status
= EXIT_USER
;
3915 return log_unit_error_errno(unit
, r
, "Failed to determine user credentials: %m");
3918 r
= get_fixed_group(context
, &groupname
, &gid
);
3920 *exit_status
= EXIT_GROUP
;
3921 return log_unit_error_errno(unit
, r
, "Failed to determine group credentials: %m");
3925 /* Initialize user supplementary groups and get SupplementaryGroups= ones */
3926 r
= get_supplementary_groups(context
, username
, groupname
, gid
,
3927 &supplementary_gids
, &ngids
);
3929 *exit_status
= EXIT_GROUP
;
3930 return log_unit_error_errno(unit
, r
, "Failed to determine supplementary groups: %m");
3933 r
= send_user_lookup(unit
, user_lookup_fd
, uid
, gid
);
3935 *exit_status
= EXIT_USER
;
3936 return log_unit_error_errno(unit
, r
, "Failed to send user credentials to PID1: %m");
3939 user_lookup_fd
= safe_close(user_lookup_fd
);
3941 r
= acquire_home(context
, uid
, &home
, &home_buffer
);
3943 *exit_status
= EXIT_CHDIR
;
3944 return log_unit_error_errno(unit
, r
, "Failed to determine $HOME for user: %m");
3947 /* If a socket is connected to STDIN/STDOUT/STDERR, we
3948 * must sure to drop O_NONBLOCK */
3950 (void) fd_nonblock(socket_fd
, false);
3952 /* Journald will try to look-up our cgroup in order to populate _SYSTEMD_CGROUP and _SYSTEMD_UNIT fields.
3953 * Hence we need to migrate to the target cgroup from init.scope before connecting to journald */
3954 if (params
->cgroup_path
) {
3955 _cleanup_free_
char *p
= NULL
;
3957 r
= exec_parameters_get_cgroup_path(params
, &p
);
3959 *exit_status
= EXIT_CGROUP
;
3960 return log_unit_error_errno(unit
, r
, "Failed to acquire cgroup path: %m");
3963 r
= cg_attach_everywhere(params
->cgroup_supported
, p
, 0, NULL
, NULL
);
3965 *exit_status
= EXIT_CGROUP
;
3966 return log_unit_error_errno(unit
, r
, "Failed to attach to cgroup %s: %m", p
);
3970 if (context
->network_namespace_path
&& runtime
&& runtime
->netns_storage_socket
[0] >= 0) {
3971 r
= open_shareable_ns_path(runtime
->netns_storage_socket
, context
->network_namespace_path
, CLONE_NEWNET
);
3973 *exit_status
= EXIT_NETWORK
;
3974 return log_unit_error_errno(unit
, r
, "Failed to open network namespace path %s: %m", context
->network_namespace_path
);
3978 if (context
->ipc_namespace_path
&& runtime
&& runtime
->ipcns_storage_socket
[0] >= 0) {
3979 r
= open_shareable_ns_path(runtime
->ipcns_storage_socket
, context
->ipc_namespace_path
, CLONE_NEWIPC
);
3981 *exit_status
= EXIT_NAMESPACE
;
3982 return log_unit_error_errno(unit
, r
, "Failed to open IPC namespace path %s: %m", context
->ipc_namespace_path
);
3986 r
= setup_input(context
, params
, socket_fd
, named_iofds
);
3988 *exit_status
= EXIT_STDIN
;
3989 return log_unit_error_errno(unit
, r
, "Failed to set up standard input: %m");
3992 r
= setup_output(unit
, context
, params
, STDOUT_FILENO
, socket_fd
, named_iofds
, basename(command
->path
), uid
, gid
, &journal_stream_dev
, &journal_stream_ino
);
3994 *exit_status
= EXIT_STDOUT
;
3995 return log_unit_error_errno(unit
, r
, "Failed to set up standard output: %m");
3998 r
= setup_output(unit
, context
, params
, STDERR_FILENO
, socket_fd
, named_iofds
, basename(command
->path
), uid
, gid
, &journal_stream_dev
, &journal_stream_ino
);
4000 *exit_status
= EXIT_STDERR
;
4001 return log_unit_error_errno(unit
, r
, "Failed to set up standard error output: %m");
4004 if (context
->oom_score_adjust_set
) {
4005 /* When we can't make this change due to EPERM, then let's silently skip over it. User namespaces
4006 * prohibit write access to this file, and we shouldn't trip up over that. */
4007 r
= set_oom_score_adjust(context
->oom_score_adjust
);
4008 if (ERRNO_IS_PRIVILEGE(r
))
4009 log_unit_debug_errno(unit
, r
, "Failed to adjust OOM setting, assuming containerized execution, ignoring: %m");
4011 *exit_status
= EXIT_OOM_ADJUST
;
4012 return log_unit_error_errno(unit
, r
, "Failed to adjust OOM setting: %m");
4016 if (context
->coredump_filter_set
) {
4017 r
= set_coredump_filter(context
->coredump_filter
);
4018 if (ERRNO_IS_PRIVILEGE(r
))
4019 log_unit_debug_errno(unit
, r
, "Failed to adjust coredump_filter, ignoring: %m");
4021 return log_unit_error_errno(unit
, r
, "Failed to adjust coredump_filter: %m");
4024 if (context
->nice_set
) {
4025 r
= setpriority_closest(context
->nice
);
4027 return log_unit_error_errno(unit
, r
, "Failed to set up process scheduling priority (nice level): %m");
4030 if (context
->cpu_sched_set
) {
4031 struct sched_param param
= {
4032 .sched_priority
= context
->cpu_sched_priority
,
4035 r
= sched_setscheduler(0,
4036 context
->cpu_sched_policy
|
4037 (context
->cpu_sched_reset_on_fork
?
4038 SCHED_RESET_ON_FORK
: 0),
4041 *exit_status
= EXIT_SETSCHEDULER
;
4042 return log_unit_error_errno(unit
, errno
, "Failed to set up CPU scheduling: %m");
4046 if (context
->cpu_affinity_from_numa
|| context
->cpu_set
.set
) {
4047 _cleanup_(cpu_set_reset
) CPUSet converted_cpu_set
= {};
4048 const CPUSet
*cpu_set
;
4050 if (context
->cpu_affinity_from_numa
) {
4051 r
= exec_context_cpu_affinity_from_numa(context
, &converted_cpu_set
);
4053 *exit_status
= EXIT_CPUAFFINITY
;
4054 return log_unit_error_errno(unit
, r
, "Failed to derive CPU affinity mask from NUMA mask: %m");
4057 cpu_set
= &converted_cpu_set
;
4059 cpu_set
= &context
->cpu_set
;
4061 if (sched_setaffinity(0, cpu_set
->allocated
, cpu_set
->set
) < 0) {
4062 *exit_status
= EXIT_CPUAFFINITY
;
4063 return log_unit_error_errno(unit
, errno
, "Failed to set up CPU affinity: %m");
4067 if (mpol_is_valid(numa_policy_get_type(&context
->numa_policy
))) {
4068 r
= apply_numa_policy(&context
->numa_policy
);
4069 if (r
== -EOPNOTSUPP
)
4070 log_unit_debug_errno(unit
, r
, "NUMA support not available, ignoring.");
4072 *exit_status
= EXIT_NUMA_POLICY
;
4073 return log_unit_error_errno(unit
, r
, "Failed to set NUMA memory policy: %m");
4077 if (context
->ioprio_set
)
4078 if (ioprio_set(IOPRIO_WHO_PROCESS
, 0, context
->ioprio
) < 0) {
4079 *exit_status
= EXIT_IOPRIO
;
4080 return log_unit_error_errno(unit
, errno
, "Failed to set up IO scheduling priority: %m");
4083 if (context
->timer_slack_nsec
!= NSEC_INFINITY
)
4084 if (prctl(PR_SET_TIMERSLACK
, context
->timer_slack_nsec
) < 0) {
4085 *exit_status
= EXIT_TIMERSLACK
;
4086 return log_unit_error_errno(unit
, errno
, "Failed to set up timer slack: %m");
4089 if (context
->personality
!= PERSONALITY_INVALID
) {
4090 r
= safe_personality(context
->personality
);
4092 *exit_status
= EXIT_PERSONALITY
;
4093 return log_unit_error_errno(unit
, r
, "Failed to set up execution domain (personality): %m");
4097 if (context
->utmp_id
)
4098 utmp_put_init_process(context
->utmp_id
, getpid_cached(), getsid(0),
4100 context
->utmp_mode
== EXEC_UTMP_INIT
? INIT_PROCESS
:
4101 context
->utmp_mode
== EXEC_UTMP_LOGIN
? LOGIN_PROCESS
:
4105 if (uid_is_valid(uid
)) {
4106 r
= chown_terminal(STDIN_FILENO
, uid
);
4108 *exit_status
= EXIT_STDIN
;
4109 return log_unit_error_errno(unit
, r
, "Failed to change ownership of terminal: %m");
4113 /* If delegation is enabled we'll pass ownership of the cgroup to the user of the new process. On cgroup v1
4114 * this is only about systemd's own hierarchy, i.e. not the controller hierarchies, simply because that's not
4115 * safe. On cgroup v2 there's only one hierarchy anyway, and delegation is safe there, hence in that case only
4116 * touch a single hierarchy too. */
4117 if (params
->cgroup_path
&& context
->user
&& (params
->flags
& EXEC_CGROUP_DELEGATE
)) {
4118 r
= cg_set_access(SYSTEMD_CGROUP_CONTROLLER
, params
->cgroup_path
, uid
, gid
);
4120 *exit_status
= EXIT_CGROUP
;
4121 return log_unit_error_errno(unit
, r
, "Failed to adjust control group access: %m");
4125 for (ExecDirectoryType dt
= 0; dt
< _EXEC_DIRECTORY_TYPE_MAX
; dt
++) {
4126 r
= setup_exec_directory(context
, params
, uid
, gid
, dt
, exit_status
);
4128 return log_unit_error_errno(unit
, r
, "Failed to set up special execution directory in %s: %m", params
->prefix
[dt
]);
4131 if (FLAGS_SET(params
->flags
, EXEC_WRITE_CREDENTIALS
)) {
4132 r
= setup_credentials(context
, params
, unit
->id
, uid
);
4134 *exit_status
= EXIT_CREDENTIALS
;
4135 return log_unit_error_errno(unit
, r
, "Failed to set up credentials: %m");
4139 r
= build_environment(
4151 *exit_status
= EXIT_MEMORY
;
4155 r
= build_pass_environment(context
, &pass_env
);
4157 *exit_status
= EXIT_MEMORY
;
4161 accum_env
= strv_env_merge(params
->environment
,
4164 context
->environment
,
4167 *exit_status
= EXIT_MEMORY
;
4170 accum_env
= strv_env_clean(accum_env
);
4172 (void) umask(context
->umask
);
4174 r
= setup_keyring(unit
, context
, params
, uid
, gid
);
4176 *exit_status
= EXIT_KEYRING
;
4177 return log_unit_error_errno(unit
, r
, "Failed to set up kernel keyring: %m");
4180 /* We need sandboxing if the caller asked us to apply it and the command isn't explicitly excepted from it */
4181 needs_sandboxing
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && !(command
->flags
& EXEC_COMMAND_FULLY_PRIVILEGED
);
4183 /* 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 */
4184 needs_ambient_hack
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && (command
->flags
& EXEC_COMMAND_AMBIENT_MAGIC
) && !ambient_capabilities_supported();
4186 /* 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 */
4187 if (needs_ambient_hack
)
4188 needs_setuid
= false;
4190 needs_setuid
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && !(command
->flags
& (EXEC_COMMAND_FULLY_PRIVILEGED
|EXEC_COMMAND_NO_SETUID
));
4192 if (needs_sandboxing
) {
4193 /* MAC enablement checks need to be done before a new mount ns is created, as they rely on /sys being
4194 * present. The actual MAC context application will happen later, as late as possible, to avoid
4195 * impacting our own code paths. */
4198 use_selinux
= mac_selinux_use();
4201 use_smack
= mac_smack_use();
4204 use_apparmor
= mac_apparmor_use();
4208 if (needs_sandboxing
) {
4211 /* Let's set the resource limits before we call into PAM, so that pam_limits wins over what
4212 * is set here. (See below.) */
4214 r
= setrlimit_closest_all((const struct rlimit
* const *) context
->rlimit
, &which_failed
);
4216 *exit_status
= EXIT_LIMITS
;
4217 return log_unit_error_errno(unit
, r
, "Failed to adjust resource limit RLIMIT_%s: %m", rlimit_to_string(which_failed
));
4221 if (needs_setuid
&& context
->pam_name
&& username
) {
4222 /* Let's call into PAM after we set up our own idea of resource limits to that pam_limits
4223 * wins here. (See above.) */
4225 /* All fds passed in the fds array will be closed in the pam child process. */
4226 r
= setup_pam(context
->pam_name
, username
, uid
, gid
, context
->tty_path
, &accum_env
, fds
, n_fds
);
4228 *exit_status
= EXIT_PAM
;
4229 return log_unit_error_errno(unit
, r
, "Failed to set up PAM session: %m");
4232 ngids_after_pam
= getgroups_alloc(&gids_after_pam
);
4233 if (ngids_after_pam
< 0) {
4234 *exit_status
= EXIT_MEMORY
;
4235 return log_unit_error_errno(unit
, ngids_after_pam
, "Failed to obtain groups after setting up PAM: %m");
4239 if (needs_sandboxing
&& context
->private_users
&& !have_effective_cap(CAP_SYS_ADMIN
)) {
4240 /* If we're unprivileged, set up the user namespace first to enable use of the other namespaces.
4241 * Users with CAP_SYS_ADMIN can set up user namespaces last because they will be able to
4242 * set up the all of the other namespaces (i.e. network, mount, UTS) without a user namespace. */
4244 userns_set_up
= true;
4245 r
= setup_private_users(saved_uid
, saved_gid
, uid
, gid
);
4247 *exit_status
= EXIT_USER
;
4248 return log_unit_error_errno(unit
, r
, "Failed to set up user namespacing for unprivileged user: %m");
4252 if ((context
->private_network
|| context
->network_namespace_path
) && runtime
&& runtime
->netns_storage_socket
[0] >= 0) {
4254 if (ns_type_supported(NAMESPACE_NET
)) {
4255 r
= setup_shareable_ns(runtime
->netns_storage_socket
, CLONE_NEWNET
);
4257 log_unit_warning_errno(unit
, r
,
4258 "PrivateNetwork=yes is configured, but network namespace setup failed, ignoring: %m");
4260 *exit_status
= EXIT_NETWORK
;
4261 return log_unit_error_errno(unit
, r
, "Failed to set up network namespacing: %m");
4263 } else if (context
->network_namespace_path
) {
4264 *exit_status
= EXIT_NETWORK
;
4265 return log_unit_error_errno(unit
, SYNTHETIC_ERRNO(EOPNOTSUPP
),
4266 "NetworkNamespacePath= is not supported, refusing.");
4268 log_unit_warning(unit
, "PrivateNetwork=yes is configured, but the kernel does not support network namespaces, ignoring.");
4271 if ((context
->private_ipc
|| context
->ipc_namespace_path
) && runtime
&& runtime
->ipcns_storage_socket
[0] >= 0) {
4273 if (ns_type_supported(NAMESPACE_IPC
)) {
4274 r
= setup_shareable_ns(runtime
->ipcns_storage_socket
, CLONE_NEWIPC
);
4276 log_unit_warning_errno(unit
, r
,
4277 "PrivateIPC=yes is configured, but IPC namespace setup failed, ignoring: %m");
4279 *exit_status
= EXIT_NAMESPACE
;
4280 return log_unit_error_errno(unit
, r
, "Failed to set up IPC namespacing: %m");
4282 } else if (context
->ipc_namespace_path
) {
4283 *exit_status
= EXIT_NAMESPACE
;
4284 return log_unit_error_errno(unit
, SYNTHETIC_ERRNO(EOPNOTSUPP
),
4285 "IPCNamespacePath= is not supported, refusing.");
4287 log_unit_warning(unit
, "PrivateIPC=yes is configured, but the kernel does not support IPC namespaces, ignoring.");
4290 needs_mount_namespace
= exec_needs_mount_namespace(context
, params
, runtime
);
4291 if (needs_mount_namespace
) {
4292 _cleanup_free_
char *error_path
= NULL
;
4294 r
= apply_mount_namespace(unit
, command
->flags
, context
, params
, runtime
, &error_path
);
4296 *exit_status
= EXIT_NAMESPACE
;
4297 return log_unit_error_errno(unit
, r
, "Failed to set up mount namespacing%s%s: %m",
4298 error_path
? ": " : "", strempty(error_path
));
4302 if (needs_sandboxing
) {
4303 r
= apply_protect_hostname(unit
, context
, exit_status
);
4308 /* Drop groups as early as possible.
4309 * This needs to be done after PrivateDevices=y setup as device nodes should be owned by the host's root.
4310 * For non-root in a userns, devices will be owned by the user/group before the group change, and nobody. */
4312 _cleanup_free_ gid_t
*gids_to_enforce
= NULL
;
4313 int ngids_to_enforce
= 0;
4315 ngids_to_enforce
= merge_gid_lists(supplementary_gids
,
4320 if (ngids_to_enforce
< 0) {
4321 *exit_status
= EXIT_MEMORY
;
4322 return log_unit_error_errno(unit
,
4324 "Failed to merge group lists. Group membership might be incorrect: %m");
4327 r
= enforce_groups(gid
, gids_to_enforce
, ngids_to_enforce
);
4329 *exit_status
= EXIT_GROUP
;
4330 return log_unit_error_errno(unit
, r
, "Changing group credentials failed: %m");
4334 /* If the user namespace was not set up above, try to do it now.
4335 * It's preferred to set up the user namespace later (after all other namespaces) so as not to be
4336 * restricted by rules pertaining to combining user namspaces with other namespaces (e.g. in the
4337 * case of mount namespaces being less privileged when the mount point list is copied from a
4338 * different user namespace). */
4340 if (needs_sandboxing
&& context
->private_users
&& !userns_set_up
) {
4341 r
= setup_private_users(saved_uid
, saved_gid
, uid
, gid
);
4343 *exit_status
= EXIT_USER
;
4344 return log_unit_error_errno(unit
, r
, "Failed to set up user namespacing: %m");
4348 /* Now that the mount namespace has been set up and privileges adjusted, let's look for the thing we
4351 _cleanup_free_
char *executable
= NULL
;
4352 _cleanup_close_
int executable_fd
= -1;
4353 r
= find_executable_full(command
->path
, false, &executable
, &executable_fd
);
4355 if (r
!= -ENOMEM
&& (command
->flags
& EXEC_COMMAND_IGNORE_FAILURE
)) {
4356 log_unit_struct_errno(unit
, LOG_INFO
, r
,
4357 "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR
,
4358 LOG_UNIT_INVOCATION_ID(unit
),
4359 LOG_UNIT_MESSAGE(unit
, "Executable %s missing, skipping: %m",
4361 "EXECUTABLE=%s", command
->path
);
4365 *exit_status
= EXIT_EXEC
;
4367 return log_unit_struct_errno(unit
, LOG_INFO
, r
,
4368 "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR
,
4369 LOG_UNIT_INVOCATION_ID(unit
),
4370 LOG_UNIT_MESSAGE(unit
, "Failed to locate executable %s: %m",
4372 "EXECUTABLE=%s", command
->path
);
4375 r
= add_shifted_fd(keep_fds
, ELEMENTSOF(keep_fds
), &n_keep_fds
, executable_fd
, &executable_fd
);
4377 *exit_status
= EXIT_FDS
;
4378 return log_unit_error_errno(unit
, r
, "Failed to shift fd and set FD_CLOEXEC: %m");
4382 if (needs_sandboxing
&& use_selinux
&& params
->selinux_context_net
) {
4387 else if (params
->n_socket_fds
== 1)
4388 /* If stdin is not connected to a socket but we are triggered by exactly one socket unit then we
4389 * use context from that fd to compute the label. */
4390 fd
= params
->fds
[0];
4393 r
= mac_selinux_get_child_mls_label(fd
, executable
, context
->selinux_context
, &mac_selinux_context_net
);
4395 *exit_status
= EXIT_SELINUX_CONTEXT
;
4396 return log_unit_error_errno(unit
, r
, "Failed to determine SELinux context: %m");
4402 /* We repeat the fd closing here, to make sure that nothing is leaked from the PAM modules. Note that we are
4403 * more aggressive this time since socket_fd and the netns and ipcns fds we don't need anymore. We do keep the exec_fd
4404 * however if we have it as we want to keep it open until the final execve(). */
4406 r
= close_all_fds(keep_fds
, n_keep_fds
);
4408 r
= shift_fds(fds
, n_fds
);
4410 r
= flags_fds(fds
, n_socket_fds
, n_storage_fds
, context
->non_blocking
);
4412 *exit_status
= EXIT_FDS
;
4413 return log_unit_error_errno(unit
, r
, "Failed to adjust passed file descriptors: %m");
4416 /* At this point, the fds we want to pass to the program are all ready and set up, with O_CLOEXEC turned off
4417 * and at the right fd numbers. The are no other fds open, with one exception: the exec_fd if it is defined,
4418 * and it has O_CLOEXEC set, after all we want it to be closed by the execve(), so that our parent knows we
4421 secure_bits
= context
->secure_bits
;
4423 if (needs_sandboxing
) {
4426 /* Set the RTPRIO resource limit to 0, but only if nothing else was explicitly
4427 * requested. (Note this is placed after the general resource limit initialization, see
4428 * above, in order to take precedence.) */
4429 if (context
->restrict_realtime
&& !context
->rlimit
[RLIMIT_RTPRIO
]) {
4430 if (setrlimit(RLIMIT_RTPRIO
, &RLIMIT_MAKE_CONST(0)) < 0) {
4431 *exit_status
= EXIT_LIMITS
;
4432 return log_unit_error_errno(unit
, errno
, "Failed to adjust RLIMIT_RTPRIO resource limit: %m");
4437 /* LSM Smack needs the capability CAP_MAC_ADMIN to change the current execution security context of the
4438 * process. This is the latest place before dropping capabilities. Other MAC context are set later. */
4440 r
= setup_smack(context
, executable_fd
);
4442 *exit_status
= EXIT_SMACK_PROCESS_LABEL
;
4443 return log_unit_error_errno(unit
, r
, "Failed to set SMACK process label: %m");
4448 bset
= context
->capability_bounding_set
;
4449 /* If the ambient caps hack is enabled (which means the kernel can't do them, and the user asked for
4450 * our magic fallback), then let's add some extra caps, so that the service can drop privs of its own,
4451 * instead of us doing that */
4452 if (needs_ambient_hack
)
4453 bset
|= (UINT64_C(1) << CAP_SETPCAP
) |
4454 (UINT64_C(1) << CAP_SETUID
) |
4455 (UINT64_C(1) << CAP_SETGID
);
4457 if (!cap_test_all(bset
)) {
4458 r
= capability_bounding_set_drop(bset
, false);
4460 *exit_status
= EXIT_CAPABILITIES
;
4461 return log_unit_error_errno(unit
, r
, "Failed to drop capabilities: %m");
4465 /* Ambient capabilities are cleared during setresuid() (in enforce_user()) even with
4467 * To be able to raise the ambient capabilities after setresuid() they have to be
4468 * added to the inherited set and keep caps has to be set (done in enforce_user()).
4469 * After setresuid() the ambient capabilities can be raised as they are present in
4470 * the permitted and inhertiable set. However it is possible that someone wants to
4471 * set ambient capabilities without changing the user, so we also set the ambient
4472 * capabilities here.
4473 * The requested ambient capabilities are raised in the inheritable set if the
4474 * second argument is true. */
4475 if (!needs_ambient_hack
) {
4476 r
= capability_ambient_set_apply(context
->capability_ambient_set
, true);
4478 *exit_status
= EXIT_CAPABILITIES
;
4479 return log_unit_error_errno(unit
, r
, "Failed to apply ambient capabilities (before UID change): %m");
4484 /* chroot to root directory first, before we lose the ability to chroot */
4485 r
= apply_root_directory(context
, params
, needs_mount_namespace
, exit_status
);
4487 return log_unit_error_errno(unit
, r
, "Chrooting to the requested root directory failed: %m");
4490 if (uid_is_valid(uid
)) {
4491 r
= enforce_user(context
, uid
);
4493 *exit_status
= EXIT_USER
;
4494 return log_unit_error_errno(unit
, r
, "Failed to change UID to " UID_FMT
": %m", uid
);
4497 if (!needs_ambient_hack
&&
4498 context
->capability_ambient_set
!= 0) {
4500 /* Raise the ambient capabilities after user change. */
4501 r
= capability_ambient_set_apply(context
->capability_ambient_set
, false);
4503 *exit_status
= EXIT_CAPABILITIES
;
4504 return log_unit_error_errno(unit
, r
, "Failed to apply ambient capabilities (after UID change): %m");
4510 /* Apply working directory here, because the working directory might be on NFS and only the user running
4511 * this service might have the correct privilege to change to the working directory */
4512 r
= apply_working_directory(context
, params
, home
, exit_status
);
4514 return log_unit_error_errno(unit
, r
, "Changing to the requested working directory failed: %m");
4516 if (needs_sandboxing
) {
4517 /* Apply other MAC contexts late, but before seccomp syscall filtering, as those should really be last to
4518 * influence our own codepaths as little as possible. Moreover, applying MAC contexts usually requires
4519 * syscalls that are subject to seccomp filtering, hence should probably be applied before the syscalls
4520 * are restricted. */
4524 char *exec_context
= mac_selinux_context_net
?: context
->selinux_context
;
4527 r
= setexeccon(exec_context
);
4529 *exit_status
= EXIT_SELINUX_CONTEXT
;
4530 return log_unit_error_errno(unit
, r
, "Failed to change SELinux context to %s: %m", exec_context
);
4537 if (use_apparmor
&& context
->apparmor_profile
) {
4538 r
= aa_change_onexec(context
->apparmor_profile
);
4539 if (r
< 0 && !context
->apparmor_profile_ignore
) {
4540 *exit_status
= EXIT_APPARMOR_PROFILE
;
4541 return log_unit_error_errno(unit
, errno
, "Failed to prepare AppArmor profile change to %s: %m", context
->apparmor_profile
);
4546 /* PR_GET_SECUREBITS is not privileged, while PR_SET_SECUREBITS is. So to suppress potential EPERMs
4547 * we'll try not to call PR_SET_SECUREBITS unless necessary. Setting securebits requires
4549 if (prctl(PR_GET_SECUREBITS
) != secure_bits
) {
4550 /* CAP_SETPCAP is required to set securebits. This capability is raised into the
4551 * effective set here.
4552 * The effective set is overwritten during execve with the following values:
4553 * - ambient set (for non-root processes)
4554 * - (inheritable | bounding) set for root processes)
4556 * Hence there is no security impact to raise it in the effective set before execve
4558 r
= capability_gain_cap_setpcap(NULL
);
4560 *exit_status
= EXIT_CAPABILITIES
;
4561 return log_unit_error_errno(unit
, r
, "Failed to gain CAP_SETPCAP for setting secure bits");
4563 if (prctl(PR_SET_SECUREBITS
, secure_bits
) < 0) {
4564 *exit_status
= EXIT_SECUREBITS
;
4565 return log_unit_error_errno(unit
, errno
, "Failed to set process secure bits: %m");
4569 if (context_has_no_new_privileges(context
))
4570 if (prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0) < 0) {
4571 *exit_status
= EXIT_NO_NEW_PRIVILEGES
;
4572 return log_unit_error_errno(unit
, errno
, "Failed to disable new privileges: %m");
4576 r
= apply_address_families(unit
, context
);
4578 *exit_status
= EXIT_ADDRESS_FAMILIES
;
4579 return log_unit_error_errno(unit
, r
, "Failed to restrict address families: %m");
4582 r
= apply_memory_deny_write_execute(unit
, context
);
4584 *exit_status
= EXIT_SECCOMP
;
4585 return log_unit_error_errno(unit
, r
, "Failed to disable writing to executable memory: %m");
4588 r
= apply_restrict_realtime(unit
, context
);
4590 *exit_status
= EXIT_SECCOMP
;
4591 return log_unit_error_errno(unit
, r
, "Failed to apply realtime restrictions: %m");
4594 r
= apply_restrict_suid_sgid(unit
, context
);
4596 *exit_status
= EXIT_SECCOMP
;
4597 return log_unit_error_errno(unit
, r
, "Failed to apply SUID/SGID restrictions: %m");
4600 r
= apply_restrict_namespaces(unit
, context
);
4602 *exit_status
= EXIT_SECCOMP
;
4603 return log_unit_error_errno(unit
, r
, "Failed to apply namespace restrictions: %m");
4606 r
= apply_protect_sysctl(unit
, context
);
4608 *exit_status
= EXIT_SECCOMP
;
4609 return log_unit_error_errno(unit
, r
, "Failed to apply sysctl restrictions: %m");
4612 r
= apply_protect_kernel_modules(unit
, context
);
4614 *exit_status
= EXIT_SECCOMP
;
4615 return log_unit_error_errno(unit
, r
, "Failed to apply module loading restrictions: %m");
4618 r
= apply_protect_kernel_logs(unit
, context
);
4620 *exit_status
= EXIT_SECCOMP
;
4621 return log_unit_error_errno(unit
, r
, "Failed to apply kernel log restrictions: %m");
4624 r
= apply_protect_clock(unit
, context
);
4626 *exit_status
= EXIT_SECCOMP
;
4627 return log_unit_error_errno(unit
, r
, "Failed to apply clock restrictions: %m");
4630 r
= apply_private_devices(unit
, context
);
4632 *exit_status
= EXIT_SECCOMP
;
4633 return log_unit_error_errno(unit
, r
, "Failed to set up private devices: %m");
4636 r
= apply_syscall_archs(unit
, context
);
4638 *exit_status
= EXIT_SECCOMP
;
4639 return log_unit_error_errno(unit
, r
, "Failed to apply syscall architecture restrictions: %m");
4642 r
= apply_lock_personality(unit
, context
);
4644 *exit_status
= EXIT_SECCOMP
;
4645 return log_unit_error_errno(unit
, r
, "Failed to lock personalities: %m");
4648 r
= apply_syscall_log(unit
, context
);
4650 *exit_status
= EXIT_SECCOMP
;
4651 return log_unit_error_errno(unit
, r
, "Failed to apply system call log filters: %m");
4654 /* This really should remain the last step before the execve(), to make sure our own code is unaffected
4655 * by the filter as little as possible. */
4656 r
= apply_syscall_filter(unit
, context
, needs_ambient_hack
);
4658 *exit_status
= EXIT_SECCOMP
;
4659 return log_unit_error_errno(unit
, r
, "Failed to apply system call filters: %m");
4664 if (!strv_isempty(context
->unset_environment
)) {
4667 ee
= strv_env_delete(accum_env
, 1, context
->unset_environment
);
4669 *exit_status
= EXIT_MEMORY
;
4673 strv_free_and_replace(accum_env
, ee
);
4676 if (!FLAGS_SET(command
->flags
, EXEC_COMMAND_NO_ENV_EXPAND
)) {
4677 replaced_argv
= replace_env_argv(command
->argv
, accum_env
);
4678 if (!replaced_argv
) {
4679 *exit_status
= EXIT_MEMORY
;
4682 final_argv
= replaced_argv
;
4684 final_argv
= command
->argv
;
4686 if (DEBUG_LOGGING
) {
4687 _cleanup_free_
char *line
= NULL
;
4689 line
= quote_command_line(final_argv
);
4691 *exit_status
= EXIT_MEMORY
;
4695 log_unit_struct(unit
, LOG_DEBUG
,
4696 "EXECUTABLE=%s", executable
,
4697 LOG_UNIT_MESSAGE(unit
, "Executing: %s", line
));
4703 /* We have finished with all our initializations. Let's now let the manager know that. From this point
4704 * on, if the manager sees POLLHUP on the exec_fd, then execve() was successful. */
4706 if (write(exec_fd
, &hot
, sizeof(hot
)) < 0) {
4707 *exit_status
= EXIT_EXEC
;
4708 return log_unit_error_errno(unit
, errno
, "Failed to enable exec_fd: %m");
4712 r
= fexecve_or_execve(executable_fd
, executable
, final_argv
, accum_env
);
4717 /* The execve() failed. This means the exec_fd is still open. Which means we need to tell the manager
4718 * that POLLHUP on it no longer means execve() succeeded. */
4720 if (write(exec_fd
, &hot
, sizeof(hot
)) < 0) {
4721 *exit_status
= EXIT_EXEC
;
4722 return log_unit_error_errno(unit
, errno
, "Failed to disable exec_fd: %m");
4726 *exit_status
= EXIT_EXEC
;
4727 return log_unit_error_errno(unit
, r
, "Failed to execute %s: %m", executable
);
4730 static int exec_context_load_environment(const Unit
*unit
, const ExecContext
*c
, char ***l
);
4731 static int exec_context_named_iofds(const ExecContext
*c
, const ExecParameters
*p
, int named_iofds
[static 3]);
4733 int exec_spawn(Unit
*unit
,
4734 ExecCommand
*command
,
4735 const ExecContext
*context
,
4736 const ExecParameters
*params
,
4737 ExecRuntime
*runtime
,
4738 DynamicCreds
*dcreds
,
4741 int socket_fd
, r
, named_iofds
[3] = { -1, -1, -1 }, *fds
= NULL
;
4742 _cleanup_free_
char *subcgroup_path
= NULL
;
4743 _cleanup_strv_free_
char **files_env
= NULL
;
4744 size_t n_storage_fds
= 0, n_socket_fds
= 0;
4745 _cleanup_free_
char *line
= NULL
;
4753 assert(params
->fds
|| (params
->n_socket_fds
+ params
->n_storage_fds
<= 0));
4755 if (context
->std_input
== EXEC_INPUT_SOCKET
||
4756 context
->std_output
== EXEC_OUTPUT_SOCKET
||
4757 context
->std_error
== EXEC_OUTPUT_SOCKET
) {
4759 if (params
->n_socket_fds
> 1)
4760 return log_unit_error_errno(unit
, SYNTHETIC_ERRNO(EINVAL
), "Got more than one socket.");
4762 if (params
->n_socket_fds
== 0)
4763 return log_unit_error_errno(unit
, SYNTHETIC_ERRNO(EINVAL
), "Got no socket.");
4765 socket_fd
= params
->fds
[0];
4769 n_socket_fds
= params
->n_socket_fds
;
4770 n_storage_fds
= params
->n_storage_fds
;
4773 r
= exec_context_named_iofds(context
, params
, named_iofds
);
4775 return log_unit_error_errno(unit
, r
, "Failed to load a named file descriptor: %m");
4777 r
= exec_context_load_environment(unit
, context
, &files_env
);
4779 return log_unit_error_errno(unit
, r
, "Failed to load environment files: %m");
4781 line
= quote_command_line(command
->argv
);
4785 /* Fork with up-to-date SELinux label database, so the child inherits the up-to-date db
4786 and, until the next SELinux policy changes, we save further reloads in future children. */
4787 mac_selinux_maybe_reload();
4789 log_unit_struct(unit
, LOG_DEBUG
,
4790 LOG_UNIT_MESSAGE(unit
, "About to execute %s", line
),
4791 "EXECUTABLE=%s", command
->path
, /* We won't know the real executable path until we create
4792 the mount namespace in the child, but we want to log
4793 from the parent, so we need to use the (possibly
4794 inaccurate) path here. */
4795 LOG_UNIT_INVOCATION_ID(unit
));
4797 if (params
->cgroup_path
) {
4798 r
= exec_parameters_get_cgroup_path(params
, &subcgroup_path
);
4800 return log_unit_error_errno(unit
, r
, "Failed to acquire subcgroup path: %m");
4801 if (r
> 0) { /* We are using a child cgroup */
4802 r
= cg_create(SYSTEMD_CGROUP_CONTROLLER
, subcgroup_path
);
4804 return log_unit_error_errno(unit
, r
, "Failed to create control group '%s': %m", subcgroup_path
);
4806 /* Normally we would not propagate the oomd xattrs to children but since we created this
4807 * sub-cgroup internally we should do it. */
4808 cgroup_oomd_xattr_apply(unit
, subcgroup_path
);
4814 return log_unit_error_errno(unit
, errno
, "Failed to fork: %m");
4817 int exit_status
= EXIT_SUCCESS
;
4819 r
= exec_child(unit
,
4831 unit
->manager
->user_lookup_fds
[1],
4835 const char *status
=
4836 exit_status_to_string(exit_status
,
4837 EXIT_STATUS_LIBC
| EXIT_STATUS_SYSTEMD
);
4839 log_unit_struct_errno(unit
, LOG_ERR
, r
,
4840 "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR
,
4841 LOG_UNIT_INVOCATION_ID(unit
),
4842 LOG_UNIT_MESSAGE(unit
, "Failed at step %s spawning %s: %m",
4843 status
, command
->path
),
4844 "EXECUTABLE=%s", command
->path
);
4850 log_unit_debug(unit
, "Forked %s as "PID_FMT
, command
->path
, pid
);
4852 /* We add the new process to the cgroup both in the child (so that we can be sure that no user code is ever
4853 * executed outside of the cgroup) and in the parent (so that we can be sure that when we kill the cgroup the
4854 * process will be killed too). */
4856 (void) cg_attach(SYSTEMD_CGROUP_CONTROLLER
, subcgroup_path
, pid
);
4858 exec_status_start(&command
->exec_status
, pid
);
4864 void exec_context_init(ExecContext
*c
) {
4868 c
->ioprio
= IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE
, 0);
4869 c
->cpu_sched_policy
= SCHED_OTHER
;
4870 c
->syslog_priority
= LOG_DAEMON
|LOG_INFO
;
4871 c
->syslog_level_prefix
= true;
4872 c
->ignore_sigpipe
= true;
4873 c
->timer_slack_nsec
= NSEC_INFINITY
;
4874 c
->personality
= PERSONALITY_INVALID
;
4875 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++)
4876 c
->directories
[t
].mode
= 0755;
4877 c
->timeout_clean_usec
= USEC_INFINITY
;
4878 c
->capability_bounding_set
= CAP_ALL
;
4879 assert_cc(NAMESPACE_FLAGS_INITIAL
!= NAMESPACE_FLAGS_ALL
);
4880 c
->restrict_namespaces
= NAMESPACE_FLAGS_INITIAL
;
4881 c
->log_level_max
= -1;
4883 c
->syscall_errno
= SECCOMP_ERROR_NUMBER_KILL
;
4885 numa_policy_reset(&c
->numa_policy
);
4888 void exec_context_done(ExecContext
*c
) {
4891 c
->environment
= strv_free(c
->environment
);
4892 c
->environment_files
= strv_free(c
->environment_files
);
4893 c
->pass_environment
= strv_free(c
->pass_environment
);
4894 c
->unset_environment
= strv_free(c
->unset_environment
);
4896 rlimit_free_all(c
->rlimit
);
4898 for (size_t l
= 0; l
< 3; l
++) {
4899 c
->stdio_fdname
[l
] = mfree(c
->stdio_fdname
[l
]);
4900 c
->stdio_file
[l
] = mfree(c
->stdio_file
[l
]);
4903 c
->working_directory
= mfree(c
->working_directory
);
4904 c
->root_directory
= mfree(c
->root_directory
);
4905 c
->root_image
= mfree(c
->root_image
);
4906 c
->root_image_options
= mount_options_free_all(c
->root_image_options
);
4907 c
->root_hash
= mfree(c
->root_hash
);
4908 c
->root_hash_size
= 0;
4909 c
->root_hash_path
= mfree(c
->root_hash_path
);
4910 c
->root_hash_sig
= mfree(c
->root_hash_sig
);
4911 c
->root_hash_sig_size
= 0;
4912 c
->root_hash_sig_path
= mfree(c
->root_hash_sig_path
);
4913 c
->root_verity
= mfree(c
->root_verity
);
4914 c
->extension_images
= mount_image_free_many(c
->extension_images
, &c
->n_extension_images
);
4915 c
->tty_path
= mfree(c
->tty_path
);
4916 c
->syslog_identifier
= mfree(c
->syslog_identifier
);
4917 c
->user
= mfree(c
->user
);
4918 c
->group
= mfree(c
->group
);
4920 c
->supplementary_groups
= strv_free(c
->supplementary_groups
);
4922 c
->pam_name
= mfree(c
->pam_name
);
4924 c
->read_only_paths
= strv_free(c
->read_only_paths
);
4925 c
->read_write_paths
= strv_free(c
->read_write_paths
);
4926 c
->inaccessible_paths
= strv_free(c
->inaccessible_paths
);
4927 c
->exec_paths
= strv_free(c
->exec_paths
);
4928 c
->no_exec_paths
= strv_free(c
->no_exec_paths
);
4930 bind_mount_free_many(c
->bind_mounts
, c
->n_bind_mounts
);
4931 c
->bind_mounts
= NULL
;
4932 c
->n_bind_mounts
= 0;
4933 temporary_filesystem_free_many(c
->temporary_filesystems
, c
->n_temporary_filesystems
);
4934 c
->temporary_filesystems
= NULL
;
4935 c
->n_temporary_filesystems
= 0;
4936 c
->mount_images
= mount_image_free_many(c
->mount_images
, &c
->n_mount_images
);
4938 cpu_set_reset(&c
->cpu_set
);
4939 numa_policy_reset(&c
->numa_policy
);
4941 c
->utmp_id
= mfree(c
->utmp_id
);
4942 c
->selinux_context
= mfree(c
->selinux_context
);
4943 c
->apparmor_profile
= mfree(c
->apparmor_profile
);
4944 c
->smack_process_label
= mfree(c
->smack_process_label
);
4946 c
->syscall_filter
= hashmap_free(c
->syscall_filter
);
4947 c
->syscall_archs
= set_free(c
->syscall_archs
);
4948 c
->address_families
= set_free(c
->address_families
);
4950 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++)
4951 c
->directories
[t
].paths
= strv_free(c
->directories
[t
].paths
);
4953 c
->log_level_max
= -1;
4955 exec_context_free_log_extra_fields(c
);
4957 c
->log_ratelimit_interval_usec
= 0;
4958 c
->log_ratelimit_burst
= 0;
4960 c
->stdin_data
= mfree(c
->stdin_data
);
4961 c
->stdin_data_size
= 0;
4963 c
->network_namespace_path
= mfree(c
->network_namespace_path
);
4964 c
->ipc_namespace_path
= mfree(c
->ipc_namespace_path
);
4966 c
->log_namespace
= mfree(c
->log_namespace
);
4968 c
->load_credentials
= hashmap_free(c
->load_credentials
);
4969 c
->set_credentials
= hashmap_free(c
->set_credentials
);
4972 int exec_context_destroy_runtime_directory(const ExecContext
*c
, const char *runtime_prefix
) {
4977 if (!runtime_prefix
)
4980 STRV_FOREACH(i
, c
->directories
[EXEC_DIRECTORY_RUNTIME
].paths
) {
4981 _cleanup_free_
char *p
= NULL
;
4983 if (exec_directory_is_private(c
, EXEC_DIRECTORY_RUNTIME
))
4984 p
= path_join(runtime_prefix
, "private", *i
);
4986 p
= path_join(runtime_prefix
, *i
);
4990 /* We execute this synchronously, since we need to be sure this is gone when we start the
4992 (void) rm_rf(p
, REMOVE_ROOT
);
4998 int exec_context_destroy_credentials(const ExecContext
*c
, const char *runtime_prefix
, const char *unit
) {
4999 _cleanup_free_
char *p
= NULL
;
5003 if (!runtime_prefix
|| !unit
)
5006 p
= path_join(runtime_prefix
, "credentials", unit
);
5010 /* This is either a tmpfs/ramfs of its own, or a plain directory. Either way, let's first try to
5011 * unmount it, and afterwards remove the mount point */
5012 (void) umount2(p
, MNT_DETACH
|UMOUNT_NOFOLLOW
);
5013 (void) rm_rf(p
, REMOVE_ROOT
|REMOVE_CHMOD
);
5018 static void exec_command_done(ExecCommand
*c
) {
5021 c
->path
= mfree(c
->path
);
5022 c
->argv
= strv_free(c
->argv
);
5025 void exec_command_done_array(ExecCommand
*c
, size_t n
) {
5026 for (size_t i
= 0; i
< n
; i
++)
5027 exec_command_done(c
+i
);
5030 ExecCommand
* exec_command_free_list(ExecCommand
*c
) {
5034 LIST_REMOVE(command
, c
, i
);
5035 exec_command_done(i
);
5042 void exec_command_free_array(ExecCommand
**c
, size_t n
) {
5043 for (size_t i
= 0; i
< n
; i
++)
5044 c
[i
] = exec_command_free_list(c
[i
]);
5047 void exec_command_reset_status_array(ExecCommand
*c
, size_t n
) {
5048 for (size_t i
= 0; i
< n
; i
++)
5049 exec_status_reset(&c
[i
].exec_status
);
5052 void exec_command_reset_status_list_array(ExecCommand
**c
, size_t n
) {
5053 for (size_t i
= 0; i
< n
; i
++) {
5056 LIST_FOREACH(command
, z
, c
[i
])
5057 exec_status_reset(&z
->exec_status
);
5061 typedef struct InvalidEnvInfo
{
5066 static void invalid_env(const char *p
, void *userdata
) {
5067 InvalidEnvInfo
*info
= userdata
;
5069 log_unit_error(info
->unit
, "Ignoring invalid environment assignment '%s': %s", p
, info
->path
);
5072 const char* exec_context_fdname(const ExecContext
*c
, int fd_index
) {
5078 if (c
->std_input
!= EXEC_INPUT_NAMED_FD
)
5081 return c
->stdio_fdname
[STDIN_FILENO
] ?: "stdin";
5084 if (c
->std_output
!= EXEC_OUTPUT_NAMED_FD
)
5087 return c
->stdio_fdname
[STDOUT_FILENO
] ?: "stdout";
5090 if (c
->std_error
!= EXEC_OUTPUT_NAMED_FD
)
5093 return c
->stdio_fdname
[STDERR_FILENO
] ?: "stderr";
5100 static int exec_context_named_iofds(
5101 const ExecContext
*c
,
5102 const ExecParameters
*p
,
5103 int named_iofds
[static 3]) {
5106 const char* stdio_fdname
[3];
5111 assert(named_iofds
);
5113 targets
= (c
->std_input
== EXEC_INPUT_NAMED_FD
) +
5114 (c
->std_output
== EXEC_OUTPUT_NAMED_FD
) +
5115 (c
->std_error
== EXEC_OUTPUT_NAMED_FD
);
5117 for (size_t i
= 0; i
< 3; i
++)
5118 stdio_fdname
[i
] = exec_context_fdname(c
, i
);
5120 n_fds
= p
->n_storage_fds
+ p
->n_socket_fds
;
5122 for (size_t i
= 0; i
< n_fds
&& targets
> 0; i
++)
5123 if (named_iofds
[STDIN_FILENO
] < 0 &&
5124 c
->std_input
== EXEC_INPUT_NAMED_FD
&&
5125 stdio_fdname
[STDIN_FILENO
] &&
5126 streq(p
->fd_names
[i
], stdio_fdname
[STDIN_FILENO
])) {
5128 named_iofds
[STDIN_FILENO
] = p
->fds
[i
];
5131 } else if (named_iofds
[STDOUT_FILENO
] < 0 &&
5132 c
->std_output
== EXEC_OUTPUT_NAMED_FD
&&
5133 stdio_fdname
[STDOUT_FILENO
] &&
5134 streq(p
->fd_names
[i
], stdio_fdname
[STDOUT_FILENO
])) {
5136 named_iofds
[STDOUT_FILENO
] = p
->fds
[i
];
5139 } else if (named_iofds
[STDERR_FILENO
] < 0 &&
5140 c
->std_error
== EXEC_OUTPUT_NAMED_FD
&&
5141 stdio_fdname
[STDERR_FILENO
] &&
5142 streq(p
->fd_names
[i
], stdio_fdname
[STDERR_FILENO
])) {
5144 named_iofds
[STDERR_FILENO
] = p
->fds
[i
];
5148 return targets
== 0 ? 0 : -ENOENT
;
5151 static int exec_context_load_environment(const Unit
*unit
, const ExecContext
*c
, char ***l
) {
5152 char **i
, **r
= NULL
;
5157 STRV_FOREACH(i
, c
->environment_files
) {
5160 bool ignore
= false;
5162 _cleanup_globfree_ glob_t pglob
= {};
5171 if (!path_is_absolute(fn
)) {
5179 /* Filename supports globbing, take all matching files */
5180 k
= safe_glob(fn
, 0, &pglob
);
5189 /* When we don't match anything, -ENOENT should be returned */
5190 assert(pglob
.gl_pathc
> 0);
5192 for (unsigned n
= 0; n
< pglob
.gl_pathc
; n
++) {
5193 k
= load_env_file(NULL
, pglob
.gl_pathv
[n
], &p
);
5201 /* Log invalid environment variables with filename */
5203 InvalidEnvInfo info
= {
5205 .path
= pglob
.gl_pathv
[n
]
5208 p
= strv_env_clean_with_callback(p
, invalid_env
, &info
);
5216 m
= strv_env_merge(r
, p
);
5232 static bool tty_may_match_dev_console(const char *tty
) {
5233 _cleanup_free_
char *resolved
= NULL
;
5238 tty
= skip_dev_prefix(tty
);
5240 /* trivial identity? */
5241 if (streq(tty
, "console"))
5244 if (resolve_dev_console(&resolved
) < 0)
5245 return true; /* if we could not resolve, assume it may */
5247 /* "tty0" means the active VC, so it may be the same sometimes */
5248 return path_equal(resolved
, tty
) || (streq(resolved
, "tty0") && tty_is_vc(tty
));
5251 static bool exec_context_may_touch_tty(const ExecContext
*ec
) {
5254 return ec
->tty_reset
||
5256 ec
->tty_vt_disallocate
||
5257 is_terminal_input(ec
->std_input
) ||
5258 is_terminal_output(ec
->std_output
) ||
5259 is_terminal_output(ec
->std_error
);
5262 bool exec_context_may_touch_console(const ExecContext
*ec
) {
5264 return exec_context_may_touch_tty(ec
) &&
5265 tty_may_match_dev_console(exec_context_tty_path(ec
));
5268 static void strv_fprintf(FILE *f
, char **l
) {
5274 fprintf(f
, " %s", *g
);
5277 static void strv_dump(FILE* f
, const char *prefix
, const char *name
, char **strv
) {
5282 if (!strv_isempty(strv
)) {
5283 fprintf(f
, "%s%s:", prefix
, name
);
5284 strv_fprintf(f
, strv
);
5289 void exec_context_dump(const ExecContext
*c
, FILE* f
, const char *prefix
) {
5296 prefix
= strempty(prefix
);
5300 "%sWorkingDirectory: %s\n"
5301 "%sRootDirectory: %s\n"
5302 "%sNonBlocking: %s\n"
5303 "%sPrivateTmp: %s\n"
5304 "%sPrivateDevices: %s\n"
5305 "%sProtectKernelTunables: %s\n"
5306 "%sProtectKernelModules: %s\n"
5307 "%sProtectKernelLogs: %s\n"
5308 "%sProtectClock: %s\n"
5309 "%sProtectControlGroups: %s\n"
5310 "%sPrivateNetwork: %s\n"
5311 "%sPrivateUsers: %s\n"
5312 "%sProtectHome: %s\n"
5313 "%sProtectSystem: %s\n"
5314 "%sMountAPIVFS: %s\n"
5315 "%sIgnoreSIGPIPE: %s\n"
5316 "%sMemoryDenyWriteExecute: %s\n"
5317 "%sRestrictRealtime: %s\n"
5318 "%sRestrictSUIDSGID: %s\n"
5319 "%sKeyringMode: %s\n"
5320 "%sProtectHostname: %s\n"
5321 "%sProtectProc: %s\n"
5322 "%sProcSubset: %s\n",
5324 prefix
, empty_to_root(c
->working_directory
),
5325 prefix
, empty_to_root(c
->root_directory
),
5326 prefix
, yes_no(c
->non_blocking
),
5327 prefix
, yes_no(c
->private_tmp
),
5328 prefix
, yes_no(c
->private_devices
),
5329 prefix
, yes_no(c
->protect_kernel_tunables
),
5330 prefix
, yes_no(c
->protect_kernel_modules
),
5331 prefix
, yes_no(c
->protect_kernel_logs
),
5332 prefix
, yes_no(c
->protect_clock
),
5333 prefix
, yes_no(c
->protect_control_groups
),
5334 prefix
, yes_no(c
->private_network
),
5335 prefix
, yes_no(c
->private_users
),
5336 prefix
, protect_home_to_string(c
->protect_home
),
5337 prefix
, protect_system_to_string(c
->protect_system
),
5338 prefix
, yes_no(exec_context_get_effective_mount_apivfs(c
)),
5339 prefix
, yes_no(c
->ignore_sigpipe
),
5340 prefix
, yes_no(c
->memory_deny_write_execute
),
5341 prefix
, yes_no(c
->restrict_realtime
),
5342 prefix
, yes_no(c
->restrict_suid_sgid
),
5343 prefix
, exec_keyring_mode_to_string(c
->keyring_mode
),
5344 prefix
, yes_no(c
->protect_hostname
),
5345 prefix
, protect_proc_to_string(c
->protect_proc
),
5346 prefix
, proc_subset_to_string(c
->proc_subset
));
5349 fprintf(f
, "%sRootImage: %s\n", prefix
, c
->root_image
);
5351 if (c
->root_image_options
) {
5354 fprintf(f
, "%sRootImageOptions:", prefix
);
5355 LIST_FOREACH(mount_options
, o
, c
->root_image_options
)
5356 if (!isempty(o
->options
))
5357 fprintf(f
, " %s:%s",
5358 partition_designator_to_string(o
->partition_designator
),
5364 _cleanup_free_
char *encoded
= NULL
;
5365 encoded
= hexmem(c
->root_hash
, c
->root_hash_size
);
5367 fprintf(f
, "%sRootHash: %s\n", prefix
, encoded
);
5370 if (c
->root_hash_path
)
5371 fprintf(f
, "%sRootHash: %s\n", prefix
, c
->root_hash_path
);
5373 if (c
->root_hash_sig
) {
5374 _cleanup_free_
char *encoded
= NULL
;
5376 len
= base64mem(c
->root_hash_sig
, c
->root_hash_sig_size
, &encoded
);
5378 fprintf(f
, "%sRootHashSignature: base64:%s\n", prefix
, encoded
);
5381 if (c
->root_hash_sig_path
)
5382 fprintf(f
, "%sRootHashSignature: %s\n", prefix
, c
->root_hash_sig_path
);
5385 fprintf(f
, "%sRootVerity: %s\n", prefix
, c
->root_verity
);
5387 STRV_FOREACH(e
, c
->environment
)
5388 fprintf(f
, "%sEnvironment: %s\n", prefix
, *e
);
5390 STRV_FOREACH(e
, c
->environment_files
)
5391 fprintf(f
, "%sEnvironmentFile: %s\n", prefix
, *e
);
5393 STRV_FOREACH(e
, c
->pass_environment
)
5394 fprintf(f
, "%sPassEnvironment: %s\n", prefix
, *e
);
5396 STRV_FOREACH(e
, c
->unset_environment
)
5397 fprintf(f
, "%sUnsetEnvironment: %s\n", prefix
, *e
);
5399 fprintf(f
, "%sRuntimeDirectoryPreserve: %s\n", prefix
, exec_preserve_mode_to_string(c
->runtime_directory_preserve_mode
));
5401 for (ExecDirectoryType dt
= 0; dt
< _EXEC_DIRECTORY_TYPE_MAX
; dt
++) {
5402 fprintf(f
, "%s%sMode: %04o\n", prefix
, exec_directory_type_to_string(dt
), c
->directories
[dt
].mode
);
5404 STRV_FOREACH(d
, c
->directories
[dt
].paths
)
5405 fprintf(f
, "%s%s: %s\n", prefix
, exec_directory_type_to_string(dt
), *d
);
5408 fprintf(f
, "%sTimeoutCleanSec: %s\n", prefix
, FORMAT_TIMESPAN(c
->timeout_clean_usec
, USEC_PER_SEC
));
5411 fprintf(f
, "%sNice: %i\n", prefix
, c
->nice
);
5413 if (c
->oom_score_adjust_set
)
5414 fprintf(f
, "%sOOMScoreAdjust: %i\n", prefix
, c
->oom_score_adjust
);
5416 if (c
->coredump_filter_set
)
5417 fprintf(f
, "%sCoredumpFilter: 0x%"PRIx64
"\n", prefix
, c
->coredump_filter
);
5419 for (unsigned i
= 0; i
< RLIM_NLIMITS
; i
++)
5421 fprintf(f
, "%sLimit%s: " RLIM_FMT
"\n",
5422 prefix
, rlimit_to_string(i
), c
->rlimit
[i
]->rlim_max
);
5423 fprintf(f
, "%sLimit%sSoft: " RLIM_FMT
"\n",
5424 prefix
, rlimit_to_string(i
), c
->rlimit
[i
]->rlim_cur
);
5427 if (c
->ioprio_set
) {
5428 _cleanup_free_
char *class_str
= NULL
;
5430 r
= ioprio_class_to_string_alloc(IOPRIO_PRIO_CLASS(c
->ioprio
), &class_str
);
5432 fprintf(f
, "%sIOSchedulingClass: %s\n", prefix
, class_str
);
5434 fprintf(f
, "%sIOPriority: %lu\n", prefix
, IOPRIO_PRIO_DATA(c
->ioprio
));
5437 if (c
->cpu_sched_set
) {
5438 _cleanup_free_
char *policy_str
= NULL
;
5440 r
= sched_policy_to_string_alloc(c
->cpu_sched_policy
, &policy_str
);
5442 fprintf(f
, "%sCPUSchedulingPolicy: %s\n", prefix
, policy_str
);
5445 "%sCPUSchedulingPriority: %i\n"
5446 "%sCPUSchedulingResetOnFork: %s\n",
5447 prefix
, c
->cpu_sched_priority
,
5448 prefix
, yes_no(c
->cpu_sched_reset_on_fork
));
5451 if (c
->cpu_set
.set
) {
5452 _cleanup_free_
char *affinity
= NULL
;
5454 affinity
= cpu_set_to_range_string(&c
->cpu_set
);
5455 fprintf(f
, "%sCPUAffinity: %s\n", prefix
, affinity
);
5458 if (mpol_is_valid(numa_policy_get_type(&c
->numa_policy
))) {
5459 _cleanup_free_
char *nodes
= NULL
;
5461 nodes
= cpu_set_to_range_string(&c
->numa_policy
.nodes
);
5462 fprintf(f
, "%sNUMAPolicy: %s\n", prefix
, mpol_to_string(numa_policy_get_type(&c
->numa_policy
)));
5463 fprintf(f
, "%sNUMAMask: %s\n", prefix
, strnull(nodes
));
5466 if (c
->timer_slack_nsec
!= NSEC_INFINITY
)
5467 fprintf(f
, "%sTimerSlackNSec: "NSEC_FMT
"\n", prefix
, c
->timer_slack_nsec
);
5470 "%sStandardInput: %s\n"
5471 "%sStandardOutput: %s\n"
5472 "%sStandardError: %s\n",
5473 prefix
, exec_input_to_string(c
->std_input
),
5474 prefix
, exec_output_to_string(c
->std_output
),
5475 prefix
, exec_output_to_string(c
->std_error
));
5477 if (c
->std_input
== EXEC_INPUT_NAMED_FD
)
5478 fprintf(f
, "%sStandardInputFileDescriptorName: %s\n", prefix
, c
->stdio_fdname
[STDIN_FILENO
]);
5479 if (c
->std_output
== EXEC_OUTPUT_NAMED_FD
)
5480 fprintf(f
, "%sStandardOutputFileDescriptorName: %s\n", prefix
, c
->stdio_fdname
[STDOUT_FILENO
]);
5481 if (c
->std_error
== EXEC_OUTPUT_NAMED_FD
)
5482 fprintf(f
, "%sStandardErrorFileDescriptorName: %s\n", prefix
, c
->stdio_fdname
[STDERR_FILENO
]);
5484 if (c
->std_input
== EXEC_INPUT_FILE
)
5485 fprintf(f
, "%sStandardInputFile: %s\n", prefix
, c
->stdio_file
[STDIN_FILENO
]);
5486 if (c
->std_output
== EXEC_OUTPUT_FILE
)
5487 fprintf(f
, "%sStandardOutputFile: %s\n", prefix
, c
->stdio_file
[STDOUT_FILENO
]);
5488 if (c
->std_output
== EXEC_OUTPUT_FILE_APPEND
)
5489 fprintf(f
, "%sStandardOutputFileToAppend: %s\n", prefix
, c
->stdio_file
[STDOUT_FILENO
]);
5490 if (c
->std_output
== EXEC_OUTPUT_FILE_TRUNCATE
)
5491 fprintf(f
, "%sStandardOutputFileToTruncate: %s\n", prefix
, c
->stdio_file
[STDOUT_FILENO
]);
5492 if (c
->std_error
== EXEC_OUTPUT_FILE
)
5493 fprintf(f
, "%sStandardErrorFile: %s\n", prefix
, c
->stdio_file
[STDERR_FILENO
]);
5494 if (c
->std_error
== EXEC_OUTPUT_FILE_APPEND
)
5495 fprintf(f
, "%sStandardErrorFileToAppend: %s\n", prefix
, c
->stdio_file
[STDERR_FILENO
]);
5496 if (c
->std_error
== EXEC_OUTPUT_FILE_TRUNCATE
)
5497 fprintf(f
, "%sStandardErrorFileToTruncate: %s\n", prefix
, c
->stdio_file
[STDERR_FILENO
]);
5503 "%sTTYVHangup: %s\n"
5504 "%sTTYVTDisallocate: %s\n",
5505 prefix
, c
->tty_path
,
5506 prefix
, yes_no(c
->tty_reset
),
5507 prefix
, yes_no(c
->tty_vhangup
),
5508 prefix
, yes_no(c
->tty_vt_disallocate
));
5510 if (IN_SET(c
->std_output
,
5512 EXEC_OUTPUT_JOURNAL
,
5513 EXEC_OUTPUT_KMSG_AND_CONSOLE
,
5514 EXEC_OUTPUT_JOURNAL_AND_CONSOLE
) ||
5515 IN_SET(c
->std_error
,
5517 EXEC_OUTPUT_JOURNAL
,
5518 EXEC_OUTPUT_KMSG_AND_CONSOLE
,
5519 EXEC_OUTPUT_JOURNAL_AND_CONSOLE
)) {
5521 _cleanup_free_
char *fac_str
= NULL
, *lvl_str
= NULL
;
5523 r
= log_facility_unshifted_to_string_alloc(c
->syslog_priority
>> 3, &fac_str
);
5525 fprintf(f
, "%sSyslogFacility: %s\n", prefix
, fac_str
);
5527 r
= log_level_to_string_alloc(LOG_PRI(c
->syslog_priority
), &lvl_str
);
5529 fprintf(f
, "%sSyslogLevel: %s\n", prefix
, lvl_str
);
5532 if (c
->log_level_max
>= 0) {
5533 _cleanup_free_
char *t
= NULL
;
5535 (void) log_level_to_string_alloc(c
->log_level_max
, &t
);
5537 fprintf(f
, "%sLogLevelMax: %s\n", prefix
, strna(t
));
5540 if (c
->log_ratelimit_interval_usec
> 0)
5542 "%sLogRateLimitIntervalSec: %s\n",
5543 prefix
, FORMAT_TIMESPAN(c
->log_ratelimit_interval_usec
, USEC_PER_SEC
));
5545 if (c
->log_ratelimit_burst
> 0)
5546 fprintf(f
, "%sLogRateLimitBurst: %u\n", prefix
, c
->log_ratelimit_burst
);
5548 for (size_t j
= 0; j
< c
->n_log_extra_fields
; j
++) {
5549 fprintf(f
, "%sLogExtraFields: ", prefix
);
5550 fwrite(c
->log_extra_fields
[j
].iov_base
,
5551 1, c
->log_extra_fields
[j
].iov_len
,
5556 if (c
->log_namespace
)
5557 fprintf(f
, "%sLogNamespace: %s\n", prefix
, c
->log_namespace
);
5559 if (c
->secure_bits
) {
5560 _cleanup_free_
char *str
= NULL
;
5562 r
= secure_bits_to_string_alloc(c
->secure_bits
, &str
);
5564 fprintf(f
, "%sSecure Bits: %s\n", prefix
, str
);
5567 if (c
->capability_bounding_set
!= CAP_ALL
) {
5568 _cleanup_free_
char *str
= NULL
;
5570 r
= capability_set_to_string_alloc(c
->capability_bounding_set
, &str
);
5572 fprintf(f
, "%sCapabilityBoundingSet: %s\n", prefix
, str
);
5575 if (c
->capability_ambient_set
!= 0) {
5576 _cleanup_free_
char *str
= NULL
;
5578 r
= capability_set_to_string_alloc(c
->capability_ambient_set
, &str
);
5580 fprintf(f
, "%sAmbientCapabilities: %s\n", prefix
, str
);
5584 fprintf(f
, "%sUser: %s\n", prefix
, c
->user
);
5586 fprintf(f
, "%sGroup: %s\n", prefix
, c
->group
);
5588 fprintf(f
, "%sDynamicUser: %s\n", prefix
, yes_no(c
->dynamic_user
));
5590 strv_dump(f
, prefix
, "SupplementaryGroups", c
->supplementary_groups
);
5593 fprintf(f
, "%sPAMName: %s\n", prefix
, c
->pam_name
);
5595 strv_dump(f
, prefix
, "ReadWritePaths", c
->read_write_paths
);
5596 strv_dump(f
, prefix
, "ReadOnlyPaths", c
->read_only_paths
);
5597 strv_dump(f
, prefix
, "InaccessiblePaths", c
->inaccessible_paths
);
5598 strv_dump(f
, prefix
, "ExecPaths", c
->exec_paths
);
5599 strv_dump(f
, prefix
, "NoExecPaths", c
->no_exec_paths
);
5601 for (size_t i
= 0; i
< c
->n_bind_mounts
; i
++)
5602 fprintf(f
, "%s%s: %s%s:%s:%s\n", prefix
,
5603 c
->bind_mounts
[i
].read_only
? "BindReadOnlyPaths" : "BindPaths",
5604 c
->bind_mounts
[i
].ignore_enoent
? "-": "",
5605 c
->bind_mounts
[i
].source
,
5606 c
->bind_mounts
[i
].destination
,
5607 c
->bind_mounts
[i
].recursive
? "rbind" : "norbind");
5609 for (size_t i
= 0; i
< c
->n_temporary_filesystems
; i
++) {
5610 const TemporaryFileSystem
*t
= c
->temporary_filesystems
+ i
;
5612 fprintf(f
, "%sTemporaryFileSystem: %s%s%s\n", prefix
,
5614 isempty(t
->options
) ? "" : ":",
5615 strempty(t
->options
));
5620 "%sUtmpIdentifier: %s\n",
5621 prefix
, c
->utmp_id
);
5623 if (c
->selinux_context
)
5625 "%sSELinuxContext: %s%s\n",
5626 prefix
, c
->selinux_context_ignore
? "-" : "", c
->selinux_context
);
5628 if (c
->apparmor_profile
)
5630 "%sAppArmorProfile: %s%s\n",
5631 prefix
, c
->apparmor_profile_ignore
? "-" : "", c
->apparmor_profile
);
5633 if (c
->smack_process_label
)
5635 "%sSmackProcessLabel: %s%s\n",
5636 prefix
, c
->smack_process_label_ignore
? "-" : "", c
->smack_process_label
);
5638 if (c
->personality
!= PERSONALITY_INVALID
)
5640 "%sPersonality: %s\n",
5641 prefix
, strna(personality_to_string(c
->personality
)));
5644 "%sLockPersonality: %s\n",
5645 prefix
, yes_no(c
->lock_personality
));
5647 if (c
->syscall_filter
) {
5654 "%sSystemCallFilter: ",
5657 if (!c
->syscall_allow_list
)
5661 HASHMAP_FOREACH_KEY(val
, id
, c
->syscall_filter
) {
5662 _cleanup_free_
char *name
= NULL
;
5663 const char *errno_name
= NULL
;
5664 int num
= PTR_TO_INT(val
);
5671 name
= seccomp_syscall_resolve_num_arch(SCMP_ARCH_NATIVE
, PTR_TO_INT(id
) - 1);
5672 fputs(strna(name
), f
);
5675 errno_name
= seccomp_errno_or_action_to_string(num
);
5677 fprintf(f
, ":%s", errno_name
);
5679 fprintf(f
, ":%d", num
);
5687 if (c
->syscall_archs
) {
5693 "%sSystemCallArchitectures:",
5697 SET_FOREACH(id
, c
->syscall_archs
)
5698 fprintf(f
, " %s", strna(seccomp_arch_to_string(PTR_TO_UINT32(id
) - 1)));
5703 if (exec_context_restrict_namespaces_set(c
)) {
5704 _cleanup_free_
char *s
= NULL
;
5706 r
= namespace_flags_to_string(c
->restrict_namespaces
, &s
);
5708 fprintf(f
, "%sRestrictNamespaces: %s\n",
5712 if (c
->network_namespace_path
)
5714 "%sNetworkNamespacePath: %s\n",
5715 prefix
, c
->network_namespace_path
);
5717 if (c
->syscall_errno
> 0) {
5719 const char *errno_name
;
5722 fprintf(f
, "%sSystemCallErrorNumber: ", prefix
);
5725 errno_name
= seccomp_errno_or_action_to_string(c
->syscall_errno
);
5727 fputs(errno_name
, f
);
5729 fprintf(f
, "%d", c
->syscall_errno
);
5734 for (size_t i
= 0; i
< c
->n_mount_images
; i
++) {
5737 fprintf(f
, "%sMountImages: %s%s:%s", prefix
,
5738 c
->mount_images
[i
].ignore_enoent
? "-": "",
5739 c
->mount_images
[i
].source
,
5740 c
->mount_images
[i
].destination
);
5741 LIST_FOREACH(mount_options
, o
, c
->mount_images
[i
].mount_options
)
5742 fprintf(f
, ":%s:%s",
5743 partition_designator_to_string(o
->partition_designator
),
5744 strempty(o
->options
));
5748 for (size_t i
= 0; i
< c
->n_extension_images
; i
++) {
5751 fprintf(f
, "%sExtensionImages: %s%s", prefix
,
5752 c
->extension_images
[i
].ignore_enoent
? "-": "",
5753 c
->extension_images
[i
].source
);
5754 LIST_FOREACH(mount_options
, o
, c
->extension_images
[i
].mount_options
)
5755 fprintf(f
, ":%s:%s",
5756 partition_designator_to_string(o
->partition_designator
),
5757 strempty(o
->options
));
5762 bool exec_context_maintains_privileges(const ExecContext
*c
) {
5765 /* Returns true if the process forked off would run under
5766 * an unchanged UID or as root. */
5771 if (streq(c
->user
, "root") || streq(c
->user
, "0"))
5777 int exec_context_get_effective_ioprio(const ExecContext
*c
) {
5785 p
= ioprio_get(IOPRIO_WHO_PROCESS
, 0);
5787 return IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE
, 4);
5792 bool exec_context_get_effective_mount_apivfs(const ExecContext
*c
) {
5795 /* Explicit setting wins */
5796 if (c
->mount_apivfs_set
)
5797 return c
->mount_apivfs
;
5799 /* Default to "yes" if root directory or image are specified */
5800 if (exec_context_with_rootfs(c
))
5806 void exec_context_free_log_extra_fields(ExecContext
*c
) {
5809 for (size_t l
= 0; l
< c
->n_log_extra_fields
; l
++)
5810 free(c
->log_extra_fields
[l
].iov_base
);
5811 c
->log_extra_fields
= mfree(c
->log_extra_fields
);
5812 c
->n_log_extra_fields
= 0;
5815 void exec_context_revert_tty(ExecContext
*c
) {
5816 _cleanup_close_
int fd
= -1;
5823 /* First, reset the TTY (possibly kicking everybody else from the TTY) */
5824 exec_context_tty_reset(c
, NULL
);
5826 /* And then undo what chown_terminal() did earlier. Note that we only do this if we have a path
5827 * configured. If the TTY was passed to us as file descriptor we assume the TTY is opened and managed
5828 * by whoever passed it to us and thus knows better when and how to chmod()/chown() it back. */
5829 if (!exec_context_may_touch_tty(c
))
5832 path
= exec_context_tty_path(c
);
5836 fd
= open(path
, O_PATH
|O_CLOEXEC
);
5838 return (void) log_full_errno(errno
== ENOENT
? LOG_DEBUG
: LOG_WARNING
, errno
,
5839 "Failed to open TTY inode of '%s' to adjust ownership/access mode, ignoring: %m",
5842 if (fstat(fd
, &st
) < 0)
5843 return (void) log_warning_errno(errno
, "Failed to stat TTY '%s', ignoring: %m", path
);
5845 /* Let's add a superficial check that we only do this for stuff that looks like a TTY. We only check
5846 * if things are a character device, since a proper check either means we'd have to open the TTY and
5847 * use isatty(), but we'd rather not do that since opening TTYs comes with all kinds of side-effects
5848 * and is slow. Or we'd have to hardcode dev_t major information, which we'd rather avoid. Why bother
5849 * with this at all? → https://github.com/systemd/systemd/issues/19213 */
5850 if (!S_ISCHR(st
.st_mode
))
5851 return log_warning("Configured TTY '%s' is not actually a character device, ignoring.", path
);
5853 r
= fchmod_and_chown(fd
, TTY_MODE
, 0, TTY_GID
);
5855 log_warning_errno(r
, "Failed to reset TTY ownership/access mode of %s, ignoring: %m", path
);
5858 int exec_context_get_clean_directories(
5864 _cleanup_strv_free_
char **l
= NULL
;
5871 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
5874 if (!FLAGS_SET(mask
, 1U << t
))
5880 STRV_FOREACH(i
, c
->directories
[t
].paths
) {
5883 j
= path_join(prefix
[t
], *i
);
5887 r
= strv_consume(&l
, j
);
5891 /* Also remove private directories unconditionally. */
5892 if (t
!= EXEC_DIRECTORY_CONFIGURATION
) {
5893 j
= path_join(prefix
[t
], "private", *i
);
5897 r
= strv_consume(&l
, j
);
5908 int exec_context_get_clean_mask(ExecContext
*c
, ExecCleanMask
*ret
) {
5909 ExecCleanMask mask
= 0;
5914 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++)
5915 if (!strv_isempty(c
->directories
[t
].paths
))
5922 void exec_status_start(ExecStatus
*s
, pid_t pid
) {
5929 dual_timestamp_get(&s
->start_timestamp
);
5932 void exec_status_exit(ExecStatus
*s
, const ExecContext
*context
, pid_t pid
, int code
, int status
) {
5940 dual_timestamp_get(&s
->exit_timestamp
);
5945 if (context
&& context
->utmp_id
)
5946 (void) utmp_put_dead_process(context
->utmp_id
, pid
, code
, status
);
5949 void exec_status_reset(ExecStatus
*s
) {
5952 *s
= (ExecStatus
) {};
5955 void exec_status_dump(const ExecStatus
*s
, FILE *f
, const char *prefix
) {
5962 prefix
= strempty(prefix
);
5965 "%sPID: "PID_FMT
"\n",
5968 if (dual_timestamp_is_set(&s
->start_timestamp
))
5970 "%sStart Timestamp: %s\n",
5971 prefix
, FORMAT_TIMESTAMP(s
->start_timestamp
.realtime
));
5973 if (dual_timestamp_is_set(&s
->exit_timestamp
))
5975 "%sExit Timestamp: %s\n"
5977 "%sExit Status: %i\n",
5978 prefix
, FORMAT_TIMESTAMP(s
->exit_timestamp
.realtime
),
5979 prefix
, sigchld_code_to_string(s
->code
),
5983 static void exec_command_dump(ExecCommand
*c
, FILE *f
, const char *prefix
) {
5984 _cleanup_free_
char *cmd
= NULL
;
5985 const char *prefix2
;
5990 prefix
= strempty(prefix
);
5991 prefix2
= strjoina(prefix
, "\t");
5993 cmd
= quote_command_line(c
->argv
);
5995 "%sCommand Line: %s\n",
5996 prefix
, cmd
? cmd
: strerror_safe(ENOMEM
));
5998 exec_status_dump(&c
->exec_status
, f
, prefix2
);
6001 void exec_command_dump_list(ExecCommand
*c
, FILE *f
, const char *prefix
) {
6004 prefix
= strempty(prefix
);
6006 LIST_FOREACH(command
, c
, c
)
6007 exec_command_dump(c
, f
, prefix
);
6010 void exec_command_append_list(ExecCommand
**l
, ExecCommand
*e
) {
6017 /* It's kind of important, that we keep the order here */
6018 LIST_FIND_TAIL(command
, *l
, end
);
6019 LIST_INSERT_AFTER(command
, *l
, end
, e
);
6024 int exec_command_set(ExecCommand
*c
, const char *path
, ...) {
6032 l
= strv_new_ap(path
, ap
);
6044 free_and_replace(c
->path
, p
);
6046 return strv_free_and_replace(c
->argv
, l
);
6049 int exec_command_append(ExecCommand
*c
, const char *path
, ...) {
6050 _cleanup_strv_free_
char **l
= NULL
;
6058 l
= strv_new_ap(path
, ap
);
6064 r
= strv_extend_strv(&c
->argv
, l
, false);
6071 static void *remove_tmpdir_thread(void *p
) {
6072 _cleanup_free_
char *path
= p
;
6074 (void) rm_rf(path
, REMOVE_ROOT
|REMOVE_PHYSICAL
);
6078 static ExecRuntime
* exec_runtime_free(ExecRuntime
*rt
, bool destroy
) {
6085 (void) hashmap_remove(rt
->manager
->exec_runtime_by_id
, rt
->id
);
6087 /* When destroy is true, then rm_rf tmp_dir and var_tmp_dir. */
6089 if (destroy
&& rt
->tmp_dir
&& !streq(rt
->tmp_dir
, RUN_SYSTEMD_EMPTY
)) {
6090 log_debug("Spawning thread to nuke %s", rt
->tmp_dir
);
6092 r
= asynchronous_job(remove_tmpdir_thread
, rt
->tmp_dir
);
6094 log_warning_errno(r
, "Failed to nuke %s: %m", rt
->tmp_dir
);
6099 if (destroy
&& rt
->var_tmp_dir
&& !streq(rt
->var_tmp_dir
, RUN_SYSTEMD_EMPTY
)) {
6100 log_debug("Spawning thread to nuke %s", rt
->var_tmp_dir
);
6102 r
= asynchronous_job(remove_tmpdir_thread
, rt
->var_tmp_dir
);
6104 log_warning_errno(r
, "Failed to nuke %s: %m", rt
->var_tmp_dir
);
6106 rt
->var_tmp_dir
= NULL
;
6109 rt
->id
= mfree(rt
->id
);
6110 rt
->tmp_dir
= mfree(rt
->tmp_dir
);
6111 rt
->var_tmp_dir
= mfree(rt
->var_tmp_dir
);
6112 safe_close_pair(rt
->netns_storage_socket
);
6113 safe_close_pair(rt
->ipcns_storage_socket
);
6117 static void exec_runtime_freep(ExecRuntime
**rt
) {
6118 (void) exec_runtime_free(*rt
, false);
6121 static int exec_runtime_allocate(ExecRuntime
**ret
, const char *id
) {
6122 _cleanup_free_
char *id_copy
= NULL
;
6127 id_copy
= strdup(id
);
6131 n
= new(ExecRuntime
, 1);
6135 *n
= (ExecRuntime
) {
6136 .id
= TAKE_PTR(id_copy
),
6137 .netns_storage_socket
= { -1, -1 },
6138 .ipcns_storage_socket
= { -1, -1 },
6145 static int exec_runtime_add(
6150 int netns_storage_socket
[2],
6151 int ipcns_storage_socket
[2],
6152 ExecRuntime
**ret
) {
6154 _cleanup_(exec_runtime_freep
) ExecRuntime
*rt
= NULL
;
6160 /* tmp_dir, var_tmp_dir, {net,ipc}ns_storage_socket fds are donated on success */
6162 r
= exec_runtime_allocate(&rt
, id
);
6166 r
= hashmap_ensure_put(&m
->exec_runtime_by_id
, &string_hash_ops
, rt
->id
, rt
);
6170 assert(!!rt
->tmp_dir
== !!rt
->var_tmp_dir
); /* We require both to be set together */
6171 rt
->tmp_dir
= TAKE_PTR(*tmp_dir
);
6172 rt
->var_tmp_dir
= TAKE_PTR(*var_tmp_dir
);
6174 if (netns_storage_socket
) {
6175 rt
->netns_storage_socket
[0] = TAKE_FD(netns_storage_socket
[0]);
6176 rt
->netns_storage_socket
[1] = TAKE_FD(netns_storage_socket
[1]);
6179 if (ipcns_storage_socket
) {
6180 rt
->ipcns_storage_socket
[0] = TAKE_FD(ipcns_storage_socket
[0]);
6181 rt
->ipcns_storage_socket
[1] = TAKE_FD(ipcns_storage_socket
[1]);
6188 /* do not remove created ExecRuntime object when the operation succeeds. */
6193 static int exec_runtime_make(
6195 const ExecContext
*c
,
6197 ExecRuntime
**ret
) {
6199 _cleanup_(namespace_cleanup_tmpdirp
) char *tmp_dir
= NULL
, *var_tmp_dir
= NULL
;
6200 _cleanup_close_pair_
int netns_storage_socket
[2] = { -1, -1 }, ipcns_storage_socket
[2] = { -1, -1 };
6207 /* It is not necessary to create ExecRuntime object. */
6208 if (!c
->private_network
&& !c
->private_ipc
&& !c
->private_tmp
&& !c
->network_namespace_path
) {
6213 if (c
->private_tmp
&&
6214 !(prefixed_path_strv_contains(c
->inaccessible_paths
, "/tmp") &&
6215 (prefixed_path_strv_contains(c
->inaccessible_paths
, "/var/tmp") ||
6216 prefixed_path_strv_contains(c
->inaccessible_paths
, "/var")))) {
6217 r
= setup_tmp_dirs(id
, &tmp_dir
, &var_tmp_dir
);
6222 if (c
->private_network
|| c
->network_namespace_path
) {
6223 if (socketpair(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
, 0, netns_storage_socket
) < 0)
6227 if (c
->private_ipc
|| c
->ipc_namespace_path
) {
6228 if (socketpair(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
, 0, ipcns_storage_socket
) < 0)
6232 r
= exec_runtime_add(m
, id
, &tmp_dir
, &var_tmp_dir
, netns_storage_socket
, ipcns_storage_socket
, ret
);
6239 int exec_runtime_acquire(Manager
*m
, const ExecContext
*c
, const char *id
, bool create
, ExecRuntime
**ret
) {
6247 rt
= hashmap_get(m
->exec_runtime_by_id
, id
);
6249 /* We already have an ExecRuntime object, let's increase the ref count and reuse it */
6257 /* If not found, then create a new object. */
6258 r
= exec_runtime_make(m
, c
, id
, &rt
);
6262 /* When r == 0, it is not necessary to create ExecRuntime object. */
6268 /* increment reference counter. */
6274 ExecRuntime
*exec_runtime_unref(ExecRuntime
*rt
, bool destroy
) {
6278 assert(rt
->n_ref
> 0);
6284 return exec_runtime_free(rt
, destroy
);
6287 int exec_runtime_serialize(const Manager
*m
, FILE *f
, FDSet
*fds
) {
6294 HASHMAP_FOREACH(rt
, m
->exec_runtime_by_id
) {
6295 fprintf(f
, "exec-runtime=%s", rt
->id
);
6298 fprintf(f
, " tmp-dir=%s", rt
->tmp_dir
);
6300 if (rt
->var_tmp_dir
)
6301 fprintf(f
, " var-tmp-dir=%s", rt
->var_tmp_dir
);
6303 if (rt
->netns_storage_socket
[0] >= 0) {
6306 copy
= fdset_put_dup(fds
, rt
->netns_storage_socket
[0]);
6310 fprintf(f
, " netns-socket-0=%i", copy
);
6313 if (rt
->netns_storage_socket
[1] >= 0) {
6316 copy
= fdset_put_dup(fds
, rt
->netns_storage_socket
[1]);
6320 fprintf(f
, " netns-socket-1=%i", copy
);
6323 if (rt
->ipcns_storage_socket
[0] >= 0) {
6326 copy
= fdset_put_dup(fds
, rt
->ipcns_storage_socket
[0]);
6330 fprintf(f
, " ipcns-socket-0=%i", copy
);
6333 if (rt
->ipcns_storage_socket
[1] >= 0) {
6336 copy
= fdset_put_dup(fds
, rt
->ipcns_storage_socket
[1]);
6340 fprintf(f
, " ipcns-socket-1=%i", copy
);
6349 int exec_runtime_deserialize_compat(Unit
*u
, const char *key
, const char *value
, FDSet
*fds
) {
6350 _cleanup_(exec_runtime_freep
) ExecRuntime
*rt_create
= NULL
;
6354 /* This is for the migration from old (v237 or earlier) deserialization text.
6355 * Due to the bug #7790, this may not work with the units that use JoinsNamespaceOf=.
6356 * Even if the ExecRuntime object originally created by the other unit, we cannot judge
6357 * so or not from the serialized text, then we always creates a new object owned by this. */
6363 /* Manager manages ExecRuntime objects by the unit id.
6364 * So, we omit the serialized text when the unit does not have id (yet?)... */
6365 if (isempty(u
->id
)) {
6366 log_unit_debug(u
, "Invocation ID not found. Dropping runtime parameter.");
6370 if (hashmap_ensure_allocated(&u
->manager
->exec_runtime_by_id
, &string_hash_ops
) < 0)
6373 rt
= hashmap_get(u
->manager
->exec_runtime_by_id
, u
->id
);
6375 if (exec_runtime_allocate(&rt_create
, u
->id
) < 0)
6381 if (streq(key
, "tmp-dir")) {
6382 if (free_and_strdup_warn(&rt
->tmp_dir
, value
) < 0)
6385 } else if (streq(key
, "var-tmp-dir")) {
6386 if (free_and_strdup_warn(&rt
->var_tmp_dir
, value
) < 0)
6389 } else if (streq(key
, "netns-socket-0")) {
6392 if (safe_atoi(value
, &fd
) < 0 || !fdset_contains(fds
, fd
)) {
6393 log_unit_debug(u
, "Failed to parse netns socket value: %s", value
);
6397 safe_close(rt
->netns_storage_socket
[0]);
6398 rt
->netns_storage_socket
[0] = fdset_remove(fds
, fd
);
6400 } else if (streq(key
, "netns-socket-1")) {
6403 if (safe_atoi(value
, &fd
) < 0 || !fdset_contains(fds
, fd
)) {
6404 log_unit_debug(u
, "Failed to parse netns socket value: %s", value
);
6408 safe_close(rt
->netns_storage_socket
[1]);
6409 rt
->netns_storage_socket
[1] = fdset_remove(fds
, fd
);
6414 /* If the object is newly created, then put it to the hashmap which manages ExecRuntime objects. */
6416 r
= hashmap_put(u
->manager
->exec_runtime_by_id
, rt_create
->id
, rt_create
);
6418 log_unit_debug_errno(u
, r
, "Failed to put runtime parameter to manager's storage: %m");
6422 rt_create
->manager
= u
->manager
;
6425 TAKE_PTR(rt_create
);
6431 int exec_runtime_deserialize_one(Manager
*m
, const char *value
, FDSet
*fds
) {
6432 _cleanup_free_
char *tmp_dir
= NULL
, *var_tmp_dir
= NULL
;
6434 int r
, netns_fdpair
[] = {-1, -1}, ipcns_fdpair
[] = {-1, -1};
6435 const char *p
, *v
= value
;
6442 n
= strcspn(v
, " ");
6443 id
= strndupa(v
, n
);
6448 v
= startswith(p
, "tmp-dir=");
6450 n
= strcspn(v
, " ");
6451 tmp_dir
= strndup(v
, n
);
6459 v
= startswith(p
, "var-tmp-dir=");
6461 n
= strcspn(v
, " ");
6462 var_tmp_dir
= strndup(v
, n
);
6470 v
= startswith(p
, "netns-socket-0=");
6474 n
= strcspn(v
, " ");
6475 buf
= strndupa(v
, n
);
6477 r
= safe_atoi(buf
, &netns_fdpair
[0]);
6479 return log_debug_errno(r
, "Unable to parse exec-runtime specification netns-socket-0=%s: %m", buf
);
6480 if (!fdset_contains(fds
, netns_fdpair
[0]))
6481 return log_debug_errno(SYNTHETIC_ERRNO(EBADF
),
6482 "exec-runtime specification netns-socket-0= refers to unknown fd %d: %m", netns_fdpair
[0]);
6483 netns_fdpair
[0] = fdset_remove(fds
, netns_fdpair
[0]);
6489 v
= startswith(p
, "netns-socket-1=");
6493 n
= strcspn(v
, " ");
6494 buf
= strndupa(v
, n
);
6496 r
= safe_atoi(buf
, &netns_fdpair
[1]);
6498 return log_debug_errno(r
, "Unable to parse exec-runtime specification netns-socket-1=%s: %m", buf
);
6499 if (!fdset_contains(fds
, netns_fdpair
[1]))
6500 return log_debug_errno(SYNTHETIC_ERRNO(EBADF
),
6501 "exec-runtime specification netns-socket-1= refers to unknown fd %d: %m", netns_fdpair
[1]);
6502 netns_fdpair
[1] = fdset_remove(fds
, netns_fdpair
[1]);
6508 v
= startswith(p
, "ipcns-socket-0=");
6512 n
= strcspn(v
, " ");
6513 buf
= strndupa(v
, n
);
6515 r
= safe_atoi(buf
, &ipcns_fdpair
[0]);
6517 return log_debug_errno(r
, "Unable to parse exec-runtime specification ipcns-socket-0=%s: %m", buf
);
6518 if (!fdset_contains(fds
, ipcns_fdpair
[0]))
6519 return log_debug_errno(SYNTHETIC_ERRNO(EBADF
),
6520 "exec-runtime specification ipcns-socket-0= refers to unknown fd %d: %m", ipcns_fdpair
[0]);
6521 ipcns_fdpair
[0] = fdset_remove(fds
, ipcns_fdpair
[0]);
6527 v
= startswith(p
, "ipcns-socket-1=");
6531 n
= strcspn(v
, " ");
6532 buf
= strndupa(v
, n
);
6534 r
= safe_atoi(buf
, &ipcns_fdpair
[1]);
6536 return log_debug_errno(r
, "Unable to parse exec-runtime specification ipcns-socket-1=%s: %m", buf
);
6537 if (!fdset_contains(fds
, ipcns_fdpair
[1]))
6538 return log_debug_errno(SYNTHETIC_ERRNO(EBADF
),
6539 "exec-runtime specification ipcns-socket-1= refers to unknown fd %d: %m", ipcns_fdpair
[1]);
6540 ipcns_fdpair
[1] = fdset_remove(fds
, ipcns_fdpair
[1]);
6544 r
= exec_runtime_add(m
, id
, &tmp_dir
, &var_tmp_dir
, netns_fdpair
, ipcns_fdpair
, NULL
);
6546 return log_debug_errno(r
, "Failed to add exec-runtime: %m");
6550 void exec_runtime_vacuum(Manager
*m
) {
6555 /* Free unreferenced ExecRuntime objects. This is used after manager deserialization process. */
6557 HASHMAP_FOREACH(rt
, m
->exec_runtime_by_id
) {
6561 (void) exec_runtime_free(rt
, false);
6565 void exec_params_clear(ExecParameters
*p
) {
6569 p
->environment
= strv_free(p
->environment
);
6570 p
->fd_names
= strv_free(p
->fd_names
);
6571 p
->fds
= mfree(p
->fds
);
6572 p
->exec_fd
= safe_close(p
->exec_fd
);
6575 ExecSetCredential
*exec_set_credential_free(ExecSetCredential
*sc
) {
6584 ExecLoadCredential
*exec_load_credential_free(ExecLoadCredential
*lc
) {
6593 DEFINE_HASH_OPS_WITH_VALUE_DESTRUCTOR(exec_set_credential_hash_ops
, char, string_hash_func
, string_compare_func
, ExecSetCredential
, exec_set_credential_free
);
6594 DEFINE_HASH_OPS_WITH_VALUE_DESTRUCTOR(exec_load_credential_hash_ops
, char, string_hash_func
, string_compare_func
, ExecLoadCredential
, exec_load_credential_free
);
6596 static const char* const exec_input_table
[_EXEC_INPUT_MAX
] = {
6597 [EXEC_INPUT_NULL
] = "null",
6598 [EXEC_INPUT_TTY
] = "tty",
6599 [EXEC_INPUT_TTY_FORCE
] = "tty-force",
6600 [EXEC_INPUT_TTY_FAIL
] = "tty-fail",
6601 [EXEC_INPUT_SOCKET
] = "socket",
6602 [EXEC_INPUT_NAMED_FD
] = "fd",
6603 [EXEC_INPUT_DATA
] = "data",
6604 [EXEC_INPUT_FILE
] = "file",
6607 DEFINE_STRING_TABLE_LOOKUP(exec_input
, ExecInput
);
6609 static const char* const exec_output_table
[_EXEC_OUTPUT_MAX
] = {
6610 [EXEC_OUTPUT_INHERIT
] = "inherit",
6611 [EXEC_OUTPUT_NULL
] = "null",
6612 [EXEC_OUTPUT_TTY
] = "tty",
6613 [EXEC_OUTPUT_KMSG
] = "kmsg",
6614 [EXEC_OUTPUT_KMSG_AND_CONSOLE
] = "kmsg+console",
6615 [EXEC_OUTPUT_JOURNAL
] = "journal",
6616 [EXEC_OUTPUT_JOURNAL_AND_CONSOLE
] = "journal+console",
6617 [EXEC_OUTPUT_SOCKET
] = "socket",
6618 [EXEC_OUTPUT_NAMED_FD
] = "fd",
6619 [EXEC_OUTPUT_FILE
] = "file",
6620 [EXEC_OUTPUT_FILE_APPEND
] = "append",
6621 [EXEC_OUTPUT_FILE_TRUNCATE
] = "truncate",
6624 DEFINE_STRING_TABLE_LOOKUP(exec_output
, ExecOutput
);
6626 static const char* const exec_utmp_mode_table
[_EXEC_UTMP_MODE_MAX
] = {
6627 [EXEC_UTMP_INIT
] = "init",
6628 [EXEC_UTMP_LOGIN
] = "login",
6629 [EXEC_UTMP_USER
] = "user",
6632 DEFINE_STRING_TABLE_LOOKUP(exec_utmp_mode
, ExecUtmpMode
);
6634 static const char* const exec_preserve_mode_table
[_EXEC_PRESERVE_MODE_MAX
] = {
6635 [EXEC_PRESERVE_NO
] = "no",
6636 [EXEC_PRESERVE_YES
] = "yes",
6637 [EXEC_PRESERVE_RESTART
] = "restart",
6640 DEFINE_STRING_TABLE_LOOKUP_WITH_BOOLEAN(exec_preserve_mode
, ExecPreserveMode
, EXEC_PRESERVE_YES
);
6642 /* This table maps ExecDirectoryType to the setting it is configured with in the unit */
6643 static const char* const exec_directory_type_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
6644 [EXEC_DIRECTORY_RUNTIME
] = "RuntimeDirectory",
6645 [EXEC_DIRECTORY_STATE
] = "StateDirectory",
6646 [EXEC_DIRECTORY_CACHE
] = "CacheDirectory",
6647 [EXEC_DIRECTORY_LOGS
] = "LogsDirectory",
6648 [EXEC_DIRECTORY_CONFIGURATION
] = "ConfigurationDirectory",
6651 DEFINE_STRING_TABLE_LOOKUP(exec_directory_type
, ExecDirectoryType
);
6653 /* And this table maps ExecDirectoryType too, but to a generic term identifying the type of resource. This
6654 * one is supposed to be generic enough to be used for unit types that don't use ExecContext and per-unit
6655 * directories, specifically .timer units with their timestamp touch file. */
6656 static const char* const exec_resource_type_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
6657 [EXEC_DIRECTORY_RUNTIME
] = "runtime",
6658 [EXEC_DIRECTORY_STATE
] = "state",
6659 [EXEC_DIRECTORY_CACHE
] = "cache",
6660 [EXEC_DIRECTORY_LOGS
] = "logs",
6661 [EXEC_DIRECTORY_CONFIGURATION
] = "configuration",
6664 DEFINE_STRING_TABLE_LOOKUP(exec_resource_type
, ExecDirectoryType
);
6666 /* And this table also maps ExecDirectoryType, to the environment variable we pass the selected directory to
6667 * the service payload in. */
6668 static const char* const exec_directory_env_name_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
6669 [EXEC_DIRECTORY_RUNTIME
] = "RUNTIME_DIRECTORY",
6670 [EXEC_DIRECTORY_STATE
] = "STATE_DIRECTORY",
6671 [EXEC_DIRECTORY_CACHE
] = "CACHE_DIRECTORY",
6672 [EXEC_DIRECTORY_LOGS
] = "LOGS_DIRECTORY",
6673 [EXEC_DIRECTORY_CONFIGURATION
] = "CONFIGURATION_DIRECTORY",
6676 DEFINE_PRIVATE_STRING_TABLE_LOOKUP_TO_STRING(exec_directory_env_name
, ExecDirectoryType
);
6678 static const char* const exec_keyring_mode_table
[_EXEC_KEYRING_MODE_MAX
] = {
6679 [EXEC_KEYRING_INHERIT
] = "inherit",
6680 [EXEC_KEYRING_PRIVATE
] = "private",
6681 [EXEC_KEYRING_SHARED
] = "shared",
6684 DEFINE_STRING_TABLE_LOOKUP(exec_keyring_mode
, ExecKeyringMode
);