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>
18 #include <linux/fs.h> /* Must be included after <sys/mount.h> */
21 #include <security/pam_appl.h>
25 #include <selinux/selinux.h>
33 #include <sys/apparmor.h>
36 #include "sd-messages.h"
40 #include "alloc-util.h"
42 #include "apparmor-util.h"
44 #include "argv-util.h"
48 #include "btrfs-util.h"
50 #include "capability-util.h"
51 #include "chattr-util.h"
52 #include "cgroup-setup.h"
54 #include "chown-recursive.h"
55 #include "constants.h"
56 #include "cpu-set-util.h"
57 #include "creds-util.h"
58 #include "data-fd-util.h"
61 #include "errno-list.h"
64 #include "exit-status.h"
67 #include "format-util.h"
68 #include "glob-util.h"
69 #include "hexdecoct.h"
71 #include "ioprio-util.h"
72 #include "label-util.h"
73 #include "lock-util.h"
77 #include "manager-dump.h"
78 #include "memory-util.h"
79 #include "missing_fs.h"
80 #include "missing_ioprio.h"
81 #include "missing_prctl.h"
82 #include "mkdir-label.h"
83 #include "mount-util.h"
84 #include "mountpoint-util.h"
85 #include "namespace.h"
86 #include "parse-util.h"
87 #include "path-util.h"
88 #include "proc-cmdline.h"
89 #include "process-util.h"
91 #include "random-util.h"
92 #include "recurse-dir.h"
93 #include "rlimit-util.h"
96 #include "seccomp-util.h"
98 #include "securebits-util.h"
99 #include "selinux-util.h"
100 #include "signal-util.h"
101 #include "smack-util.h"
102 #include "socket-util.h"
103 #include "sort-util.h"
105 #include "stat-util.h"
106 #include "string-table.h"
107 #include "string-util.h"
109 #include "syslog-util.h"
110 #include "terminal-util.h"
111 #include "tmpfile-util.h"
112 #include "umask-util.h"
113 #include "unit-serialize.h"
114 #include "user-util.h"
115 #include "utmp-wtmp.h"
117 #define IDLE_TIMEOUT_USEC (5*USEC_PER_SEC)
118 #define IDLE_TIMEOUT2_USEC (1*USEC_PER_SEC)
120 #define SNDBUF_SIZE (8*1024*1024)
122 static int shift_fds(int fds
[], size_t n_fds
) {
126 /* Modifies the fds array! (sorts it) */
130 for (int start
= 0;;) {
131 int restart_from
= -1;
133 for (int i
= start
; i
< (int) n_fds
; i
++) {
136 /* Already at right index? */
140 nfd
= fcntl(fds
[i
], F_DUPFD
, i
+ 3);
147 /* Hmm, the fd we wanted isn't free? Then
148 * let's remember that and try again from here */
149 if (nfd
!= i
+3 && restart_from
< 0)
153 if (restart_from
< 0)
156 start
= restart_from
;
162 static int flags_fds(
175 /* Drops/Sets O_NONBLOCK and FD_CLOEXEC from the file flags.
176 * O_NONBLOCK only applies to socket activation though. */
178 for (size_t i
= 0; i
< n_fds
; i
++) {
180 if (i
< n_socket_fds
) {
181 r
= fd_nonblock(fds
[i
], nonblock
);
186 /* We unconditionally drop FD_CLOEXEC from the fds,
187 * since after all we want to pass these fds to our
190 r
= fd_cloexec(fds
[i
], false);
198 static const char *exec_context_tty_path(const ExecContext
*context
) {
201 if (context
->stdio_as_fds
)
204 if (context
->tty_path
)
205 return context
->tty_path
;
207 return "/dev/console";
210 static int exec_context_tty_size(const ExecContext
*context
, unsigned *ret_rows
, unsigned *ret_cols
) {
218 rows
= context
->tty_rows
;
219 cols
= context
->tty_cols
;
221 tty
= exec_context_tty_path(context
);
223 (void) proc_cmdline_tty_size(tty
, rows
== UINT_MAX
? &rows
: NULL
, cols
== UINT_MAX
? &cols
: NULL
);
231 static void exec_context_tty_reset(const ExecContext
*context
, const ExecParameters
*p
) {
232 _cleanup_close_
int fd
= -EBADF
;
233 const char *path
= exec_context_tty_path(ASSERT_PTR(context
));
235 /* Take a lock around the device for the duration of the setup that we do here.
236 * systemd-vconsole-setup.service also takes the lock to avoid being interrupted.
237 * We open a new fd that will be closed automatically, and operate on it for convenience.
240 if (p
&& p
->stdin_fd
>= 0) {
241 fd
= xopenat_lock(p
->stdin_fd
, NULL
,
242 O_RDONLY
|O_CLOEXEC
|O_NONBLOCK
|O_NOCTTY
, 0, 0, LOCK_BSD
, LOCK_EX
);
246 fd
= open_terminal(path
, O_RDWR
|O_NOCTTY
|O_CLOEXEC
|O_NONBLOCK
);
250 if (lock_generic(fd
, LOCK_BSD
, LOCK_EX
) < 0)
253 return; /* nothing to do */
255 if (context
->tty_vhangup
)
256 (void) terminal_vhangup_fd(fd
);
258 if (context
->tty_reset
)
259 (void) reset_terminal_fd(fd
, true);
261 if (p
&& p
->stdin_fd
>= 0) {
262 unsigned rows
= context
->tty_rows
, cols
= context
->tty_cols
;
264 (void) exec_context_tty_size(context
, &rows
, &cols
);
265 (void) terminal_set_size_fd(p
->stdin_fd
, path
, rows
, cols
);
268 if (context
->tty_vt_disallocate
&& path
)
269 (void) vt_disallocate(path
);
272 static bool is_terminal_input(ExecInput i
) {
275 EXEC_INPUT_TTY_FORCE
,
276 EXEC_INPUT_TTY_FAIL
);
279 static bool is_terminal_output(ExecOutput o
) {
282 EXEC_OUTPUT_KMSG_AND_CONSOLE
,
283 EXEC_OUTPUT_JOURNAL_AND_CONSOLE
);
286 static bool is_kmsg_output(ExecOutput o
) {
289 EXEC_OUTPUT_KMSG_AND_CONSOLE
);
292 static bool exec_context_needs_term(const ExecContext
*c
) {
295 /* Return true if the execution context suggests we should set $TERM to something useful. */
297 if (is_terminal_input(c
->std_input
))
300 if (is_terminal_output(c
->std_output
))
303 if (is_terminal_output(c
->std_error
))
306 return !!c
->tty_path
;
309 static int open_null_as(int flags
, int nfd
) {
314 fd
= open("/dev/null", flags
|O_NOCTTY
);
318 return move_fd(fd
, nfd
, false);
321 static int connect_journal_socket(
323 const char *log_namespace
,
327 uid_t olduid
= UID_INVALID
;
328 gid_t oldgid
= GID_INVALID
;
333 strjoina("/run/systemd/journal.", log_namespace
, "/stdout") :
334 "/run/systemd/journal/stdout";
336 if (gid_is_valid(gid
)) {
339 if (setegid(gid
) < 0)
343 if (uid_is_valid(uid
)) {
346 if (seteuid(uid
) < 0) {
352 r
= connect_unix_path(fd
, AT_FDCWD
, j
);
354 /* If we fail to restore the uid or gid, things will likely fail later on. This should only happen if
355 an LSM interferes. */
357 if (uid_is_valid(uid
))
358 (void) seteuid(olduid
);
361 if (gid_is_valid(gid
))
362 (void) setegid(oldgid
);
367 static int connect_logger_as(
369 const ExecContext
*context
,
370 const ExecParameters
*params
,
377 _cleanup_close_
int fd
= -EBADF
;
382 assert(output
< _EXEC_OUTPUT_MAX
);
386 fd
= socket(AF_UNIX
, SOCK_STREAM
, 0);
390 r
= connect_journal_socket(fd
, context
->log_namespace
, uid
, gid
);
394 if (shutdown(fd
, SHUT_RD
) < 0)
397 (void) fd_inc_sndbuf(fd
, SNDBUF_SIZE
);
407 context
->syslog_identifier
?: ident
,
408 params
->flags
& EXEC_PASS_LOG_UNIT
? unit
->id
: "",
409 context
->syslog_priority
,
410 !!context
->syslog_level_prefix
,
412 is_kmsg_output(output
),
413 is_terminal_output(output
)) < 0)
416 return move_fd(TAKE_FD(fd
), nfd
, false);
419 static int open_terminal_as(const char *path
, int flags
, int nfd
) {
425 fd
= open_terminal(path
, flags
| O_NOCTTY
);
429 return move_fd(fd
, nfd
, false);
432 static int acquire_path(const char *path
, int flags
, mode_t mode
) {
433 _cleanup_close_
int fd
= -EBADF
;
438 if (IN_SET(flags
& O_ACCMODE
, O_WRONLY
, O_RDWR
))
441 fd
= open(path
, flags
|O_NOCTTY
, mode
);
445 if (errno
!= ENXIO
) /* ENXIO is returned when we try to open() an AF_UNIX file system socket on Linux */
448 /* So, it appears the specified path could be an AF_UNIX socket. Let's see if we can connect to it. */
450 fd
= socket(AF_UNIX
, SOCK_STREAM
, 0);
454 r
= connect_unix_path(fd
, AT_FDCWD
, path
);
455 if (IN_SET(r
, -ENOTSOCK
, -EINVAL
))
456 /* Propagate initial error if we get ENOTSOCK or EINVAL, i.e. we have indication that this
457 * wasn't an AF_UNIX socket after all */
462 if ((flags
& O_ACCMODE
) == O_RDONLY
)
463 r
= shutdown(fd
, SHUT_WR
);
464 else if ((flags
& O_ACCMODE
) == O_WRONLY
)
465 r
= shutdown(fd
, SHUT_RD
);
474 static int fixup_input(
475 const ExecContext
*context
,
477 bool apply_tty_stdin
) {
483 std_input
= context
->std_input
;
485 if (is_terminal_input(std_input
) && !apply_tty_stdin
)
486 return EXEC_INPUT_NULL
;
488 if (std_input
== EXEC_INPUT_SOCKET
&& socket_fd
< 0)
489 return EXEC_INPUT_NULL
;
491 if (std_input
== EXEC_INPUT_DATA
&& context
->stdin_data_size
== 0)
492 return EXEC_INPUT_NULL
;
497 static int fixup_output(ExecOutput output
, int socket_fd
) {
499 if (output
== EXEC_OUTPUT_SOCKET
&& socket_fd
< 0)
500 return EXEC_OUTPUT_INHERIT
;
505 static int setup_input(
506 const ExecContext
*context
,
507 const ExecParameters
*params
,
509 const int named_iofds
[static 3]) {
518 if (params
->stdin_fd
>= 0) {
519 if (dup2(params
->stdin_fd
, STDIN_FILENO
) < 0)
522 /* Try to make this the controlling tty, if it is a tty, and reset it */
523 if (isatty(STDIN_FILENO
)) {
524 unsigned rows
= context
->tty_rows
, cols
= context
->tty_cols
;
526 (void) exec_context_tty_size(context
, &rows
, &cols
);
527 (void) ioctl(STDIN_FILENO
, TIOCSCTTY
, context
->std_input
== EXEC_INPUT_TTY_FORCE
);
528 (void) reset_terminal_fd(STDIN_FILENO
, true);
529 (void) terminal_set_size_fd(STDIN_FILENO
, NULL
, rows
, cols
);
535 i
= fixup_input(context
, socket_fd
, params
->flags
& EXEC_APPLY_TTY_STDIN
);
539 case EXEC_INPUT_NULL
:
540 return open_null_as(O_RDONLY
, STDIN_FILENO
);
543 case EXEC_INPUT_TTY_FORCE
:
544 case EXEC_INPUT_TTY_FAIL
: {
548 fd
= acquire_terminal(exec_context_tty_path(context
),
549 i
== EXEC_INPUT_TTY_FAIL
? ACQUIRE_TERMINAL_TRY
:
550 i
== EXEC_INPUT_TTY_FORCE
? ACQUIRE_TERMINAL_FORCE
:
551 ACQUIRE_TERMINAL_WAIT
,
556 r
= exec_context_tty_size(context
, &rows
, &cols
);
560 r
= terminal_set_size_fd(fd
, exec_context_tty_path(context
), rows
, cols
);
564 return move_fd(fd
, STDIN_FILENO
, false);
567 case EXEC_INPUT_SOCKET
:
568 assert(socket_fd
>= 0);
570 return RET_NERRNO(dup2(socket_fd
, STDIN_FILENO
));
572 case EXEC_INPUT_NAMED_FD
:
573 assert(named_iofds
[STDIN_FILENO
] >= 0);
575 (void) fd_nonblock(named_iofds
[STDIN_FILENO
], false);
576 return RET_NERRNO(dup2(named_iofds
[STDIN_FILENO
], STDIN_FILENO
));
578 case EXEC_INPUT_DATA
: {
581 fd
= acquire_data_fd(context
->stdin_data
, context
->stdin_data_size
, 0);
585 return move_fd(fd
, STDIN_FILENO
, false);
588 case EXEC_INPUT_FILE
: {
592 assert(context
->stdio_file
[STDIN_FILENO
]);
594 rw
= (context
->std_output
== EXEC_OUTPUT_FILE
&& streq_ptr(context
->stdio_file
[STDIN_FILENO
], context
->stdio_file
[STDOUT_FILENO
])) ||
595 (context
->std_error
== EXEC_OUTPUT_FILE
&& streq_ptr(context
->stdio_file
[STDIN_FILENO
], context
->stdio_file
[STDERR_FILENO
]));
597 fd
= acquire_path(context
->stdio_file
[STDIN_FILENO
], rw
? O_RDWR
: O_RDONLY
, 0666 & ~context
->umask
);
601 return move_fd(fd
, STDIN_FILENO
, false);
605 assert_not_reached();
609 static bool can_inherit_stderr_from_stdout(
610 const ExecContext
*context
,
616 /* Returns true, if given the specified STDERR and STDOUT output we can directly dup() the stdout fd to the
619 if (e
== EXEC_OUTPUT_INHERIT
)
624 if (e
== EXEC_OUTPUT_NAMED_FD
)
625 return streq_ptr(context
->stdio_fdname
[STDOUT_FILENO
], context
->stdio_fdname
[STDERR_FILENO
]);
627 if (IN_SET(e
, EXEC_OUTPUT_FILE
, EXEC_OUTPUT_FILE_APPEND
, EXEC_OUTPUT_FILE_TRUNCATE
))
628 return streq_ptr(context
->stdio_file
[STDOUT_FILENO
], context
->stdio_file
[STDERR_FILENO
]);
633 static int setup_output(
635 const ExecContext
*context
,
636 const ExecParameters
*params
,
639 const int named_iofds
[static 3],
643 dev_t
*journal_stream_dev
,
644 ino_t
*journal_stream_ino
) {
654 assert(journal_stream_dev
);
655 assert(journal_stream_ino
);
657 if (fileno
== STDOUT_FILENO
&& params
->stdout_fd
>= 0) {
659 if (dup2(params
->stdout_fd
, STDOUT_FILENO
) < 0)
662 return STDOUT_FILENO
;
665 if (fileno
== STDERR_FILENO
&& params
->stderr_fd
>= 0) {
666 if (dup2(params
->stderr_fd
, STDERR_FILENO
) < 0)
669 return STDERR_FILENO
;
672 i
= fixup_input(context
, socket_fd
, params
->flags
& EXEC_APPLY_TTY_STDIN
);
673 o
= fixup_output(context
->std_output
, socket_fd
);
675 if (fileno
== STDERR_FILENO
) {
677 e
= fixup_output(context
->std_error
, socket_fd
);
679 /* This expects the input and output are already set up */
681 /* Don't change the stderr file descriptor if we inherit all
682 * the way and are not on a tty */
683 if (e
== EXEC_OUTPUT_INHERIT
&&
684 o
== EXEC_OUTPUT_INHERIT
&&
685 i
== EXEC_INPUT_NULL
&&
686 !is_terminal_input(context
->std_input
) &&
690 /* Duplicate from stdout if possible */
691 if (can_inherit_stderr_from_stdout(context
, o
, e
))
692 return RET_NERRNO(dup2(STDOUT_FILENO
, fileno
));
696 } else if (o
== EXEC_OUTPUT_INHERIT
) {
697 /* If input got downgraded, inherit the original value */
698 if (i
== EXEC_INPUT_NULL
&& is_terminal_input(context
->std_input
))
699 return open_terminal_as(exec_context_tty_path(context
), O_WRONLY
, fileno
);
701 /* If the input is connected to anything that's not a /dev/null or a data fd, inherit that... */
702 if (!IN_SET(i
, EXEC_INPUT_NULL
, EXEC_INPUT_DATA
))
703 return RET_NERRNO(dup2(STDIN_FILENO
, fileno
));
705 /* If we are not started from PID 1 we just inherit STDOUT from our parent process. */
709 /* We need to open /dev/null here anew, to get the right access mode. */
710 return open_null_as(O_WRONLY
, fileno
);
715 case EXEC_OUTPUT_NULL
:
716 return open_null_as(O_WRONLY
, fileno
);
718 case EXEC_OUTPUT_TTY
:
719 if (is_terminal_input(i
))
720 return RET_NERRNO(dup2(STDIN_FILENO
, fileno
));
722 /* We don't reset the terminal if this is just about output */
723 return open_terminal_as(exec_context_tty_path(context
), O_WRONLY
, fileno
);
725 case EXEC_OUTPUT_KMSG
:
726 case EXEC_OUTPUT_KMSG_AND_CONSOLE
:
727 case EXEC_OUTPUT_JOURNAL
:
728 case EXEC_OUTPUT_JOURNAL_AND_CONSOLE
:
729 r
= connect_logger_as(unit
, context
, params
, o
, ident
, fileno
, uid
, gid
);
731 log_unit_warning_errno(unit
, r
, "Failed to connect %s to the journal socket, ignoring: %m",
732 fileno
== STDOUT_FILENO
? "stdout" : "stderr");
733 r
= open_null_as(O_WRONLY
, fileno
);
737 /* If we connected this fd to the journal via a stream, patch the device/inode into the passed
738 * parameters, but only then. This is useful so that we can set $JOURNAL_STREAM that permits
739 * services to detect whether they are connected to the journal or not.
741 * If both stdout and stderr are connected to a stream then let's make sure to store the data
742 * about STDERR as that's usually the best way to do logging. */
744 if (fstat(fileno
, &st
) >= 0 &&
745 (*journal_stream_ino
== 0 || fileno
== STDERR_FILENO
)) {
746 *journal_stream_dev
= st
.st_dev
;
747 *journal_stream_ino
= st
.st_ino
;
752 case EXEC_OUTPUT_SOCKET
:
753 assert(socket_fd
>= 0);
755 return RET_NERRNO(dup2(socket_fd
, fileno
));
757 case EXEC_OUTPUT_NAMED_FD
:
758 assert(named_iofds
[fileno
] >= 0);
760 (void) fd_nonblock(named_iofds
[fileno
], false);
761 return RET_NERRNO(dup2(named_iofds
[fileno
], fileno
));
763 case EXEC_OUTPUT_FILE
:
764 case EXEC_OUTPUT_FILE_APPEND
:
765 case EXEC_OUTPUT_FILE_TRUNCATE
: {
769 assert(context
->stdio_file
[fileno
]);
771 rw
= context
->std_input
== EXEC_INPUT_FILE
&&
772 streq_ptr(context
->stdio_file
[fileno
], context
->stdio_file
[STDIN_FILENO
]);
775 return RET_NERRNO(dup2(STDIN_FILENO
, fileno
));
778 if (o
== EXEC_OUTPUT_FILE_APPEND
)
780 else if (o
== EXEC_OUTPUT_FILE_TRUNCATE
)
783 fd
= acquire_path(context
->stdio_file
[fileno
], flags
, 0666 & ~context
->umask
);
787 return move_fd(fd
, fileno
, 0);
791 assert_not_reached();
795 static int chown_terminal(int fd
, uid_t uid
) {
800 /* Before we chown/chmod the TTY, let's ensure this is actually a tty */
801 if (isatty(fd
) < 1) {
802 if (IN_SET(errno
, EINVAL
, ENOTTY
))
803 return 0; /* not a tty */
808 /* This might fail. What matters are the results. */
809 r
= fchmod_and_chown(fd
, TTY_MODE
, uid
, GID_INVALID
);
816 static int setup_confirm_stdio(
817 const ExecContext
*context
,
819 int *ret_saved_stdin
,
820 int *ret_saved_stdout
) {
822 _cleanup_close_
int fd
= -EBADF
, saved_stdin
= -EBADF
, saved_stdout
= -EBADF
;
826 assert(ret_saved_stdin
);
827 assert(ret_saved_stdout
);
829 saved_stdin
= fcntl(STDIN_FILENO
, F_DUPFD
, 3);
833 saved_stdout
= fcntl(STDOUT_FILENO
, F_DUPFD
, 3);
834 if (saved_stdout
< 0)
837 fd
= acquire_terminal(vc
, ACQUIRE_TERMINAL_WAIT
, DEFAULT_CONFIRM_USEC
);
841 r
= chown_terminal(fd
, getuid());
845 r
= reset_terminal_fd(fd
, true);
849 r
= exec_context_tty_size(context
, &rows
, &cols
);
853 r
= terminal_set_size_fd(fd
, vc
, rows
, cols
);
857 r
= rearrange_stdio(fd
, fd
, STDERR_FILENO
); /* Invalidates 'fd' also on failure */
862 *ret_saved_stdin
= TAKE_FD(saved_stdin
);
863 *ret_saved_stdout
= TAKE_FD(saved_stdout
);
867 static void write_confirm_error_fd(int err
, int fd
, const Unit
*u
) {
870 if (err
== -ETIMEDOUT
)
871 dprintf(fd
, "Confirmation question timed out for %s, assuming positive response.\n", u
->id
);
874 dprintf(fd
, "Couldn't ask confirmation for %s: %m, assuming positive response.\n", u
->id
);
878 static void write_confirm_error(int err
, const char *vc
, const Unit
*u
) {
879 _cleanup_close_
int fd
= -EBADF
;
883 fd
= open_terminal(vc
, O_WRONLY
|O_NOCTTY
|O_CLOEXEC
);
887 write_confirm_error_fd(err
, fd
, u
);
890 static int restore_confirm_stdio(int *saved_stdin
, int *saved_stdout
) {
894 assert(saved_stdout
);
898 if (*saved_stdin
>= 0)
899 if (dup2(*saved_stdin
, STDIN_FILENO
) < 0)
902 if (*saved_stdout
>= 0)
903 if (dup2(*saved_stdout
, STDOUT_FILENO
) < 0)
906 *saved_stdin
= safe_close(*saved_stdin
);
907 *saved_stdout
= safe_close(*saved_stdout
);
913 CONFIRM_PRETEND_FAILURE
= -1,
914 CONFIRM_PRETEND_SUCCESS
= 0,
918 static int ask_for_confirmation(const ExecContext
*context
, const char *vc
, Unit
*u
, const char *cmdline
) {
919 int saved_stdout
= -1, saved_stdin
= -1, r
;
920 _cleanup_free_
char *e
= NULL
;
923 /* For any internal errors, assume a positive response. */
924 r
= setup_confirm_stdio(context
, vc
, &saved_stdin
, &saved_stdout
);
926 write_confirm_error(r
, vc
, u
);
927 return CONFIRM_EXECUTE
;
930 /* confirm_spawn might have been disabled while we were sleeping. */
931 if (manager_is_confirm_spawn_disabled(u
->manager
)) {
936 e
= ellipsize(cmdline
, 60, 100);
944 r
= ask_char(&c
, "yfshiDjcn", "Execute %s? [y, f, s – h for help] ", e
);
946 write_confirm_error_fd(r
, STDOUT_FILENO
, u
);
953 printf("Resuming normal execution.\n");
954 manager_disable_confirm_spawn();
958 unit_dump(u
, stdout
, " ");
959 continue; /* ask again */
961 printf("Failing execution.\n");
962 r
= CONFIRM_PRETEND_FAILURE
;
965 printf(" c - continue, proceed without asking anymore\n"
966 " D - dump, show the state of the unit\n"
967 " f - fail, don't execute the command and pretend it failed\n"
969 " i - info, show a short summary of the unit\n"
970 " j - jobs, show jobs that are in progress\n"
971 " s - skip, don't execute the command and pretend it succeeded\n"
972 " y - yes, execute the command\n");
973 continue; /* ask again */
975 printf(" Description: %s\n"
978 u
->id
, u
->description
, cmdline
);
979 continue; /* ask again */
981 manager_dump_jobs(u
->manager
, stdout
, /* patterns= */ NULL
, " ");
982 continue; /* ask again */
984 /* 'n' was removed in favor of 'f'. */
985 printf("Didn't understand 'n', did you mean 'f'?\n");
986 continue; /* ask again */
988 printf("Skipping execution.\n");
989 r
= CONFIRM_PRETEND_SUCCESS
;
995 assert_not_reached();
1001 restore_confirm_stdio(&saved_stdin
, &saved_stdout
);
1005 static int get_fixed_user(const ExecContext
*c
, const char **user
,
1006 uid_t
*uid
, gid_t
*gid
,
1007 const char **home
, const char **shell
) {
1016 /* Note that we don't set $HOME or $SHELL if they are not particularly enlightening anyway
1017 * (i.e. are "/" or "/bin/nologin"). */
1020 r
= get_user_creds(&name
, uid
, gid
, home
, shell
, USER_CREDS_CLEAN
);
1028 static int get_fixed_group(const ExecContext
*c
, const char **group
, gid_t
*gid
) {
1038 r
= get_group_creds(&name
, gid
, 0);
1046 static int get_supplementary_groups(const ExecContext
*c
, const char *user
,
1047 const char *group
, gid_t gid
,
1048 gid_t
**supplementary_gids
, int *ngids
) {
1051 bool keep_groups
= false;
1052 gid_t
*groups
= NULL
;
1053 _cleanup_free_ gid_t
*l_gids
= NULL
;
1058 * If user is given, then lookup GID and supplementary groups list.
1059 * We avoid NSS lookups for gid=0. Also we have to initialize groups
1060 * here and as early as possible so we keep the list of supplementary
1061 * groups of the caller.
1063 if (user
&& gid_is_valid(gid
) && gid
!= 0) {
1064 /* First step, initialize groups from /etc/groups */
1065 if (initgroups(user
, gid
) < 0)
1071 if (strv_isempty(c
->supplementary_groups
))
1075 * If SupplementaryGroups= was passed then NGROUPS_MAX has to
1076 * be positive, otherwise fail.
1079 ngroups_max
= (int) sysconf(_SC_NGROUPS_MAX
);
1080 if (ngroups_max
<= 0)
1081 return errno_or_else(EOPNOTSUPP
);
1083 l_gids
= new(gid_t
, ngroups_max
);
1089 * Lookup the list of groups that the user belongs to, we
1090 * avoid NSS lookups here too for gid=0.
1093 if (getgrouplist(user
, gid
, l_gids
, &k
) < 0)
1098 STRV_FOREACH(i
, c
->supplementary_groups
) {
1101 if (k
>= ngroups_max
)
1105 r
= get_group_creds(&g
, l_gids
+k
, 0);
1113 * Sets ngids to zero to drop all supplementary groups, happens
1114 * when we are under root and SupplementaryGroups= is empty.
1121 /* Otherwise get the final list of supplementary groups */
1122 groups
= memdup(l_gids
, sizeof(gid_t
) * k
);
1126 *supplementary_gids
= groups
;
1134 static int enforce_groups(gid_t gid
, const gid_t
*supplementary_gids
, int ngids
) {
1137 /* Handle SupplementaryGroups= if it is not empty */
1139 r
= maybe_setgroups(ngids
, supplementary_gids
);
1144 if (gid_is_valid(gid
)) {
1145 /* Then set our gids */
1146 if (setresgid(gid
, gid
, gid
) < 0)
1153 static int set_securebits(unsigned bits
, unsigned mask
) {
1157 current
= prctl(PR_GET_SECUREBITS
);
1161 /* Clear all securebits defined in mask and set bits */
1162 applied
= ((unsigned) current
& ~mask
) | bits
;
1163 if ((unsigned) current
== applied
)
1166 if (prctl(PR_SET_SECUREBITS
, applied
) < 0)
1172 static int enforce_user(
1173 const ExecContext
*context
,
1175 uint64_t capability_ambient_set
) {
1179 if (!uid_is_valid(uid
))
1182 /* Sets (but doesn't look up) the UIS and makes sure we keep the capabilities while doing so. For
1183 * setting secure bits the capability CAP_SETPCAP is required, so we also need keep-caps in this
1186 if ((capability_ambient_set
!= 0 || context
->secure_bits
!= 0) && uid
!= 0) {
1188 /* First step: If we need to keep capabilities but drop privileges we need to make sure we
1189 * keep our caps, while we drop privileges. Add KEEP_CAPS to the securebits */
1190 r
= set_securebits(1U << SECURE_KEEP_CAPS
, 0);
1195 /* Second step: actually set the uids */
1196 if (setresuid(uid
, uid
, uid
) < 0)
1199 /* At this point we should have all necessary capabilities but are otherwise a normal user. However,
1200 * the caps might got corrupted due to the setresuid() so we need clean them up later. This is done
1201 * outside of this call. */
1207 static int null_conv(
1209 const struct pam_message
**msg
,
1210 struct pam_response
**resp
,
1211 void *appdata_ptr
) {
1213 /* We don't support conversations */
1215 return PAM_CONV_ERR
;
1220 static int setup_pam(
1226 char ***env
, /* updated on success */
1227 const int fds
[], size_t n_fds
) {
1231 static const struct pam_conv conv
= {
1236 _cleanup_(barrier_destroy
) Barrier barrier
= BARRIER_NULL
;
1237 _cleanup_strv_free_
char **e
= NULL
;
1238 pam_handle_t
*handle
= NULL
;
1240 int pam_code
= PAM_SUCCESS
, r
;
1241 bool close_session
= false;
1242 pid_t pam_pid
= 0, parent_pid
;
1249 /* We set up PAM in the parent process, then fork. The child
1250 * will then stay around until killed via PR_GET_PDEATHSIG or
1251 * systemd via the cgroup logic. It will then remove the PAM
1252 * session again. The parent process will exec() the actual
1253 * daemon. We do things this way to ensure that the main PID
1254 * of the daemon is the one we initially fork()ed. */
1256 r
= barrier_create(&barrier
);
1260 if (log_get_max_level() < LOG_DEBUG
)
1261 flags
|= PAM_SILENT
;
1263 pam_code
= pam_start(name
, user
, &conv
, &handle
);
1264 if (pam_code
!= PAM_SUCCESS
) {
1270 _cleanup_free_
char *q
= NULL
;
1272 /* Hmm, so no TTY was explicitly passed, but an fd passed to us directly might be a TTY. Let's figure
1273 * out if that's the case, and read the TTY off it. */
1275 if (getttyname_malloc(STDIN_FILENO
, &q
) >= 0)
1276 tty
= strjoina("/dev/", q
);
1280 pam_code
= pam_set_item(handle
, PAM_TTY
, tty
);
1281 if (pam_code
!= PAM_SUCCESS
)
1285 STRV_FOREACH(nv
, *env
) {
1286 pam_code
= pam_putenv(handle
, *nv
);
1287 if (pam_code
!= PAM_SUCCESS
)
1291 pam_code
= pam_acct_mgmt(handle
, flags
);
1292 if (pam_code
!= PAM_SUCCESS
)
1295 pam_code
= pam_setcred(handle
, PAM_ESTABLISH_CRED
| flags
);
1296 if (pam_code
!= PAM_SUCCESS
)
1297 log_debug("pam_setcred() failed, ignoring: %s", pam_strerror(handle
, pam_code
));
1299 pam_code
= pam_open_session(handle
, flags
);
1300 if (pam_code
!= PAM_SUCCESS
)
1303 close_session
= true;
1305 e
= pam_getenvlist(handle
);
1307 pam_code
= PAM_BUF_ERR
;
1311 /* Block SIGTERM, so that we know that it won't get lost in the child */
1313 assert_se(sigprocmask_many(SIG_BLOCK
, &old_ss
, SIGTERM
, -1) >= 0);
1315 parent_pid
= getpid_cached();
1317 r
= safe_fork("(sd-pam)", 0, &pam_pid
);
1321 int sig
, ret
= EXIT_PAM
;
1323 /* The child's job is to reset the PAM session on termination */
1324 barrier_set_role(&barrier
, BARRIER_CHILD
);
1326 /* Make sure we don't keep open the passed fds in this child. We assume that otherwise only
1327 * those fds are open here that have been opened by PAM. */
1328 (void) close_many(fds
, n_fds
);
1330 /* Drop privileges - we don't need any to pam_close_session and this will make
1331 * PR_SET_PDEATHSIG work in most cases. If this fails, ignore the error - but expect sd-pam
1332 * threads to fail to exit normally */
1334 r
= maybe_setgroups(0, NULL
);
1336 log_warning_errno(r
, "Failed to setgroups() in sd-pam: %m");
1337 if (setresgid(gid
, gid
, gid
) < 0)
1338 log_warning_errno(errno
, "Failed to setresgid() in sd-pam: %m");
1339 if (setresuid(uid
, uid
, uid
) < 0)
1340 log_warning_errno(errno
, "Failed to setresuid() in sd-pam: %m");
1342 (void) ignore_signals(SIGPIPE
);
1344 /* Wait until our parent died. This will only work if the above setresuid() succeeds,
1345 * otherwise the kernel will not allow unprivileged parents kill their privileged children
1346 * this way. We rely on the control groups kill logic to do the rest for us. */
1347 if (prctl(PR_SET_PDEATHSIG
, SIGTERM
) < 0)
1350 /* Tell the parent that our setup is done. This is especially important regarding dropping
1351 * privileges. Otherwise, unit setup might race against our setresuid(2) call.
1353 * If the parent aborted, we'll detect this below, hence ignore return failure here. */
1354 (void) barrier_place(&barrier
);
1356 /* Check if our parent process might already have died? */
1357 if (getppid() == parent_pid
) {
1360 assert_se(sigemptyset(&ss
) >= 0);
1361 assert_se(sigaddset(&ss
, SIGTERM
) >= 0);
1364 if (sigwait(&ss
, &sig
) < 0) {
1371 assert(sig
== SIGTERM
);
1376 pam_code
= pam_setcred(handle
, PAM_DELETE_CRED
| flags
);
1377 if (pam_code
!= PAM_SUCCESS
)
1380 /* If our parent died we'll end the session */
1381 if (getppid() != parent_pid
) {
1382 pam_code
= pam_close_session(handle
, flags
);
1383 if (pam_code
!= PAM_SUCCESS
)
1390 /* NB: pam_end() when called in child processes should set PAM_DATA_SILENT to let the module
1391 * know about this. See pam_end(3) */
1392 (void) pam_end(handle
, pam_code
| flags
| PAM_DATA_SILENT
);
1396 barrier_set_role(&barrier
, BARRIER_PARENT
);
1398 /* If the child was forked off successfully it will do all the cleanups, so forget about the handle
1402 /* Unblock SIGTERM again in the parent */
1403 assert_se(sigprocmask(SIG_SETMASK
, &old_ss
, NULL
) >= 0);
1405 /* We close the log explicitly here, since the PAM modules might have opened it, but we don't want
1406 * this fd around. */
1409 /* Synchronously wait for the child to initialize. We don't care for errors as we cannot
1410 * recover. However, warn loudly if it happens. */
1411 if (!barrier_place_and_sync(&barrier
))
1412 log_error("PAM initialization failed");
1414 return strv_free_and_replace(*env
, e
);
1417 if (pam_code
!= PAM_SUCCESS
) {
1418 log_error("PAM failed: %s", pam_strerror(handle
, pam_code
));
1419 r
= -EPERM
; /* PAM errors do not map to errno */
1421 log_error_errno(r
, "PAM failed: %m");
1425 pam_code
= pam_close_session(handle
, flags
);
1427 (void) pam_end(handle
, pam_code
| flags
);
1437 static void rename_process_from_path(const char *path
) {
1438 _cleanup_free_
char *buf
= NULL
;
1443 /* This resulting string must fit in 10 chars (i.e. the length of "/sbin/init") to look pretty in
1446 if (path_extract_filename(path
, &buf
) < 0) {
1447 rename_process("(...)");
1451 size_t l
= strlen(buf
);
1453 /* The end of the process name is usually more interesting, since the first bit might just be
1460 char process_name
[11];
1461 process_name
[0] = '(';
1462 memcpy(process_name
+1, p
, l
);
1463 process_name
[1+l
] = ')';
1464 process_name
[1+l
+1] = 0;
1466 rename_process(process_name
);
1469 static bool context_has_address_families(const ExecContext
*c
) {
1472 return c
->address_families_allow_list
||
1473 !set_isempty(c
->address_families
);
1476 static bool context_has_syscall_filters(const ExecContext
*c
) {
1479 return c
->syscall_allow_list
||
1480 !hashmap_isempty(c
->syscall_filter
);
1483 static bool context_has_syscall_logs(const ExecContext
*c
) {
1486 return c
->syscall_log_allow_list
||
1487 !hashmap_isempty(c
->syscall_log
);
1490 static bool context_has_no_new_privileges(const ExecContext
*c
) {
1493 if (c
->no_new_privileges
)
1496 if (have_effective_cap(CAP_SYS_ADMIN
) > 0) /* if we are privileged, we don't need NNP */
1499 /* We need NNP if we have any form of seccomp and are unprivileged */
1500 return c
->lock_personality
||
1501 c
->memory_deny_write_execute
||
1502 c
->private_devices
||
1504 c
->protect_hostname
||
1505 c
->protect_kernel_tunables
||
1506 c
->protect_kernel_modules
||
1507 c
->protect_kernel_logs
||
1508 context_has_address_families(c
) ||
1509 exec_context_restrict_namespaces_set(c
) ||
1510 c
->restrict_realtime
||
1511 c
->restrict_suid_sgid
||
1512 !set_isempty(c
->syscall_archs
) ||
1513 context_has_syscall_filters(c
) ||
1514 context_has_syscall_logs(c
);
1517 bool exec_context_has_credentials(const ExecContext
*context
) {
1521 return !hashmap_isempty(context
->set_credentials
) ||
1522 !hashmap_isempty(context
->load_credentials
) ||
1523 !set_isempty(context
->import_credentials
);
1528 static bool skip_seccomp_unavailable(const Unit
* u
, const char* msg
) {
1530 if (is_seccomp_available())
1533 log_unit_debug(u
, "SECCOMP features not detected in the kernel, skipping %s", msg
);
1537 static int apply_syscall_filter(const Unit
* u
, const ExecContext
*c
, bool needs_ambient_hack
) {
1538 uint32_t negative_action
, default_action
, action
;
1544 if (!context_has_syscall_filters(c
))
1547 if (skip_seccomp_unavailable(u
, "SystemCallFilter="))
1550 negative_action
= c
->syscall_errno
== SECCOMP_ERROR_NUMBER_KILL
? scmp_act_kill_process() : SCMP_ACT_ERRNO(c
->syscall_errno
);
1552 if (c
->syscall_allow_list
) {
1553 default_action
= negative_action
;
1554 action
= SCMP_ACT_ALLOW
;
1556 default_action
= SCMP_ACT_ALLOW
;
1557 action
= negative_action
;
1560 if (needs_ambient_hack
) {
1561 r
= seccomp_filter_set_add(c
->syscall_filter
, c
->syscall_allow_list
, syscall_filter_sets
+ SYSCALL_FILTER_SET_SETUID
);
1566 return seccomp_load_syscall_filter_set_raw(default_action
, c
->syscall_filter
, action
, false);
1569 static int apply_syscall_log(const Unit
* u
, const ExecContext
*c
) {
1571 uint32_t default_action
, action
;
1577 if (!context_has_syscall_logs(c
))
1581 if (skip_seccomp_unavailable(u
, "SystemCallLog="))
1584 if (c
->syscall_log_allow_list
) {
1585 /* Log nothing but the ones listed */
1586 default_action
= SCMP_ACT_ALLOW
;
1587 action
= SCMP_ACT_LOG
;
1589 /* Log everything but the ones listed */
1590 default_action
= SCMP_ACT_LOG
;
1591 action
= SCMP_ACT_ALLOW
;
1594 return seccomp_load_syscall_filter_set_raw(default_action
, c
->syscall_log
, action
, false);
1596 /* old libseccomp */
1597 log_unit_debug(u
, "SECCOMP feature SCMP_ACT_LOG not available, skipping SystemCallLog=");
1602 static int apply_syscall_archs(const Unit
*u
, const ExecContext
*c
) {
1606 if (set_isempty(c
->syscall_archs
))
1609 if (skip_seccomp_unavailable(u
, "SystemCallArchitectures="))
1612 return seccomp_restrict_archs(c
->syscall_archs
);
1615 static int apply_address_families(const Unit
* u
, const ExecContext
*c
) {
1619 if (!context_has_address_families(c
))
1622 if (skip_seccomp_unavailable(u
, "RestrictAddressFamilies="))
1625 return seccomp_restrict_address_families(c
->address_families
, c
->address_families_allow_list
);
1628 static int apply_memory_deny_write_execute(const Unit
* u
, const ExecContext
*c
) {
1634 if (!c
->memory_deny_write_execute
)
1637 /* use prctl() if kernel supports it (6.3) */
1638 r
= prctl(PR_SET_MDWE
, PR_MDWE_REFUSE_EXEC_GAIN
, 0, 0, 0);
1640 log_unit_debug(u
, "Enabled MemoryDenyWriteExecute= with PR_SET_MDWE");
1643 if (r
< 0 && errno
!= EINVAL
)
1644 return log_unit_debug_errno(u
, errno
, "Failed to enable MemoryDenyWriteExecute= with PR_SET_MDWE: %m");
1645 /* else use seccomp */
1646 log_unit_debug(u
, "Kernel doesn't support PR_SET_MDWE: falling back to seccomp");
1648 if (skip_seccomp_unavailable(u
, "MemoryDenyWriteExecute="))
1651 return seccomp_memory_deny_write_execute();
1654 static int apply_restrict_realtime(const Unit
* u
, const ExecContext
*c
) {
1658 if (!c
->restrict_realtime
)
1661 if (skip_seccomp_unavailable(u
, "RestrictRealtime="))
1664 return seccomp_restrict_realtime();
1667 static int apply_restrict_suid_sgid(const Unit
* u
, const ExecContext
*c
) {
1671 if (!c
->restrict_suid_sgid
)
1674 if (skip_seccomp_unavailable(u
, "RestrictSUIDSGID="))
1677 return seccomp_restrict_suid_sgid();
1680 static int apply_protect_sysctl(const Unit
*u
, const ExecContext
*c
) {
1684 /* Turn off the legacy sysctl() system call. Many distributions turn this off while building the kernel, but
1685 * let's protect even those systems where this is left on in the kernel. */
1687 if (!c
->protect_kernel_tunables
)
1690 if (skip_seccomp_unavailable(u
, "ProtectKernelTunables="))
1693 return seccomp_protect_sysctl();
1696 static int apply_protect_kernel_modules(const Unit
*u
, const ExecContext
*c
) {
1700 /* Turn off module syscalls on ProtectKernelModules=yes */
1702 if (!c
->protect_kernel_modules
)
1705 if (skip_seccomp_unavailable(u
, "ProtectKernelModules="))
1708 return seccomp_load_syscall_filter_set(SCMP_ACT_ALLOW
, syscall_filter_sets
+ SYSCALL_FILTER_SET_MODULE
, SCMP_ACT_ERRNO(EPERM
), false);
1711 static int apply_protect_kernel_logs(const Unit
*u
, const ExecContext
*c
) {
1715 if (!c
->protect_kernel_logs
)
1718 if (skip_seccomp_unavailable(u
, "ProtectKernelLogs="))
1721 return seccomp_protect_syslog();
1724 static int apply_protect_clock(const Unit
*u
, const ExecContext
*c
) {
1728 if (!c
->protect_clock
)
1731 if (skip_seccomp_unavailable(u
, "ProtectClock="))
1734 return seccomp_load_syscall_filter_set(SCMP_ACT_ALLOW
, syscall_filter_sets
+ SYSCALL_FILTER_SET_CLOCK
, SCMP_ACT_ERRNO(EPERM
), false);
1737 static int apply_private_devices(const Unit
*u
, const ExecContext
*c
) {
1741 /* If PrivateDevices= is set, also turn off iopl and all @raw-io syscalls. */
1743 if (!c
->private_devices
)
1746 if (skip_seccomp_unavailable(u
, "PrivateDevices="))
1749 return seccomp_load_syscall_filter_set(SCMP_ACT_ALLOW
, syscall_filter_sets
+ SYSCALL_FILTER_SET_RAW_IO
, SCMP_ACT_ERRNO(EPERM
), false);
1752 static int apply_restrict_namespaces(const Unit
*u
, const ExecContext
*c
) {
1756 if (!exec_context_restrict_namespaces_set(c
))
1759 if (skip_seccomp_unavailable(u
, "RestrictNamespaces="))
1762 return seccomp_restrict_namespaces(c
->restrict_namespaces
);
1765 static int apply_lock_personality(const Unit
* u
, const ExecContext
*c
) {
1766 unsigned long personality
;
1772 if (!c
->lock_personality
)
1775 if (skip_seccomp_unavailable(u
, "LockPersonality="))
1778 personality
= c
->personality
;
1780 /* If personality is not specified, use either PER_LINUX or PER_LINUX32 depending on what is currently set. */
1781 if (personality
== PERSONALITY_INVALID
) {
1783 r
= opinionated_personality(&personality
);
1788 return seccomp_lock_personality(personality
);
1794 static int apply_restrict_filesystems(Unit
*u
, const ExecContext
*c
) {
1798 if (!exec_context_restrict_filesystems_set(c
))
1801 if (!u
->manager
->restrict_fs
) {
1802 /* LSM BPF is unsupported or lsm_bpf_setup failed */
1803 log_unit_debug(u
, "LSM BPF not supported, skipping RestrictFileSystems=");
1807 return lsm_bpf_unit_restrict_filesystems(u
, c
->restrict_filesystems
, c
->restrict_filesystems_allow_list
);
1811 static int apply_protect_hostname(const Unit
*u
, const ExecContext
*c
, int *ret_exit_status
) {
1815 if (!c
->protect_hostname
)
1818 if (ns_type_supported(NAMESPACE_UTS
)) {
1819 if (unshare(CLONE_NEWUTS
) < 0) {
1820 if (!ERRNO_IS_NOT_SUPPORTED(errno
) && !ERRNO_IS_PRIVILEGE(errno
)) {
1821 *ret_exit_status
= EXIT_NAMESPACE
;
1822 return log_unit_error_errno(u
, errno
, "Failed to set up UTS namespacing: %m");
1825 log_unit_warning(u
, "ProtectHostname=yes is configured, but UTS namespace setup is prohibited (container manager?), ignoring namespace setup.");
1828 log_unit_warning(u
, "ProtectHostname=yes is configured, but the kernel does not support UTS namespaces, ignoring namespace setup.");
1833 if (skip_seccomp_unavailable(u
, "ProtectHostname="))
1836 r
= seccomp_protect_hostname();
1838 *ret_exit_status
= EXIT_SECCOMP
;
1839 return log_unit_error_errno(u
, r
, "Failed to apply hostname restrictions: %m");
1846 static void do_idle_pipe_dance(int idle_pipe
[static 4]) {
1849 idle_pipe
[1] = safe_close(idle_pipe
[1]);
1850 idle_pipe
[2] = safe_close(idle_pipe
[2]);
1852 if (idle_pipe
[0] >= 0) {
1855 r
= fd_wait_for_event(idle_pipe
[0], POLLHUP
, IDLE_TIMEOUT_USEC
);
1857 if (idle_pipe
[3] >= 0 && r
== 0 /* timeout */) {
1860 /* Signal systemd that we are bored and want to continue. */
1861 n
= write(idle_pipe
[3], "x", 1);
1863 /* Wait for systemd to react to the signal above. */
1864 (void) fd_wait_for_event(idle_pipe
[0], POLLHUP
, IDLE_TIMEOUT2_USEC
);
1867 idle_pipe
[0] = safe_close(idle_pipe
[0]);
1871 idle_pipe
[3] = safe_close(idle_pipe
[3]);
1874 static const char *exec_directory_env_name_to_string(ExecDirectoryType t
);
1876 static int build_environment(
1878 const ExecContext
*c
,
1879 const ExecParameters
*p
,
1880 const CGroupContext
*cgroup_context
,
1884 const char *username
,
1886 dev_t journal_stream_dev
,
1887 ino_t journal_stream_ino
,
1888 const char *memory_pressure_path
,
1891 _cleanup_strv_free_
char **our_env
= NULL
;
1901 #define N_ENV_VARS 19
1902 our_env
= new0(char*, N_ENV_VARS
+ _EXEC_DIRECTORY_TYPE_MAX
);
1907 _cleanup_free_
char *joined
= NULL
;
1909 if (asprintf(&x
, "LISTEN_PID="PID_FMT
, getpid_cached()) < 0)
1911 our_env
[n_env
++] = x
;
1913 if (asprintf(&x
, "LISTEN_FDS=%zu", n_fds
) < 0)
1915 our_env
[n_env
++] = x
;
1917 joined
= strv_join(fdnames
, ":");
1921 x
= strjoin("LISTEN_FDNAMES=", joined
);
1924 our_env
[n_env
++] = x
;
1927 if ((p
->flags
& EXEC_SET_WATCHDOG
) && p
->watchdog_usec
> 0) {
1928 if (asprintf(&x
, "WATCHDOG_PID="PID_FMT
, getpid_cached()) < 0)
1930 our_env
[n_env
++] = x
;
1932 if (asprintf(&x
, "WATCHDOG_USEC="USEC_FMT
, p
->watchdog_usec
) < 0)
1934 our_env
[n_env
++] = x
;
1937 /* If this is D-Bus, tell the nss-systemd module, since it relies on being able to use blocking
1938 * Varlink calls back to us for look up dynamic users in PID 1. Break the deadlock between D-Bus and
1939 * PID 1 by disabling use of PID1' NSS interface for looking up dynamic users. */
1940 if (p
->flags
& EXEC_NSS_DYNAMIC_BYPASS
) {
1941 x
= strdup("SYSTEMD_NSS_DYNAMIC_BYPASS=1");
1944 our_env
[n_env
++] = x
;
1948 x
= strjoin("HOME=", home
);
1952 path_simplify(x
+ 5);
1953 our_env
[n_env
++] = x
;
1957 x
= strjoin("LOGNAME=", username
);
1960 our_env
[n_env
++] = x
;
1962 x
= strjoin("USER=", username
);
1965 our_env
[n_env
++] = x
;
1969 x
= strjoin("SHELL=", shell
);
1973 path_simplify(x
+ 6);
1974 our_env
[n_env
++] = x
;
1977 if (!sd_id128_is_null(u
->invocation_id
)) {
1978 if (asprintf(&x
, "INVOCATION_ID=" SD_ID128_FORMAT_STR
, SD_ID128_FORMAT_VAL(u
->invocation_id
)) < 0)
1981 our_env
[n_env
++] = x
;
1984 if (exec_context_needs_term(c
)) {
1985 _cleanup_free_
char *cmdline
= NULL
;
1986 const char *tty_path
, *term
= NULL
;
1988 tty_path
= exec_context_tty_path(c
);
1990 /* If we are forked off PID 1 and we are supposed to operate on /dev/console, then let's try
1991 * to inherit the $TERM set for PID 1. This is useful for containers so that the $TERM the
1992 * container manager passes to PID 1 ends up all the way in the console login shown. */
1994 if (path_equal_ptr(tty_path
, "/dev/console") && getppid() == 1)
1995 term
= getenv("TERM");
1996 else if (tty_path
&& in_charset(skip_dev_prefix(tty_path
), ALPHANUMERICAL
)) {
1997 _cleanup_free_
char *key
= NULL
;
1999 key
= strjoin("systemd.tty.term.", skip_dev_prefix(tty_path
));
2003 r
= proc_cmdline_get_key(key
, 0, &cmdline
);
2005 log_debug_errno(r
, "Failed to read %s from kernel cmdline, ignoring: %m", key
);
2011 term
= default_term_for_tty(tty_path
);
2013 x
= strjoin("TERM=", term
);
2016 our_env
[n_env
++] = x
;
2019 if (journal_stream_dev
!= 0 && journal_stream_ino
!= 0) {
2020 if (asprintf(&x
, "JOURNAL_STREAM=" DEV_FMT
":" INO_FMT
, journal_stream_dev
, journal_stream_ino
) < 0)
2023 our_env
[n_env
++] = x
;
2026 if (c
->log_namespace
) {
2027 x
= strjoin("LOG_NAMESPACE=", c
->log_namespace
);
2031 our_env
[n_env
++] = x
;
2034 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
2035 _cleanup_free_
char *joined
= NULL
;
2041 if (c
->directories
[t
].n_items
== 0)
2044 n
= exec_directory_env_name_to_string(t
);
2048 for (size_t i
= 0; i
< c
->directories
[t
].n_items
; i
++) {
2049 _cleanup_free_
char *prefixed
= NULL
;
2051 prefixed
= path_join(p
->prefix
[t
], c
->directories
[t
].items
[i
].path
);
2055 if (!strextend_with_separator(&joined
, ":", prefixed
))
2059 x
= strjoin(n
, "=", joined
);
2063 our_env
[n_env
++] = x
;
2066 if (exec_context_has_credentials(c
) && p
->prefix
[EXEC_DIRECTORY_RUNTIME
]) {
2067 x
= strjoin("CREDENTIALS_DIRECTORY=", p
->prefix
[EXEC_DIRECTORY_RUNTIME
], "/credentials/", u
->id
);
2071 our_env
[n_env
++] = x
;
2074 if (asprintf(&x
, "SYSTEMD_EXEC_PID=" PID_FMT
, getpid_cached()) < 0)
2077 our_env
[n_env
++] = x
;
2079 if (memory_pressure_path
) {
2080 x
= strjoin("MEMORY_PRESSURE_WATCH=", memory_pressure_path
);
2084 our_env
[n_env
++] = x
;
2086 if (cgroup_context
&& !path_equal(memory_pressure_path
, "/dev/null")) {
2087 _cleanup_free_
char *b
= NULL
, *e
= NULL
;
2089 if (asprintf(&b
, "%s " USEC_FMT
" " USEC_FMT
,
2090 MEMORY_PRESSURE_DEFAULT_TYPE
,
2091 cgroup_context
->memory_pressure_threshold_usec
== USEC_INFINITY
? MEMORY_PRESSURE_DEFAULT_THRESHOLD_USEC
:
2092 CLAMP(cgroup_context
->memory_pressure_threshold_usec
, 1U, MEMORY_PRESSURE_DEFAULT_WINDOW_USEC
),
2093 MEMORY_PRESSURE_DEFAULT_WINDOW_USEC
) < 0)
2096 if (base64mem(b
, strlen(b
) + 1, &e
) < 0)
2099 x
= strjoin("MEMORY_PRESSURE_WRITE=", e
);
2103 our_env
[n_env
++] = x
;
2107 assert(n_env
< N_ENV_VARS
+ _EXEC_DIRECTORY_TYPE_MAX
);
2110 *ret
= TAKE_PTR(our_env
);
2115 static int build_pass_environment(const ExecContext
*c
, char ***ret
) {
2116 _cleanup_strv_free_
char **pass_env
= NULL
;
2119 STRV_FOREACH(i
, c
->pass_environment
) {
2120 _cleanup_free_
char *x
= NULL
;
2126 x
= strjoin(*i
, "=", v
);
2130 if (!GREEDY_REALLOC(pass_env
, n_env
+ 2))
2133 pass_env
[n_env
++] = TAKE_PTR(x
);
2134 pass_env
[n_env
] = NULL
;
2137 *ret
= TAKE_PTR(pass_env
);
2142 bool exec_needs_network_namespace(const ExecContext
*context
) {
2145 return context
->private_network
|| context
->network_namespace_path
;
2148 static bool exec_needs_ephemeral(const ExecContext
*context
) {
2149 return (context
->root_image
|| context
->root_directory
) && context
->root_ephemeral
;
2152 static bool exec_needs_ipc_namespace(const ExecContext
*context
) {
2155 return context
->private_ipc
|| context
->ipc_namespace_path
;
2158 bool exec_needs_mount_namespace(
2159 const ExecContext
*context
,
2160 const ExecParameters
*params
,
2161 const ExecRuntime
*runtime
) {
2165 if (context
->root_image
)
2168 if (!strv_isempty(context
->read_write_paths
) ||
2169 !strv_isempty(context
->read_only_paths
) ||
2170 !strv_isempty(context
->inaccessible_paths
) ||
2171 !strv_isempty(context
->exec_paths
) ||
2172 !strv_isempty(context
->no_exec_paths
))
2175 if (context
->n_bind_mounts
> 0)
2178 if (context
->n_temporary_filesystems
> 0)
2181 if (context
->n_mount_images
> 0)
2184 if (context
->n_extension_images
> 0)
2187 if (!strv_isempty(context
->extension_directories
))
2190 if (!IN_SET(context
->mount_propagation_flag
, 0, MS_SHARED
))
2193 if (context
->private_tmp
&& runtime
&& runtime
->shared
&& (runtime
->shared
->tmp_dir
|| runtime
->shared
->var_tmp_dir
))
2196 if (context
->private_devices
||
2197 context
->private_mounts
> 0 ||
2198 (context
->private_mounts
< 0 && exec_needs_network_namespace(context
)) ||
2199 context
->protect_system
!= PROTECT_SYSTEM_NO
||
2200 context
->protect_home
!= PROTECT_HOME_NO
||
2201 context
->protect_kernel_tunables
||
2202 context
->protect_kernel_modules
||
2203 context
->protect_kernel_logs
||
2204 context
->protect_control_groups
||
2205 context
->protect_proc
!= PROTECT_PROC_DEFAULT
||
2206 context
->proc_subset
!= PROC_SUBSET_ALL
||
2207 exec_needs_ipc_namespace(context
))
2210 if (context
->root_directory
) {
2211 if (exec_context_get_effective_mount_apivfs(context
))
2214 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
2215 if (params
&& !params
->prefix
[t
])
2218 if (context
->directories
[t
].n_items
> 0)
2223 if (context
->dynamic_user
&&
2224 (context
->directories
[EXEC_DIRECTORY_STATE
].n_items
> 0 ||
2225 context
->directories
[EXEC_DIRECTORY_CACHE
].n_items
> 0 ||
2226 context
->directories
[EXEC_DIRECTORY_LOGS
].n_items
> 0))
2229 if (context
->log_namespace
)
2235 static int setup_private_users(uid_t ouid
, gid_t ogid
, uid_t uid
, gid_t gid
) {
2236 _cleanup_free_
char *uid_map
= NULL
, *gid_map
= NULL
;
2237 _cleanup_close_pair_
int errno_pipe
[2] = PIPE_EBADF
;
2238 _cleanup_close_
int unshare_ready_fd
= -EBADF
;
2239 _cleanup_(sigkill_waitp
) pid_t pid
= 0;
2244 /* Set up a user namespace and map the original UID/GID (IDs from before any user or group changes, i.e.
2245 * the IDs from the user or system manager(s)) to itself, the selected UID/GID to itself, and everything else to
2246 * nobody. In order to be able to write this mapping we need CAP_SETUID in the original user namespace, which
2247 * we however lack after opening the user namespace. To work around this we fork() a temporary child process,
2248 * which waits for the parent to create the new user namespace while staying in the original namespace. The
2249 * child then writes the UID mapping, under full privileges. The parent waits for the child to finish and
2250 * continues execution normally.
2251 * For unprivileged users (i.e. without capabilities), the root to root mapping is excluded. As such, it
2252 * does not need CAP_SETUID to write the single line mapping to itself. */
2254 /* Can only set up multiple mappings with CAP_SETUID. */
2255 if (have_effective_cap(CAP_SETUID
) > 0 && uid
!= ouid
&& uid_is_valid(uid
))
2256 r
= asprintf(&uid_map
,
2257 UID_FMT
" " UID_FMT
" 1\n" /* Map $OUID → $OUID */
2258 UID_FMT
" " UID_FMT
" 1\n", /* Map $UID → $UID */
2259 ouid
, ouid
, uid
, uid
);
2261 r
= asprintf(&uid_map
,
2262 UID_FMT
" " UID_FMT
" 1\n", /* Map $OUID → $OUID */
2268 /* Can only set up multiple mappings with CAP_SETGID. */
2269 if (have_effective_cap(CAP_SETGID
) > 0 && gid
!= ogid
&& gid_is_valid(gid
))
2270 r
= asprintf(&gid_map
,
2271 GID_FMT
" " GID_FMT
" 1\n" /* Map $OGID → $OGID */
2272 GID_FMT
" " GID_FMT
" 1\n", /* Map $GID → $GID */
2273 ogid
, ogid
, gid
, gid
);
2275 r
= asprintf(&gid_map
,
2276 GID_FMT
" " GID_FMT
" 1\n", /* Map $OGID -> $OGID */
2282 /* Create a communication channel so that the parent can tell the child when it finished creating the user
2284 unshare_ready_fd
= eventfd(0, EFD_CLOEXEC
);
2285 if (unshare_ready_fd
< 0)
2288 /* Create a communication channel so that the child can tell the parent a proper error code in case it
2290 if (pipe2(errno_pipe
, O_CLOEXEC
) < 0)
2293 r
= safe_fork("(sd-userns)", FORK_RESET_SIGNALS
|FORK_DEATHSIG
, &pid
);
2297 _cleanup_close_
int fd
= -EBADF
;
2301 /* Child process, running in the original user namespace. Let's update the parent's UID/GID map from
2302 * here, after the parent opened its own user namespace. */
2305 errno_pipe
[0] = safe_close(errno_pipe
[0]);
2307 /* Wait until the parent unshared the user namespace */
2308 if (read(unshare_ready_fd
, &c
, sizeof(c
)) < 0) {
2313 /* Disable the setgroups() system call in the child user namespace, for good. */
2314 a
= procfs_file_alloca(ppid
, "setgroups");
2315 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
2317 if (errno
!= ENOENT
) {
2322 /* If the file is missing the kernel is too old, let's continue anyway. */
2324 if (write(fd
, "deny\n", 5) < 0) {
2329 fd
= safe_close(fd
);
2332 /* First write the GID map */
2333 a
= procfs_file_alloca(ppid
, "gid_map");
2334 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
2339 if (write(fd
, gid_map
, strlen(gid_map
)) < 0) {
2343 fd
= safe_close(fd
);
2345 /* The write the UID map */
2346 a
= procfs_file_alloca(ppid
, "uid_map");
2347 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
2352 if (write(fd
, uid_map
, strlen(uid_map
)) < 0) {
2357 _exit(EXIT_SUCCESS
);
2360 (void) write(errno_pipe
[1], &r
, sizeof(r
));
2361 _exit(EXIT_FAILURE
);
2364 errno_pipe
[1] = safe_close(errno_pipe
[1]);
2366 if (unshare(CLONE_NEWUSER
) < 0)
2369 /* Let the child know that the namespace is ready now */
2370 if (write(unshare_ready_fd
, &c
, sizeof(c
)) < 0)
2373 /* Try to read an error code from the child */
2374 n
= read(errno_pipe
[0], &r
, sizeof(r
));
2377 if (n
== sizeof(r
)) { /* an error code was sent to us */
2382 if (n
!= 0) /* on success we should have read 0 bytes */
2385 r
= wait_for_terminate_and_check("(sd-userns)", TAKE_PID(pid
), 0);
2388 if (r
!= EXIT_SUCCESS
) /* If something strange happened with the child, let's consider this fatal, too */
2394 static bool exec_directory_is_private(const ExecContext
*context
, ExecDirectoryType type
) {
2397 if (!context
->dynamic_user
)
2400 if (type
== EXEC_DIRECTORY_CONFIGURATION
)
2403 if (type
== EXEC_DIRECTORY_RUNTIME
&& context
->runtime_directory_preserve_mode
== EXEC_PRESERVE_NO
)
2409 static int create_many_symlinks(const char *root
, const char *source
, char **symlinks
) {
2410 _cleanup_free_
char *src_abs
= NULL
;
2415 src_abs
= path_join(root
, source
);
2419 STRV_FOREACH(dst
, symlinks
) {
2420 _cleanup_free_
char *dst_abs
= NULL
;
2422 dst_abs
= path_join(root
, *dst
);
2426 r
= mkdir_parents_label(dst_abs
, 0755);
2430 r
= symlink_idempotent(src_abs
, dst_abs
, true);
2438 static int setup_exec_directory(
2440 const ExecContext
*context
,
2441 const ExecParameters
*params
,
2444 ExecDirectoryType type
,
2445 bool needs_mount_namespace
,
2448 static const int exit_status_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
2449 [EXEC_DIRECTORY_RUNTIME
] = EXIT_RUNTIME_DIRECTORY
,
2450 [EXEC_DIRECTORY_STATE
] = EXIT_STATE_DIRECTORY
,
2451 [EXEC_DIRECTORY_CACHE
] = EXIT_CACHE_DIRECTORY
,
2452 [EXEC_DIRECTORY_LOGS
] = EXIT_LOGS_DIRECTORY
,
2453 [EXEC_DIRECTORY_CONFIGURATION
] = EXIT_CONFIGURATION_DIRECTORY
,
2459 assert(type
>= 0 && type
< _EXEC_DIRECTORY_TYPE_MAX
);
2460 assert(exit_status
);
2462 if (!params
->prefix
[type
])
2465 if (params
->flags
& EXEC_CHOWN_DIRECTORIES
) {
2466 if (!uid_is_valid(uid
))
2468 if (!gid_is_valid(gid
))
2472 for (size_t i
= 0; i
< context
->directories
[type
].n_items
; i
++) {
2473 _cleanup_free_
char *p
= NULL
, *pp
= NULL
;
2475 p
= path_join(params
->prefix
[type
], context
->directories
[type
].items
[i
].path
);
2481 r
= mkdir_parents_label(p
, 0755);
2485 if (IN_SET(type
, EXEC_DIRECTORY_STATE
, EXEC_DIRECTORY_LOGS
) && params
->runtime_scope
== RUNTIME_SCOPE_USER
) {
2487 /* If we are in user mode, and a configuration directory exists but a state directory
2488 * doesn't exist, then we likely are upgrading from an older systemd version that
2489 * didn't know the more recent addition to the xdg-basedir spec: the $XDG_STATE_HOME
2490 * directory. In older systemd versions EXEC_DIRECTORY_STATE was aliased to
2491 * EXEC_DIRECTORY_CONFIGURATION, with the advent of $XDG_STATE_HOME is is now
2492 * separated. If a service has both dirs configured but only the configuration dir
2493 * exists and the state dir does not, we assume we are looking at an update
2494 * situation. Hence, create a compatibility symlink, so that all expectations are
2497 * (We also do something similar with the log directory, which still doesn't exist in
2498 * the xdg basedir spec. We'll make it a subdir of the state dir.) */
2500 /* this assumes the state dir is always created before the configuration dir */
2501 assert_cc(EXEC_DIRECTORY_STATE
< EXEC_DIRECTORY_LOGS
);
2502 assert_cc(EXEC_DIRECTORY_LOGS
< EXEC_DIRECTORY_CONFIGURATION
);
2504 r
= laccess(p
, F_OK
);
2506 _cleanup_free_
char *q
= NULL
;
2508 /* OK, we know that the state dir does not exist. Let's see if the dir exists
2509 * under the configuration hierarchy. */
2511 if (type
== EXEC_DIRECTORY_STATE
)
2512 q
= path_join(params
->prefix
[EXEC_DIRECTORY_CONFIGURATION
], context
->directories
[type
].items
[i
].path
);
2513 else if (type
== EXEC_DIRECTORY_LOGS
)
2514 q
= path_join(params
->prefix
[EXEC_DIRECTORY_CONFIGURATION
], "log", context
->directories
[type
].items
[i
].path
);
2516 assert_not_reached();
2522 r
= laccess(q
, F_OK
);
2524 /* It does exist! This hence looks like an update. Symlink the
2525 * configuration directory into the state directory. */
2527 r
= symlink_idempotent(q
, p
, /* make_relative= */ true);
2531 log_unit_notice(u
, "Unit state directory %s missing but matching configuration directory %s exists, assuming update from systemd 253 or older, creating compatibility symlink.", p
, q
);
2533 } else if (r
!= -ENOENT
)
2534 log_unit_warning_errno(u
, r
, "Unable to detect whether unit configuration directory '%s' exists, assuming not: %m", q
);
2537 log_unit_warning_errno(u
, r
, "Unable to detect whether unit state directory '%s' is missing, assuming it is: %m", p
);
2540 if (exec_directory_is_private(context
, type
)) {
2541 /* So, here's one extra complication when dealing with DynamicUser=1 units. In that
2542 * case we want to avoid leaving a directory around fully accessible that is owned by
2543 * a dynamic user whose UID is later on reused. To lock this down we use the same
2544 * trick used by container managers to prohibit host users to get access to files of
2545 * the same UID in containers: we place everything inside a directory that has an
2546 * access mode of 0700 and is owned root:root, so that it acts as security boundary
2547 * for unprivileged host code. We then use fs namespacing to make this directory
2548 * permeable for the service itself.
2550 * Specifically: for a service which wants a special directory "foo/" we first create
2551 * a directory "private/" with access mode 0700 owned by root:root. Then we place
2552 * "foo" inside of that directory (i.e. "private/foo/"), and make "foo" a symlink to
2553 * "private/foo". This way, privileged host users can access "foo/" as usual, but
2554 * unprivileged host users can't look into it. Inside of the namespace of the unit
2555 * "private/" is replaced by a more liberally accessible tmpfs, into which the host's
2556 * "private/foo/" is mounted under the same name, thus disabling the access boundary
2557 * for the service and making sure it only gets access to the dirs it needs but no
2558 * others. Tricky? Yes, absolutely, but it works!
2560 * Note that we don't do this for EXEC_DIRECTORY_CONFIGURATION as that's assumed not
2561 * to be owned by the service itself.
2563 * Also, note that we don't do this for EXEC_DIRECTORY_RUNTIME as that's often used
2564 * for sharing files or sockets with other services. */
2566 pp
= path_join(params
->prefix
[type
], "private");
2572 /* First set up private root if it doesn't exist yet, with access mode 0700 and owned by root:root */
2573 r
= mkdir_safe_label(pp
, 0700, 0, 0, MKDIR_WARN_MODE
);
2577 if (!path_extend(&pp
, context
->directories
[type
].items
[i
].path
)) {
2582 /* Create all directories between the configured directory and this private root, and mark them 0755 */
2583 r
= mkdir_parents_label(pp
, 0755);
2587 if (is_dir(p
, false) > 0 &&
2588 (laccess(pp
, F_OK
) == -ENOENT
)) {
2590 /* Hmm, the private directory doesn't exist yet, but the normal one exists? If so, move
2591 * it over. Most likely the service has been upgraded from one that didn't use
2592 * DynamicUser=1, to one that does. */
2594 log_unit_info(u
, "Found pre-existing public %s= directory %s, migrating to %s.\n"
2595 "Apparently, service previously had DynamicUser= turned off, and has now turned it on.",
2596 exec_directory_type_to_string(type
), p
, pp
);
2598 r
= RET_NERRNO(rename(p
, pp
));
2602 /* Otherwise, create the actual directory for the service */
2604 r
= mkdir_label(pp
, context
->directories
[type
].mode
);
2605 if (r
< 0 && r
!= -EEXIST
)
2609 if (!context
->directories
[type
].items
[i
].only_create
) {
2610 /* And link it up from the original place.
2612 * 1) If a mount namespace is going to be used, then this symlink remains on
2613 * the host, and a new one for the child namespace will be created later.
2614 * 2) It is not necessary to create this symlink when one of its parent
2615 * directories is specified and already created. E.g.
2616 * StateDirectory=foo foo/bar
2617 * In that case, the inode points to pp and p for "foo/bar" are the same:
2618 * pp = "/var/lib/private/foo/bar"
2619 * p = "/var/lib/foo/bar"
2620 * and, /var/lib/foo is a symlink to /var/lib/private/foo. So, not only
2621 * we do not need to create the symlink, but we cannot create the symlink.
2622 * See issue #24783. */
2623 r
= symlink_idempotent(pp
, p
, true);
2629 _cleanup_free_
char *target
= NULL
;
2631 if (type
!= EXEC_DIRECTORY_CONFIGURATION
&&
2632 readlink_and_make_absolute(p
, &target
) >= 0) {
2633 _cleanup_free_
char *q
= NULL
, *q_resolved
= NULL
, *target_resolved
= NULL
;
2635 /* This already exists and is a symlink? Interesting. Maybe it's one created
2636 * by DynamicUser=1 (see above)?
2638 * We do this for all directory types except for ConfigurationDirectory=,
2639 * since they all support the private/ symlink logic at least in some
2640 * configurations, see above. */
2642 r
= chase(target
, NULL
, 0, &target_resolved
, NULL
);
2646 q
= path_join(params
->prefix
[type
], "private", context
->directories
[type
].items
[i
].path
);
2652 /* /var/lib or friends may be symlinks. So, let's chase them also. */
2653 r
= chase(q
, NULL
, CHASE_NONEXISTENT
, &q_resolved
, NULL
);
2657 if (path_equal(q_resolved
, target_resolved
)) {
2659 /* Hmm, apparently DynamicUser= was once turned on for this service,
2660 * but is no longer. Let's move the directory back up. */
2662 log_unit_info(u
, "Found pre-existing private %s= directory %s, migrating to %s.\n"
2663 "Apparently, service previously had DynamicUser= turned on, and has now turned it off.",
2664 exec_directory_type_to_string(type
), q
, p
);
2666 r
= RET_NERRNO(unlink(p
));
2670 r
= RET_NERRNO(rename(q
, p
));
2676 r
= mkdir_label(p
, context
->directories
[type
].mode
);
2681 if (type
== EXEC_DIRECTORY_CONFIGURATION
) {
2684 /* Don't change the owner/access mode of the configuration directory,
2685 * as in the common case it is not written to by a service, and shall
2686 * not be writable. */
2688 r
= RET_NERRNO(stat(p
, &st
));
2692 /* Still complain if the access mode doesn't match */
2693 if (((st
.st_mode
^ context
->directories
[type
].mode
) & 07777) != 0)
2694 log_unit_warning(u
, "%s \'%s\' already exists but the mode is different. "
2695 "(File system: %o %sMode: %o)",
2696 exec_directory_type_to_string(type
), context
->directories
[type
].items
[i
].path
,
2697 st
.st_mode
& 07777, exec_directory_type_to_string(type
), context
->directories
[type
].mode
& 07777);
2704 /* Lock down the access mode (we use chmod_and_chown() to make this idempotent. We don't
2705 * specify UID/GID here, so that path_chown_recursive() can optimize things depending on the
2706 * current UID/GID ownership.) */
2707 r
= chmod_and_chown(pp
?: p
, context
->directories
[type
].mode
, UID_INVALID
, GID_INVALID
);
2711 /* Skip the rest (which deals with ownership) in user mode, since ownership changes are not
2712 * available to user code anyway */
2713 if (params
->runtime_scope
!= RUNTIME_SCOPE_SYSTEM
)
2716 /* Then, change the ownership of the whole tree, if necessary. When dynamic users are used we
2717 * drop the suid/sgid bits, since we really don't want SUID/SGID files for dynamic UID/GID
2718 * assignments to exist. */
2719 r
= path_chown_recursive(pp
?: p
, uid
, gid
, context
->dynamic_user
? 01777 : 07777, AT_SYMLINK_FOLLOW
);
2724 /* If we are not going to run in a namespace, set up the symlinks - otherwise
2725 * they are set up later, to allow configuring empty var/run/etc. */
2726 if (!needs_mount_namespace
)
2727 for (size_t i
= 0; i
< context
->directories
[type
].n_items
; i
++) {
2728 r
= create_many_symlinks(params
->prefix
[type
],
2729 context
->directories
[type
].items
[i
].path
,
2730 context
->directories
[type
].items
[i
].symlinks
);
2738 *exit_status
= exit_status_table
[type
];
2742 static int write_credential(
2748 bool ownership_ok
) {
2750 _cleanup_(unlink_and_freep
) char *tmp
= NULL
;
2751 _cleanup_close_
int fd
= -EBADF
;
2754 r
= tempfn_random_child("", "cred", &tmp
);
2758 fd
= openat(dfd
, tmp
, O_CREAT
|O_RDWR
|O_CLOEXEC
|O_EXCL
|O_NOFOLLOW
|O_NOCTTY
, 0600);
2764 r
= loop_write(fd
, data
, size
, /* do_poll = */ false);
2768 if (fchmod(fd
, 0400) < 0) /* Take away "w" bit */
2771 if (uid_is_valid(uid
) && uid
!= getuid()) {
2772 r
= fd_add_uid_acl_permission(fd
, uid
, ACL_READ
);
2774 if (!ERRNO_IS_NOT_SUPPORTED(r
) && !ERRNO_IS_PRIVILEGE(r
))
2777 if (!ownership_ok
) /* Ideally we use ACLs, since we can neatly express what we want
2778 * to express: that the user gets read access and nothing
2779 * else. But if the backing fs can't support that (e.g. ramfs)
2780 * then we can use file ownership instead. But that's only safe if
2781 * we can then re-mount the whole thing read-only, so that the
2782 * user can no longer chmod() the file to gain write access. */
2785 if (fchown(fd
, uid
, GID_INVALID
) < 0)
2790 if (renameat(dfd
, tmp
, dfd
, id
) < 0)
2797 typedef enum CredentialSearchPath
{
2798 CREDENTIAL_SEARCH_PATH_TRUSTED
,
2799 CREDENTIAL_SEARCH_PATH_ENCRYPTED
,
2800 CREDENTIAL_SEARCH_PATH_ALL
,
2801 _CREDENTIAL_SEARCH_PATH_MAX
,
2802 _CREDENTIAL_SEARCH_PATH_INVALID
= -EINVAL
,
2803 } CredentialSearchPath
;
2805 static char **credential_search_path(const ExecParameters
*params
, CredentialSearchPath path
) {
2807 _cleanup_strv_free_
char **l
= NULL
;
2810 assert(path
>= 0 && path
< _CREDENTIAL_SEARCH_PATH_MAX
);
2812 /* Assemble a search path to find credentials in. For non-encrypted credentials, We'll look in
2813 * /etc/credstore/ (and similar directories in /usr/lib/ + /run/). If we're looking for encrypted
2814 * credentials, we'll look in /etc/credstore.encrypted/ (and similar dirs). */
2816 if (IN_SET(path
, CREDENTIAL_SEARCH_PATH_ENCRYPTED
, CREDENTIAL_SEARCH_PATH_ALL
)) {
2817 if (strv_extend(&l
, params
->received_encrypted_credentials_directory
) < 0)
2820 if (strv_extend_strv(&l
, CONF_PATHS_STRV("credstore.encrypted"), /* filter_duplicates= */ true) < 0)
2824 if (IN_SET(path
, CREDENTIAL_SEARCH_PATH_TRUSTED
, CREDENTIAL_SEARCH_PATH_ALL
)) {
2825 if (params
->received_credentials_directory
)
2826 if (strv_extend(&l
, params
->received_credentials_directory
) < 0)
2829 if (strv_extend_strv(&l
, CONF_PATHS_STRV("credstore"), /* filter_duplicates= */ true) < 0)
2833 if (DEBUG_LOGGING
) {
2834 _cleanup_free_
char *t
= strv_join(l
, ":");
2836 log_debug("Credential search path is: %s", strempty(t
));
2842 static int maybe_decrypt_and_write_credential(
2852 _cleanup_free_
void *plaintext
= NULL
;
2857 size_t plaintext_size
= 0;
2859 r
= decrypt_credential_and_warn(id
, now(CLOCK_REALTIME
), NULL
, NULL
, data
, size
,
2860 &plaintext
, &plaintext_size
);
2865 size
= plaintext_size
;
2868 add
= strlen(id
) + size
;
2872 r
= write_credential(dir_fd
, id
, data
, size
, uid
, ownership_ok
);
2874 return log_debug_errno(r
, "Failed to write credential '%s': %m", id
);
2880 static int load_credential_glob(
2884 ReadFullFileFlags flags
,
2892 STRV_FOREACH(d
, search_path
) {
2893 _cleanup_globfree_ glob_t pglob
= {};
2894 _cleanup_free_
char *j
= NULL
;
2896 j
= path_join(*d
, path
);
2900 r
= safe_glob(j
, 0, &pglob
);
2906 for (size_t n
= 0; n
< pglob
.gl_pathc
; n
++) {
2907 _cleanup_free_
char *fn
= NULL
;
2908 _cleanup_(erase_and_freep
) char *data
= NULL
;
2911 /* path is absolute, hence pass AT_FDCWD as nop dir fd here */
2912 r
= read_full_file_full(
2916 encrypted
? CREDENTIAL_ENCRYPTED_SIZE_MAX
: CREDENTIAL_SIZE_MAX
,
2921 return log_debug_errno(r
, "Failed to read credential '%s': %m",
2924 r
= path_extract_filename(pglob
.gl_pathv
[n
], &fn
);
2926 return log_debug_errno(r
, "Failed to extract filename from '%s': %m",
2929 r
= maybe_decrypt_and_write_credential(
2947 static int load_credential(
2948 const ExecContext
*context
,
2949 const ExecParameters
*params
,
2960 ReadFullFileFlags flags
= READ_FULL_FILE_SECURE
|READ_FULL_FILE_FAIL_WHEN_LARGER
;
2961 _cleanup_strv_free_
char **search_path
= NULL
;
2962 _cleanup_(erase_and_freep
) char *data
= NULL
;
2963 _cleanup_free_
char *bindname
= NULL
;
2964 const char *source
= NULL
;
2965 bool missing_ok
= true;
2974 assert(read_dfd
>= 0 || read_dfd
== AT_FDCWD
);
2975 assert(write_dfd
>= 0);
2978 if (read_dfd
>= 0) {
2979 /* If a directory fd is specified, then read the file directly from that dir. In this case we
2980 * won't do AF_UNIX stuff (we simply don't want to recursively iterate down a tree of AF_UNIX
2981 * IPC sockets). It's OK if a file vanishes here in the time we enumerate it and intend to
2984 if (!filename_is_valid(path
)) /* safety check */
2990 } else if (path_is_absolute(path
)) {
2991 /* If this is an absolute path, read the data directly from it, and support AF_UNIX
2994 if (!path_is_valid(path
)) /* safety check */
2997 flags
|= READ_FULL_FILE_CONNECT_SOCKET
;
2999 /* Pass some minimal info about the unit and the credential name we are looking to acquire
3000 * via the source socket address in case we read off an AF_UNIX socket. */
3001 if (asprintf(&bindname
, "@%" PRIx64
"/unit/%s/%s", random_u64(), unit
, id
) < 0)
3007 } else if (credential_name_valid(path
)) {
3008 /* If this is a relative path, take it as credential name relative to the credentials
3009 * directory we received ourselves. We don't support the AF_UNIX stuff in this mode, since we
3010 * are operating on a credential store, i.e. this is guaranteed to be regular files. */
3012 search_path
= credential_search_path(params
, CREDENTIAL_SEARCH_PATH_ALL
);
3021 flags
|= READ_FULL_FILE_UNBASE64
;
3023 maxsz
= encrypted
? CREDENTIAL_ENCRYPTED_SIZE_MAX
: CREDENTIAL_SIZE_MAX
;
3026 STRV_FOREACH(d
, search_path
) {
3027 _cleanup_free_
char *j
= NULL
;
3029 j
= path_join(*d
, path
);
3033 r
= read_full_file_full(
3034 AT_FDCWD
, j
, /* path is absolute, hence pass AT_FDCWD as nop dir fd here */
3044 r
= read_full_file_full(
3054 if (r
== -ENOENT
&& (missing_ok
|| hashmap_contains(context
->set_credentials
, id
))) {
3055 /* Make a missing inherited credential non-fatal, let's just continue. After all apps
3056 * will get clear errors if we don't pass such a missing credential on as they
3057 * themselves will get ENOENT when trying to read them, which should not be much
3058 * worse than when we handle the error here and make it fatal.
3060 * Also, if the source file doesn't exist, but a fallback is set via SetCredentials=
3061 * we are fine, too. */
3062 log_debug_errno(r
, "Couldn't read inherited credential '%s', skipping: %m", path
);
3066 return log_debug_errno(r
, "Failed to read credential '%s': %m", path
);
3068 return maybe_decrypt_and_write_credential(write_dfd
, id
, encrypted
, uid
, ownership_ok
, data
, size
, left
);
3071 struct load_cred_args
{
3072 const ExecContext
*context
;
3073 const ExecParameters
*params
;
3082 static int load_cred_recurse_dir_cb(
3083 RecurseDirEvent event
,
3087 const struct dirent
*de
,
3088 const struct statx
*sx
,
3091 struct load_cred_args
*args
= ASSERT_PTR(userdata
);
3092 _cleanup_free_
char *sub_id
= NULL
;
3095 if (event
!= RECURSE_DIR_ENTRY
)
3096 return RECURSE_DIR_CONTINUE
;
3098 if (!IN_SET(de
->d_type
, DT_REG
, DT_SOCK
))
3099 return RECURSE_DIR_CONTINUE
;
3101 sub_id
= strreplace(path
, "/", "_");
3105 if (!credential_name_valid(sub_id
))
3106 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
), "Credential would get ID %s, which is not valid, refusing", sub_id
);
3108 if (faccessat(args
->dfd
, sub_id
, F_OK
, AT_SYMLINK_NOFOLLOW
) >= 0) {
3109 log_debug("Skipping credential with duplicated ID %s at %s", sub_id
, path
);
3110 return RECURSE_DIR_CONTINUE
;
3112 if (errno
!= ENOENT
)
3113 return log_debug_errno(errno
, "Failed to test if credential %s exists: %m", sub_id
);
3115 r
= load_credential(
3130 return RECURSE_DIR_CONTINUE
;
3133 static int acquire_credentials(
3134 const ExecContext
*context
,
3135 const ExecParameters
*params
,
3139 bool ownership_ok
) {
3141 uint64_t left
= CREDENTIALS_TOTAL_SIZE_MAX
;
3142 _cleanup_close_
int dfd
= -EBADF
;
3144 ExecLoadCredential
*lc
;
3145 ExecSetCredential
*sc
;
3151 dfd
= open(p
, O_DIRECTORY
|O_CLOEXEC
);
3155 r
= fd_acl_make_writable(dfd
); /* Add the "w" bit, if we are reusing an already set up credentials dir where it was unset */
3159 /* First, load credentials off disk (or acquire via AF_UNIX socket) */
3160 HASHMAP_FOREACH(lc
, context
->load_credentials
) {
3161 _cleanup_close_
int sub_fd
= -EBADF
;
3163 /* If this is an absolute path, then try to open it as a directory. If that works, then we'll
3164 * recurse into it. If it is an absolute path but it isn't a directory, then we'll open it as
3165 * a regular file. Finally, if it's a relative path we will use it as a credential name to
3166 * propagate a credential passed to us from further up. */
3168 if (path_is_absolute(lc
->path
)) {
3169 sub_fd
= open(lc
->path
, O_DIRECTORY
|O_CLOEXEC
|O_RDONLY
);
3170 if (sub_fd
< 0 && !IN_SET(errno
,
3171 ENOTDIR
, /* Not a directory */
3172 ENOENT
)) /* Doesn't exist? */
3173 return log_debug_errno(errno
, "Failed to open '%s': %m", lc
->path
);
3177 /* Regular file (incl. a credential passed in from higher up) */
3178 r
= load_credential(
3194 /* path= */ lc
->id
, /* recurse_dir() will suffix the subdir paths from here to the top-level id */
3195 /* statx_mask= */ 0,
3196 /* n_depth_max= */ UINT_MAX
,
3197 RECURSE_DIR_SORT
|RECURSE_DIR_IGNORE_DOT
|RECURSE_DIR_ENSURE_TYPE
,
3198 load_cred_recurse_dir_cb
,
3199 &(struct load_cred_args
) {
3202 .encrypted
= lc
->encrypted
,
3206 .ownership_ok
= ownership_ok
,
3213 /* Next, look for system credentials and credentials in the credentials store. Note that these do not
3214 * override any credentials found earlier. */
3215 SET_FOREACH(ic
, context
->import_credentials
) {
3216 _cleanup_free_
char **search_path
= NULL
;
3218 search_path
= credential_search_path(params
, CREDENTIAL_SEARCH_PATH_TRUSTED
);
3222 r
= load_credential_glob(
3224 /* encrypted = */ false,
3226 READ_FULL_FILE_SECURE
|READ_FULL_FILE_FAIL_WHEN_LARGER
,
3234 search_path
= strv_free(search_path
);
3235 search_path
= credential_search_path(params
, CREDENTIAL_SEARCH_PATH_ENCRYPTED
);
3239 r
= load_credential_glob(
3241 /* encrypted = */ true,
3243 READ_FULL_FILE_SECURE
|READ_FULL_FILE_FAIL_WHEN_LARGER
|READ_FULL_FILE_UNBASE64
,
3252 /* Finally, we add in literally specified credentials. If the credentials already exist, we'll not
3253 * add them, so that they can act as a "default" if the same credential is specified multiple times. */
3254 HASHMAP_FOREACH(sc
, context
->set_credentials
) {
3255 _cleanup_(erase_and_freep
) void *plaintext
= NULL
;
3259 /* Note that we check ahead of time here instead of relying on O_EXCL|O_CREAT later to return
3260 * EEXIST if the credential already exists. That's because the TPM2-based decryption is kinda
3261 * slow and involved, hence it's nice to be able to skip that if the credential already
3263 if (faccessat(dfd
, sc
->id
, F_OK
, AT_SYMLINK_NOFOLLOW
) >= 0)
3265 if (errno
!= ENOENT
)
3266 return log_debug_errno(errno
, "Failed to test if credential %s exists: %m", sc
->id
);
3268 if (sc
->encrypted
) {
3269 r
= decrypt_credential_and_warn(sc
->id
, now(CLOCK_REALTIME
), NULL
, NULL
, sc
->data
, sc
->size
, &plaintext
, &size
);
3279 add
= strlen(sc
->id
) + size
;
3283 r
= write_credential(dfd
, sc
->id
, data
, size
, uid
, ownership_ok
);
3290 r
= fd_acl_make_read_only(dfd
); /* Now take away the "w" bit */
3294 /* After we created all keys with the right perms, also make sure the credential store as a whole is
3297 if (uid_is_valid(uid
) && uid
!= getuid()) {
3298 r
= fd_add_uid_acl_permission(dfd
, uid
, ACL_READ
| ACL_EXECUTE
);
3300 if (!ERRNO_IS_NOT_SUPPORTED(r
) && !ERRNO_IS_PRIVILEGE(r
))
3306 if (fchown(dfd
, uid
, GID_INVALID
) < 0)
3314 static int setup_credentials_internal(
3315 const ExecContext
*context
,
3316 const ExecParameters
*params
,
3318 const char *final
, /* This is where the credential store shall eventually end up at */
3319 const char *workspace
, /* This is where we can prepare it before moving it to the final place */
3320 bool reuse_workspace
, /* Whether to reuse any existing workspace mount if it already is a mount */
3321 bool must_mount
, /* Whether to require that we mount something, it's not OK to use the plain directory fall back */
3324 int r
, workspace_mounted
; /* negative if we don't know yet whether we have/can mount something; true
3325 * if we mounted something; false if we definitely can't mount anything */
3333 if (reuse_workspace
) {
3334 r
= path_is_mount_point(workspace
, NULL
, 0);
3338 workspace_mounted
= true; /* If this is already a mount, and we are supposed to reuse it, let's keep this in mind */
3340 workspace_mounted
= -1; /* We need to figure out if we can mount something to the workspace */
3342 workspace_mounted
= -1; /* ditto */
3344 r
= path_is_mount_point(final
, NULL
, 0);
3348 /* If the final place already has something mounted, we use that. If the workspace also has
3349 * something mounted we assume it's actually the same mount (but with MS_RDONLY
3351 final_mounted
= true;
3353 if (workspace_mounted
< 0) {
3354 /* If the final place is mounted, but the workspace isn't, then let's bind mount
3355 * the final version to the workspace, and make it writable, so that we can make
3358 r
= mount_nofollow_verbose(LOG_DEBUG
, final
, workspace
, NULL
, MS_BIND
|MS_REC
, NULL
);
3362 r
= mount_nofollow_verbose(LOG_DEBUG
, NULL
, workspace
, NULL
, MS_BIND
|MS_REMOUNT
|credentials_fs_mount_flags(/* ro= */ false), NULL
);
3366 workspace_mounted
= true;
3369 final_mounted
= false;
3371 if (workspace_mounted
< 0) {
3372 /* Nothing is mounted on the workspace yet, let's try to mount something now */
3374 r
= mount_credentials_fs(workspace
, CREDENTIALS_TOTAL_SIZE_MAX
, /* ro= */ false);
3376 /* 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. */
3377 r
= mount_nofollow_verbose(LOG_DEBUG
, final
, workspace
, NULL
, MS_BIND
|MS_REC
, NULL
);
3379 if (!ERRNO_IS_PRIVILEGE(r
)) /* Propagate anything that isn't a permission problem */
3382 if (must_mount
) /* If we it's not OK to use the plain directory
3383 * fallback, propagate all errors too */
3386 /* If we lack privileges to bind mount stuff, then let's gracefully
3387 * proceed for compat with container envs, and just use the final dir
3390 workspace_mounted
= false;
3392 /* Make the new bind mount writable (i.e. drop MS_RDONLY) */
3393 r
= mount_nofollow_verbose(LOG_DEBUG
, NULL
, workspace
, NULL
, MS_BIND
|MS_REMOUNT
|credentials_fs_mount_flags(/* ro= */ false), NULL
);
3397 workspace_mounted
= true;
3400 workspace_mounted
= true;
3403 assert(!must_mount
|| workspace_mounted
> 0);
3404 where
= workspace_mounted
? workspace
: final
;
3406 (void) label_fix_full(AT_FDCWD
, where
, final
, 0);
3408 r
= acquire_credentials(context
, params
, unit
, where
, uid
, workspace_mounted
);
3412 if (workspace_mounted
) {
3415 /* Determine if we should actually install the prepared mount in the final location by bind
3416 * mounting it there. We do so only if the mount is not established there already, and if the
3417 * mount is actually non-empty (i.e. carries at least one credential). Not that in the best
3418 * case we are doing all this in a mount namespace, thus no one else will see that we
3419 * allocated a file system we are getting rid of again here. */
3421 install
= false; /* already installed */
3423 r
= dir_is_empty(where
, /* ignore_hidden_or_backup= */ false);
3427 install
= r
== 0; /* install only if non-empty */
3431 /* Make workspace read-only now, so that any bind mount we make from it defaults to read-only too */
3432 r
= mount_nofollow_verbose(LOG_DEBUG
, NULL
, workspace
, NULL
, MS_BIND
|MS_REMOUNT
|credentials_fs_mount_flags(/* ro= */ true), NULL
);
3436 /* And mount it to the final place, read-only */
3437 r
= mount_nofollow_verbose(LOG_DEBUG
, workspace
, final
, NULL
, MS_MOVE
, NULL
);
3439 /* Otherwise get rid of it */
3440 r
= umount_verbose(LOG_DEBUG
, workspace
, MNT_DETACH
|UMOUNT_NOFOLLOW
);
3444 _cleanup_free_
char *parent
= NULL
;
3446 /* If we do not have our own mount put used the plain directory fallback, then we need to
3447 * open access to the top-level credential directory and the per-service directory now */
3449 r
= path_extract_directory(final
, &parent
);
3452 if (chmod(parent
, 0755) < 0)
3459 static int setup_credentials(
3460 const ExecContext
*context
,
3461 const ExecParameters
*params
,
3465 _cleanup_free_
char *p
= NULL
, *q
= NULL
;
3471 if (!exec_context_has_credentials(context
))
3474 if (!params
->prefix
[EXEC_DIRECTORY_RUNTIME
])
3477 /* This where we'll place stuff when we are done; this main credentials directory is world-readable,
3478 * and the subdir we mount over with a read-only file system readable by the service's user */
3479 q
= path_join(params
->prefix
[EXEC_DIRECTORY_RUNTIME
], "credentials");
3483 r
= mkdir_label(q
, 0755); /* top-level dir: world readable/searchable */
3484 if (r
< 0 && r
!= -EEXIST
)
3487 p
= path_join(q
, unit
);
3491 r
= mkdir_label(p
, 0700); /* per-unit dir: private to user */
3492 if (r
< 0 && r
!= -EEXIST
)
3495 r
= safe_fork("(sd-mkdcreds)", FORK_DEATHSIG
|FORK_WAIT
|FORK_NEW_MOUNTNS
, NULL
);
3497 _cleanup_free_
char *t
= NULL
, *u
= NULL
;
3499 /* If this is not a privilege or support issue then propagate the error */
3500 if (!ERRNO_IS_NOT_SUPPORTED(r
) && !ERRNO_IS_PRIVILEGE(r
))
3503 /* Temporary workspace, that remains inaccessible all the time. We prepare stuff there before moving
3504 * it into place, so that users can't access half-initialized credential stores. */
3505 t
= path_join(params
->prefix
[EXEC_DIRECTORY_RUNTIME
], "systemd/temporary-credentials");
3509 /* We can't set up a mount namespace. In that case operate on a fixed, inaccessible per-unit
3510 * directory outside of /run/credentials/ first, and then move it over to /run/credentials/
3511 * after it is fully set up */
3512 u
= path_join(t
, unit
);
3516 FOREACH_STRING(i
, t
, u
) {
3517 r
= mkdir_label(i
, 0700);
3518 if (r
< 0 && r
!= -EEXIST
)
3522 r
= setup_credentials_internal(
3526 p
, /* final mount point */
3527 u
, /* temporary workspace to overmount */
3528 true, /* reuse the workspace if it is already a mount */
3529 false, /* it's OK to fall back to a plain directory if we can't mount anything */
3532 (void) rmdir(u
); /* remove the workspace again if we can. */
3537 } else if (r
== 0) {
3539 /* We managed to set up a mount namespace, and are now in a child. That's great. In this case
3540 * we can use the same directory for all cases, after turning off propagation. Question
3541 * though is: where do we turn off propagation exactly, and where do we place the workspace
3542 * directory? We need some place that is guaranteed to be a mount point in the host, and
3543 * which is guaranteed to have a subdir we can mount over. /run/ is not suitable for this,
3544 * since we ultimately want to move the resulting file system there, i.e. we need propagation
3545 * for /run/ eventually. We could use our own /run/systemd/bind mount on itself, but that
3546 * would be visible in the host mount table all the time, which we want to avoid. Hence, what
3547 * we do here instead we use /dev/ and /dev/shm/ for our purposes. We know for sure that
3548 * /dev/ is a mount point and we now for sure that /dev/shm/ exists. Hence we can turn off
3549 * propagation on the former, and then overmount the latter.
3551 * Yes it's nasty playing games with /dev/ and /dev/shm/ like this, since it does not exist
3552 * for this purpose, but there are few other candidates that work equally well for us, and
3553 * given that the we do this in a privately namespaced short-lived single-threaded process
3554 * that no one else sees this should be OK to do. */
3556 r
= mount_nofollow_verbose(LOG_DEBUG
, NULL
, "/dev", NULL
, MS_SLAVE
|MS_REC
, NULL
); /* Turn off propagation from our namespace to host */
3560 r
= setup_credentials_internal(
3564 p
, /* final mount point */
3565 "/dev/shm", /* temporary workspace to overmount */
3566 false, /* do not reuse /dev/shm if it is already a mount, under no circumstances */
3567 true, /* insist that something is mounted, do not allow fallback to plain directory */
3572 _exit(EXIT_SUCCESS
);
3575 _exit(EXIT_FAILURE
);
3578 /* If the credentials dir is empty and not a mount point, then there's no point in having it. Let's
3579 * try to remove it. This matters in particular if we created the dir as mount point but then didn't
3580 * actually end up mounting anything on it. In that case we'd rather have ENOENT than EACCESS being
3581 * seen by users when trying access this inode. */
3587 static int setup_smack(
3588 const Manager
*manager
,
3589 const ExecContext
*context
,
3590 int executable_fd
) {
3594 assert(executable_fd
>= 0);
3596 if (context
->smack_process_label
) {
3597 r
= mac_smack_apply_pid(0, context
->smack_process_label
);
3600 } else if (manager
->default_smack_process_label
) {
3601 _cleanup_free_
char *exec_label
= NULL
;
3603 r
= mac_smack_read_fd(executable_fd
, SMACK_ATTR_EXEC
, &exec_label
);
3604 if (r
< 0 && !ERRNO_IS_XATTR_ABSENT(r
))
3607 r
= mac_smack_apply_pid(0, exec_label
?: manager
->default_smack_process_label
);
3616 static int compile_bind_mounts(
3617 const ExecContext
*context
,
3618 const ExecParameters
*params
,
3619 BindMount
**ret_bind_mounts
,
3620 size_t *ret_n_bind_mounts
,
3621 char ***ret_empty_directories
) {
3623 _cleanup_strv_free_
char **empty_directories
= NULL
;
3624 BindMount
*bind_mounts
= NULL
;
3630 assert(ret_bind_mounts
);
3631 assert(ret_n_bind_mounts
);
3632 assert(ret_empty_directories
);
3634 CLEANUP_ARRAY(bind_mounts
, h
, bind_mount_free_many
);
3636 n
= context
->n_bind_mounts
;
3637 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
3638 if (!params
->prefix
[t
])
3641 for (size_t i
= 0; i
< context
->directories
[t
].n_items
; i
++)
3642 n
+= !context
->directories
[t
].items
[i
].only_create
;
3646 *ret_bind_mounts
= NULL
;
3647 *ret_n_bind_mounts
= 0;
3648 *ret_empty_directories
= NULL
;
3652 bind_mounts
= new(BindMount
, n
);
3656 for (size_t i
= 0; i
< context
->n_bind_mounts
; i
++) {
3657 BindMount
*item
= context
->bind_mounts
+ i
;
3658 _cleanup_free_
char *s
= NULL
, *d
= NULL
;
3660 s
= strdup(item
->source
);
3664 d
= strdup(item
->destination
);
3668 bind_mounts
[h
++] = (BindMount
) {
3669 .source
= TAKE_PTR(s
),
3670 .destination
= TAKE_PTR(d
),
3671 .read_only
= item
->read_only
,
3672 .recursive
= item
->recursive
,
3673 .ignore_enoent
= item
->ignore_enoent
,
3677 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
3678 if (!params
->prefix
[t
])
3681 if (context
->directories
[t
].n_items
== 0)
3684 if (exec_directory_is_private(context
, t
) &&
3685 !exec_context_with_rootfs(context
)) {
3688 /* So this is for a dynamic user, and we need to make sure the process can access its own
3689 * directory. For that we overmount the usually inaccessible "private" subdirectory with a
3690 * tmpfs that makes it accessible and is empty except for the submounts we do this for. */
3692 private_root
= path_join(params
->prefix
[t
], "private");
3696 r
= strv_consume(&empty_directories
, private_root
);
3701 for (size_t i
= 0; i
< context
->directories
[t
].n_items
; i
++) {
3702 _cleanup_free_
char *s
= NULL
, *d
= NULL
;
3704 /* When one of the parent directories is in the list, we cannot create the symlink
3705 * for the child directory. See also the comments in setup_exec_directory(). */
3706 if (context
->directories
[t
].items
[i
].only_create
)
3709 if (exec_directory_is_private(context
, t
))
3710 s
= path_join(params
->prefix
[t
], "private", context
->directories
[t
].items
[i
].path
);
3712 s
= path_join(params
->prefix
[t
], context
->directories
[t
].items
[i
].path
);
3716 if (exec_directory_is_private(context
, t
) &&
3717 exec_context_with_rootfs(context
))
3718 /* When RootDirectory= or RootImage= are set, then the symbolic link to the private
3719 * directory is not created on the root directory. So, let's bind-mount the directory
3720 * on the 'non-private' place. */
3721 d
= path_join(params
->prefix
[t
], context
->directories
[t
].items
[i
].path
);
3727 bind_mounts
[h
++] = (BindMount
) {
3728 .source
= TAKE_PTR(s
),
3729 .destination
= TAKE_PTR(d
),
3731 .nosuid
= context
->dynamic_user
, /* don't allow suid/sgid when DynamicUser= is on */
3733 .ignore_enoent
= false,
3740 *ret_bind_mounts
= TAKE_PTR(bind_mounts
);
3741 *ret_n_bind_mounts
= n
;
3742 *ret_empty_directories
= TAKE_PTR(empty_directories
);
3747 /* ret_symlinks will contain a list of pairs src:dest that describes
3748 * the symlinks to create later on. For example, the symlinks needed
3749 * to safely give private directories to DynamicUser=1 users. */
3750 static int compile_symlinks(
3751 const ExecContext
*context
,
3752 const ExecParameters
*params
,
3753 char ***ret_symlinks
) {
3755 _cleanup_strv_free_
char **symlinks
= NULL
;
3760 assert(ret_symlinks
);
3762 for (ExecDirectoryType dt
= 0; dt
< _EXEC_DIRECTORY_TYPE_MAX
; dt
++) {
3763 for (size_t i
= 0; i
< context
->directories
[dt
].n_items
; i
++) {
3764 _cleanup_free_
char *private_path
= NULL
, *path
= NULL
;
3766 STRV_FOREACH(symlink
, context
->directories
[dt
].items
[i
].symlinks
) {
3767 _cleanup_free_
char *src_abs
= NULL
, *dst_abs
= NULL
;
3769 src_abs
= path_join(params
->prefix
[dt
], context
->directories
[dt
].items
[i
].path
);
3770 dst_abs
= path_join(params
->prefix
[dt
], *symlink
);
3771 if (!src_abs
|| !dst_abs
)
3774 r
= strv_consume_pair(&symlinks
, TAKE_PTR(src_abs
), TAKE_PTR(dst_abs
));
3779 if (!exec_directory_is_private(context
, dt
) ||
3780 exec_context_with_rootfs(context
) ||
3781 context
->directories
[dt
].items
[i
].only_create
)
3784 private_path
= path_join(params
->prefix
[dt
], "private", context
->directories
[dt
].items
[i
].path
);
3788 path
= path_join(params
->prefix
[dt
], context
->directories
[dt
].items
[i
].path
);
3792 r
= strv_consume_pair(&symlinks
, TAKE_PTR(private_path
), TAKE_PTR(path
));
3798 *ret_symlinks
= TAKE_PTR(symlinks
);
3803 static bool insist_on_sandboxing(
3804 const ExecContext
*context
,
3805 const char *root_dir
,
3806 const char *root_image
,
3807 const BindMount
*bind_mounts
,
3808 size_t n_bind_mounts
) {
3811 assert(n_bind_mounts
== 0 || bind_mounts
);
3813 /* Checks whether we need to insist on fs namespacing. i.e. whether we have settings configured that
3814 * would alter the view on the file system beyond making things read-only or invisible, i.e. would
3815 * rearrange stuff in a way we cannot ignore gracefully. */
3817 if (context
->n_temporary_filesystems
> 0)
3820 if (root_dir
|| root_image
)
3823 if (context
->n_mount_images
> 0)
3826 if (context
->dynamic_user
)
3829 if (context
->n_extension_images
> 0 || !strv_isempty(context
->extension_directories
))
3832 /* If there are any bind mounts set that don't map back onto themselves, fs namespacing becomes
3834 for (size_t i
= 0; i
< n_bind_mounts
; i
++)
3835 if (!path_equal(bind_mounts
[i
].source
, bind_mounts
[i
].destination
))
3838 if (context
->log_namespace
)
3844 static int setup_ephemeral(const ExecContext
*context
, ExecRuntime
*runtime
) {
3845 _cleanup_close_
int fd
= -EBADF
;
3848 if (!runtime
|| !runtime
->ephemeral_copy
)
3851 r
= posix_lock(runtime
->ephemeral_storage_socket
[0], LOCK_EX
);
3853 return log_debug_errno(r
, "Failed to lock ephemeral storage socket: %m");
3855 CLEANUP_POSIX_UNLOCK(runtime
->ephemeral_storage_socket
[0]);
3857 fd
= receive_one_fd(runtime
->ephemeral_storage_socket
[0], MSG_PEEK
|MSG_DONTWAIT
);
3859 /* We got an fd! That means ephemeral has already been set up, so nothing to do here. */
3863 return log_debug_errno(fd
, "Failed to receive file descriptor queued on ephemeral storage socket: %m");
3865 log_debug("Making ephemeral snapshot of %s to %s",
3866 context
->root_image
?: context
->root_directory
, runtime
->ephemeral_copy
);
3868 if (context
->root_image
)
3869 fd
= copy_file(context
->root_image
, runtime
->ephemeral_copy
, O_EXCL
, 0600,
3870 COPY_LOCK_BSD
|COPY_REFLINK
|COPY_CRTIME
);
3872 fd
= btrfs_subvol_snapshot_at(AT_FDCWD
, context
->root_directory
,
3873 AT_FDCWD
, runtime
->ephemeral_copy
,
3874 BTRFS_SNAPSHOT_FALLBACK_COPY
|
3875 BTRFS_SNAPSHOT_FALLBACK_DIRECTORY
|
3876 BTRFS_SNAPSHOT_RECURSIVE
|
3877 BTRFS_SNAPSHOT_LOCK_BSD
);
3879 return log_debug_errno(fd
, "Failed to snapshot %s to %s: %m",
3880 context
->root_image
?: context
->root_directory
, runtime
->ephemeral_copy
);
3882 if (context
->root_image
) {
3883 /* A root image might be subject to lots of random writes so let's try to disable COW on it
3884 * which tends to not perform well in combination with lots of random writes.
3886 * Note: btrfs actually isn't impressed by us setting the flag after making the reflink'ed
3887 * copy, but we at least want to make the intention clear.
3889 r
= chattr_fd(fd
, FS_NOCOW_FL
, FS_NOCOW_FL
, NULL
);
3891 log_debug_errno(fd
, "Failed to disable copy-on-write for %s, ignoring: %m", runtime
->ephemeral_copy
);
3894 r
= send_one_fd(runtime
->ephemeral_storage_socket
[1], fd
, MSG_DONTWAIT
);
3896 return log_debug_errno(r
, "Failed to queue file descriptor on ephemeral storage socket: %m");
3901 static int verity_settings_prepare(
3902 VeritySettings
*verity
,
3903 const char *root_image
,
3904 const void *root_hash
,
3905 size_t root_hash_size
,
3906 const char *root_hash_path
,
3907 const void *root_hash_sig
,
3908 size_t root_hash_sig_size
,
3909 const char *root_hash_sig_path
,
3910 const char *verity_data_path
) {
3919 d
= memdup(root_hash
, root_hash_size
);
3923 free_and_replace(verity
->root_hash
, d
);
3924 verity
->root_hash_size
= root_hash_size
;
3925 verity
->designator
= PARTITION_ROOT
;
3928 if (root_hash_sig
) {
3931 d
= memdup(root_hash_sig
, root_hash_sig_size
);
3935 free_and_replace(verity
->root_hash_sig
, d
);
3936 verity
->root_hash_sig_size
= root_hash_sig_size
;
3937 verity
->designator
= PARTITION_ROOT
;
3940 if (verity_data_path
) {
3941 r
= free_and_strdup(&verity
->data_path
, verity_data_path
);
3946 r
= verity_settings_load(
3950 root_hash_sig_path
);
3952 return log_debug_errno(r
, "Failed to load root hash: %m");
3957 static int apply_mount_namespace(
3959 ExecCommandFlags command_flags
,
3960 const ExecContext
*context
,
3961 const ExecParameters
*params
,
3962 ExecRuntime
*runtime
,
3963 const char *memory_pressure_path
,
3964 char **error_path
) {
3966 _cleanup_(verity_settings_done
) VeritySettings verity
= VERITY_SETTINGS_DEFAULT
;
3967 _cleanup_strv_free_
char **empty_directories
= NULL
, **symlinks
= NULL
,
3968 **read_write_paths_cleanup
= NULL
;
3969 _cleanup_free_
char *creds_path
= NULL
, *incoming_dir
= NULL
, *propagate_dir
= NULL
,
3970 *extension_dir
= NULL
, *host_os_release
= NULL
;
3971 const char *root_dir
= NULL
, *root_image
= NULL
, *tmp_dir
= NULL
, *var_tmp_dir
= NULL
;
3972 char **read_write_paths
;
3973 NamespaceInfo ns_info
;
3974 bool needs_sandboxing
;
3975 BindMount
*bind_mounts
= NULL
;
3976 size_t n_bind_mounts
= 0;
3981 CLEANUP_ARRAY(bind_mounts
, n_bind_mounts
, bind_mount_free_many
);
3983 if (params
->flags
& EXEC_APPLY_CHROOT
) {
3984 r
= setup_ephemeral(context
, runtime
);
3988 if (context
->root_image
)
3989 root_image
= (runtime
? runtime
->ephemeral_copy
: NULL
) ?: context
->root_image
;
3991 root_dir
= (runtime
? runtime
->ephemeral_copy
: NULL
) ?: context
->root_directory
;
3994 r
= compile_bind_mounts(context
, params
, &bind_mounts
, &n_bind_mounts
, &empty_directories
);
3998 /* Symlinks for exec dirs are set up after other mounts, before they are made read-only. */
3999 r
= compile_symlinks(context
, params
, &symlinks
);
4003 /* We need to make the pressure path writable even if /sys/fs/cgroups is made read-only, as the
4004 * service will need to write to it in order to start the notifications. */
4005 if (context
->protect_control_groups
&& memory_pressure_path
&& !streq(memory_pressure_path
, "/dev/null")) {
4006 read_write_paths_cleanup
= strv_copy(context
->read_write_paths
);
4007 if (!read_write_paths_cleanup
)
4010 r
= strv_extend(&read_write_paths_cleanup
, memory_pressure_path
);
4014 read_write_paths
= read_write_paths_cleanup
;
4016 read_write_paths
= context
->read_write_paths
;
4018 needs_sandboxing
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && !(command_flags
& EXEC_COMMAND_FULLY_PRIVILEGED
);
4019 if (needs_sandboxing
) {
4020 /* The runtime struct only contains the parent of the private /tmp,
4021 * which is non-accessible to world users. Inside of it there's a /tmp
4022 * that is sticky, and that's the one we want to use here.
4023 * This does not apply when we are using /run/systemd/empty as fallback. */
4025 if (context
->private_tmp
&& runtime
&& runtime
->shared
) {
4026 if (streq_ptr(runtime
->shared
->tmp_dir
, RUN_SYSTEMD_EMPTY
))
4027 tmp_dir
= runtime
->shared
->tmp_dir
;
4028 else if (runtime
->shared
->tmp_dir
)
4029 tmp_dir
= strjoina(runtime
->shared
->tmp_dir
, "/tmp");
4031 if (streq_ptr(runtime
->shared
->var_tmp_dir
, RUN_SYSTEMD_EMPTY
))
4032 var_tmp_dir
= runtime
->shared
->var_tmp_dir
;
4033 else if (runtime
->shared
->var_tmp_dir
)
4034 var_tmp_dir
= strjoina(runtime
->shared
->var_tmp_dir
, "/tmp");
4037 ns_info
= (NamespaceInfo
) {
4038 .ignore_protect_paths
= false,
4039 .private_dev
= context
->private_devices
,
4040 .protect_control_groups
= context
->protect_control_groups
,
4041 .protect_kernel_tunables
= context
->protect_kernel_tunables
,
4042 .protect_kernel_modules
= context
->protect_kernel_modules
,
4043 .protect_kernel_logs
= context
->protect_kernel_logs
,
4044 .protect_hostname
= context
->protect_hostname
,
4045 .mount_apivfs
= exec_context_get_effective_mount_apivfs(context
),
4046 .protect_home
= context
->protect_home
,
4047 .protect_system
= context
->protect_system
,
4048 .protect_proc
= context
->protect_proc
,
4049 .proc_subset
= context
->proc_subset
,
4050 .private_network
= exec_needs_network_namespace(context
),
4051 .private_ipc
= exec_needs_ipc_namespace(context
),
4052 /* If NNP is on, we can turn on MS_NOSUID, since it won't have any effect anymore. */
4053 .mount_nosuid
= context
->no_new_privileges
&& !mac_selinux_use(),
4055 } else if (!context
->dynamic_user
&& root_dir
)
4057 * If DynamicUser=no and RootDirectory= is set then lets pass a relaxed
4058 * sandbox info, otherwise enforce it, don't ignore protected paths and
4059 * fail if we are enable to apply the sandbox inside the mount namespace.
4061 ns_info
= (NamespaceInfo
) {
4062 .ignore_protect_paths
= true,
4065 ns_info
= (NamespaceInfo
) {};
4067 if (context
->mount_propagation_flag
== MS_SHARED
)
4068 log_unit_debug(u
, "shared mount propagation hidden by other fs namespacing unit settings: ignoring");
4070 if (exec_context_has_credentials(context
) &&
4071 params
->prefix
[EXEC_DIRECTORY_RUNTIME
] &&
4072 FLAGS_SET(params
->flags
, EXEC_WRITE_CREDENTIALS
)) {
4073 creds_path
= path_join(params
->prefix
[EXEC_DIRECTORY_RUNTIME
], "credentials", u
->id
);
4078 if (params
->runtime_scope
== RUNTIME_SCOPE_SYSTEM
) {
4079 propagate_dir
= path_join("/run/systemd/propagate/", u
->id
);
4083 incoming_dir
= strdup("/run/systemd/incoming");
4087 extension_dir
= strdup("/run/systemd/unit-extensions");
4091 /* If running under a different root filesystem, propagate the host's os-release. We make a
4092 * copy rather than just bind mounting it, so that it can be updated on soft-reboot. */
4093 if (root_dir
|| root_image
) {
4094 host_os_release
= strdup("/run/systemd/propagate/os-release");
4095 if (!host_os_release
)
4099 assert(params
->runtime_scope
== RUNTIME_SCOPE_USER
);
4101 if (asprintf(&extension_dir
, "/run/user/" UID_FMT
"/systemd/unit-extensions", geteuid()) < 0)
4104 if (root_dir
|| root_image
) {
4105 if (asprintf(&host_os_release
, "/run/user/" UID_FMT
"/systemd/propagate/os-release", geteuid()) < 0)
4111 r
= verity_settings_prepare(
4114 context
->root_hash
, context
->root_hash_size
, context
->root_hash_path
,
4115 context
->root_hash_sig
, context
->root_hash_sig_size
, context
->root_hash_sig_path
,
4116 context
->root_verity
);
4121 r
= setup_namespace(
4124 context
->root_image_options
,
4125 context
->root_image_policy
?: &image_policy_service
,
4128 needs_sandboxing
? context
->read_only_paths
: NULL
,
4129 needs_sandboxing
? context
->inaccessible_paths
: NULL
,
4130 needs_sandboxing
? context
->exec_paths
: NULL
,
4131 needs_sandboxing
? context
->no_exec_paths
: NULL
,
4136 context
->temporary_filesystems
,
4137 context
->n_temporary_filesystems
,
4138 context
->mount_images
,
4139 context
->n_mount_images
,
4140 context
->mount_image_policy
?: &image_policy_service
,
4144 context
->log_namespace
,
4145 context
->mount_propagation_flag
,
4147 context
->extension_images
,
4148 context
->n_extension_images
,
4149 context
->extension_image_policy
?: &image_policy_sysext
,
4150 context
->extension_directories
,
4154 root_dir
|| root_image
? params
->notify_socket
: NULL
,
4158 /* If we couldn't set up the namespace this is probably due to a missing capability. setup_namespace() reports
4159 * that with a special, recognizable error ENOANO. In this case, silently proceed, but only if exclusively
4160 * sandboxing options were used, i.e. nothing such as RootDirectory= or BindMount= that would result in a
4161 * completely different execution environment. */
4163 if (insist_on_sandboxing(
4165 root_dir
, root_image
,
4168 return log_unit_debug_errno(u
,
4169 SYNTHETIC_ERRNO(EOPNOTSUPP
),
4170 "Failed to set up namespace, and refusing to continue since "
4171 "the selected namespacing options alter mount environment non-trivially.\n"
4172 "Bind mounts: %zu, temporary filesystems: %zu, root directory: %s, root image: %s, dynamic user: %s",
4174 context
->n_temporary_filesystems
,
4177 yes_no(context
->dynamic_user
));
4179 log_unit_debug(u
, "Failed to set up namespace, assuming containerized execution and ignoring.");
4186 static int apply_working_directory(
4187 const ExecContext
*context
,
4188 const ExecParameters
*params
,
4189 ExecRuntime
*runtime
,
4196 assert(exit_status
);
4198 if (context
->working_directory_home
) {
4201 *exit_status
= EXIT_CHDIR
;
4208 wd
= empty_to_root(context
->working_directory
);
4210 if (params
->flags
& EXEC_APPLY_CHROOT
)
4213 d
= prefix_roota((runtime
? runtime
->ephemeral_copy
: NULL
) ?: context
->root_directory
, wd
);
4215 if (chdir(d
) < 0 && !context
->working_directory_missing_ok
) {
4216 *exit_status
= EXIT_CHDIR
;
4223 static int apply_root_directory(
4224 const ExecContext
*context
,
4225 const ExecParameters
*params
,
4226 ExecRuntime
*runtime
,
4227 const bool needs_mount_ns
,
4231 assert(exit_status
);
4233 if (params
->flags
& EXEC_APPLY_CHROOT
)
4234 if (!needs_mount_ns
&& context
->root_directory
)
4235 if (chroot((runtime
? runtime
->ephemeral_copy
: NULL
) ?: context
->root_directory
) < 0) {
4236 *exit_status
= EXIT_CHROOT
;
4243 static int setup_keyring(
4245 const ExecContext
*context
,
4246 const ExecParameters
*p
,
4247 uid_t uid
, gid_t gid
) {
4249 key_serial_t keyring
;
4258 /* Let's set up a new per-service "session" kernel keyring for each system service. This has the benefit that
4259 * each service runs with its own keyring shared among all processes of the service, but with no hook-up beyond
4260 * that scope, and in particular no link to the per-UID keyring. If we don't do this the keyring will be
4261 * automatically created on-demand and then linked to the per-UID keyring, by the kernel. The kernel's built-in
4262 * on-demand behaviour is very appropriate for login users, but probably not so much for system services, where
4263 * UIDs are not necessarily specific to a service but reused (at least in the case of UID 0). */
4265 if (context
->keyring_mode
== EXEC_KEYRING_INHERIT
)
4268 /* Acquiring a reference to the user keyring is nasty. We briefly change identity in order to get things set up
4269 * properly by the kernel. If we don't do that then we can't create it atomically, and that sucks for parallel
4270 * execution. This mimics what pam_keyinit does, too. Setting up session keyring, to be owned by the right user
4271 * & group is just as nasty as acquiring a reference to the user keyring. */
4273 saved_uid
= getuid();
4274 saved_gid
= getgid();
4276 if (gid_is_valid(gid
) && gid
!= saved_gid
) {
4277 if (setregid(gid
, -1) < 0)
4278 return log_unit_error_errno(u
, errno
, "Failed to change GID for user keyring: %m");
4281 if (uid_is_valid(uid
) && uid
!= saved_uid
) {
4282 if (setreuid(uid
, -1) < 0) {
4283 r
= log_unit_error_errno(u
, errno
, "Failed to change UID for user keyring: %m");
4288 keyring
= keyctl(KEYCTL_JOIN_SESSION_KEYRING
, 0, 0, 0, 0);
4289 if (keyring
== -1) {
4290 if (errno
== ENOSYS
)
4291 log_unit_debug_errno(u
, errno
, "Kernel keyring not supported, ignoring.");
4292 else if (ERRNO_IS_PRIVILEGE(errno
))
4293 log_unit_debug_errno(u
, errno
, "Kernel keyring access prohibited, ignoring.");
4294 else if (errno
== EDQUOT
)
4295 log_unit_debug_errno(u
, errno
, "Out of kernel keyrings to allocate, ignoring.");
4297 r
= log_unit_error_errno(u
, errno
, "Setting up kernel keyring failed: %m");
4302 /* When requested link the user keyring into the session keyring. */
4303 if (context
->keyring_mode
== EXEC_KEYRING_SHARED
) {
4305 if (keyctl(KEYCTL_LINK
,
4306 KEY_SPEC_USER_KEYRING
,
4307 KEY_SPEC_SESSION_KEYRING
, 0, 0) < 0) {
4308 r
= log_unit_error_errno(u
, errno
, "Failed to link user keyring into session keyring: %m");
4313 /* Restore uid/gid back */
4314 if (uid_is_valid(uid
) && uid
!= saved_uid
) {
4315 if (setreuid(saved_uid
, -1) < 0) {
4316 r
= log_unit_error_errno(u
, errno
, "Failed to change UID back for user keyring: %m");
4321 if (gid_is_valid(gid
) && gid
!= saved_gid
) {
4322 if (setregid(saved_gid
, -1) < 0)
4323 return log_unit_error_errno(u
, errno
, "Failed to change GID back for user keyring: %m");
4326 /* Populate they keyring with the invocation ID by default, as original saved_uid. */
4327 if (!sd_id128_is_null(u
->invocation_id
)) {
4330 key
= add_key("user", "invocation_id", &u
->invocation_id
, sizeof(u
->invocation_id
), KEY_SPEC_SESSION_KEYRING
);
4332 log_unit_debug_errno(u
, errno
, "Failed to add invocation ID to keyring, ignoring: %m");
4334 if (keyctl(KEYCTL_SETPERM
, key
,
4335 KEY_POS_VIEW
|KEY_POS_READ
|KEY_POS_SEARCH
|
4336 KEY_USR_VIEW
|KEY_USR_READ
|KEY_USR_SEARCH
, 0, 0) < 0)
4337 r
= log_unit_error_errno(u
, errno
, "Failed to restrict invocation ID permission: %m");
4342 /* Revert back uid & gid for the last time, and exit */
4343 /* no extra logging, as only the first already reported error matters */
4344 if (getuid() != saved_uid
)
4345 (void) setreuid(saved_uid
, -1);
4347 if (getgid() != saved_gid
)
4348 (void) setregid(saved_gid
, -1);
4353 static void append_socket_pair(int *array
, size_t *n
, const int pair
[static 2]) {
4359 array
[(*n
)++] = pair
[0];
4361 array
[(*n
)++] = pair
[1];
4364 static int close_remaining_fds(
4365 const ExecParameters
*params
,
4366 const ExecRuntime
*runtime
,
4369 const int *fds
, size_t n_fds
) {
4371 size_t n_dont_close
= 0;
4372 int dont_close
[n_fds
+ 14];
4376 if (params
->stdin_fd
>= 0)
4377 dont_close
[n_dont_close
++] = params
->stdin_fd
;
4378 if (params
->stdout_fd
>= 0)
4379 dont_close
[n_dont_close
++] = params
->stdout_fd
;
4380 if (params
->stderr_fd
>= 0)
4381 dont_close
[n_dont_close
++] = params
->stderr_fd
;
4384 dont_close
[n_dont_close
++] = socket_fd
;
4386 memcpy(dont_close
+ n_dont_close
, fds
, sizeof(int) * n_fds
);
4387 n_dont_close
+= n_fds
;
4391 append_socket_pair(dont_close
, &n_dont_close
, runtime
->ephemeral_storage_socket
);
4393 if (runtime
&& runtime
->shared
) {
4394 append_socket_pair(dont_close
, &n_dont_close
, runtime
->shared
->netns_storage_socket
);
4395 append_socket_pair(dont_close
, &n_dont_close
, runtime
->shared
->ipcns_storage_socket
);
4398 if (runtime
&& runtime
->dynamic_creds
) {
4399 if (runtime
->dynamic_creds
->user
)
4400 append_socket_pair(dont_close
, &n_dont_close
, runtime
->dynamic_creds
->user
->storage_socket
);
4401 if (runtime
->dynamic_creds
->group
)
4402 append_socket_pair(dont_close
, &n_dont_close
, runtime
->dynamic_creds
->group
->storage_socket
);
4405 if (user_lookup_fd
>= 0)
4406 dont_close
[n_dont_close
++] = user_lookup_fd
;
4408 return close_all_fds(dont_close
, n_dont_close
);
4411 static int send_user_lookup(
4419 /* Send the resolved UID/GID to PID 1 after we learnt it. We send a single datagram, containing the UID/GID
4420 * data as well as the unit name. Note that we suppress sending this if no user/group to resolve was
4423 if (user_lookup_fd
< 0)
4426 if (!uid_is_valid(uid
) && !gid_is_valid(gid
))
4429 if (writev(user_lookup_fd
,
4431 IOVEC_MAKE(&uid
, sizeof(uid
)),
4432 IOVEC_MAKE(&gid
, sizeof(gid
)),
4433 IOVEC_MAKE_STRING(unit
->id
) }, 3) < 0)
4439 static int acquire_home(const ExecContext
*c
, uid_t uid
, const char** home
, char **buf
) {
4446 /* If WorkingDirectory=~ is set, try to acquire a usable home directory. */
4451 if (!c
->working_directory_home
)
4454 r
= get_home_dir(buf
);
4462 static int compile_suggested_paths(const ExecContext
*c
, const ExecParameters
*p
, char ***ret
) {
4463 _cleanup_strv_free_
char ** list
= NULL
;
4470 assert(c
->dynamic_user
);
4472 /* Compile a list of paths that it might make sense to read the owning UID from to use as initial candidate for
4473 * dynamic UID allocation, in order to save us from doing costly recursive chown()s of the special
4476 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
4477 if (t
== EXEC_DIRECTORY_CONFIGURATION
)
4483 for (size_t i
= 0; i
< c
->directories
[t
].n_items
; i
++) {
4486 if (exec_directory_is_private(c
, t
))
4487 e
= path_join(p
->prefix
[t
], "private", c
->directories
[t
].items
[i
].path
);
4489 e
= path_join(p
->prefix
[t
], c
->directories
[t
].items
[i
].path
);
4493 r
= strv_consume(&list
, e
);
4499 *ret
= TAKE_PTR(list
);
4504 static int exec_parameters_get_cgroup_path(
4505 const ExecParameters
*params
,
4506 const CGroupContext
*c
,
4509 const char *subgroup
= NULL
;
4515 if (!params
->cgroup_path
)
4518 /* If we are called for a unit where cgroup delegation is on, and the payload created its own populated
4519 * subcgroup (which we expect it to do, after all it asked for delegation), then we cannot place the control
4520 * processes started after the main unit's process in the unit's main cgroup because it is now an inner one,
4521 * and inner cgroups may not contain processes. Hence, if delegation is on, and this is a control process,
4522 * let's use ".control" as subcgroup instead. Note that we do so only for ExecStartPost=, ExecReload=,
4523 * ExecStop=, ExecStopPost=, i.e. for the commands where the main process is already forked. For ExecStartPre=
4524 * this is not necessary, the cgroup is still empty. We distinguish these cases with the EXEC_CONTROL_CGROUP
4525 * flag, which is only passed for the former statements, not for the latter. */
4527 if (FLAGS_SET(params
->flags
, EXEC_CGROUP_DELEGATE
) && (FLAGS_SET(params
->flags
, EXEC_CONTROL_CGROUP
) || c
->delegate_subgroup
)) {
4528 if (FLAGS_SET(params
->flags
, EXEC_IS_CONTROL
))
4529 subgroup
= ".control";
4531 subgroup
= c
->delegate_subgroup
;
4535 p
= path_join(params
->cgroup_path
, subgroup
);
4537 p
= strdup(params
->cgroup_path
);
4545 static int exec_context_cpu_affinity_from_numa(const ExecContext
*c
, CPUSet
*ret
) {
4546 _cleanup_(cpu_set_reset
) CPUSet s
= {};
4552 if (!c
->numa_policy
.nodes
.set
) {
4553 log_debug("Can't derive CPU affinity mask from NUMA mask because NUMA mask is not set, ignoring");
4557 r
= numa_to_cpu_set(&c
->numa_policy
, &s
);
4563 return cpu_set_add_all(ret
, &s
);
4566 bool exec_context_get_cpu_affinity_from_numa(const ExecContext
*c
) {
4569 return c
->cpu_affinity_from_numa
;
4572 static int add_shifted_fd(int *fds
, size_t fds_size
, size_t *n_fds
, int fd
, int *ret_fd
) {
4577 assert(*n_fds
< fds_size
);
4585 if (fd
< 3 + (int) *n_fds
) {
4586 /* Let's move the fd up, so that it's outside of the fd range we will use to store
4587 * the fds we pass to the process (or which are closed only during execve). */
4589 r
= fcntl(fd
, F_DUPFD_CLOEXEC
, 3 + (int) *n_fds
);
4593 close_and_replace(fd
, r
);
4596 *ret_fd
= fds
[*n_fds
] = fd
;
4601 static int connect_unix_harder(Unit
*u
, const OpenFile
*of
, int ofd
) {
4602 union sockaddr_union addr
= {
4603 .un
.sun_family
= AF_UNIX
,
4606 static const int socket_types
[] = { SOCK_DGRAM
, SOCK_STREAM
, SOCK_SEQPACKET
};
4613 r
= sockaddr_un_set_path(&addr
.un
, FORMAT_PROC_FD_PATH(ofd
));
4615 return log_unit_error_errno(u
, r
, "Failed to set sockaddr for %s: %m", of
->path
);
4619 for (size_t i
= 0; i
< ELEMENTSOF(socket_types
); i
++) {
4620 _cleanup_close_
int fd
= -EBADF
;
4622 fd
= socket(AF_UNIX
, socket_types
[i
] | SOCK_CLOEXEC
, 0);
4624 return log_unit_error_errno(u
, errno
, "Failed to create socket for %s: %m", of
->path
);
4626 r
= RET_NERRNO(connect(fd
, &addr
.sa
, sa_len
));
4627 if (r
== -EPROTOTYPE
)
4630 return log_unit_error_errno(u
, r
, "Failed to connect socket for %s: %m", of
->path
);
4635 return log_unit_error_errno(u
, SYNTHETIC_ERRNO(EPROTOTYPE
), "Failed to connect socket for \"%s\".", of
->path
);
4638 static int get_open_file_fd(Unit
*u
, const OpenFile
*of
) {
4640 _cleanup_close_
int fd
= -EBADF
, ofd
= -EBADF
;
4645 ofd
= open(of
->path
, O_PATH
| O_CLOEXEC
);
4647 return log_unit_error_errno(u
, errno
, "Could not open \"%s\": %m", of
->path
);
4649 if (fstat(ofd
, &st
) < 0)
4650 return log_unit_error_errno(u
, errno
, "Failed to stat %s: %m", of
->path
);
4652 if (S_ISSOCK(st
.st_mode
)) {
4653 fd
= connect_unix_harder(u
, of
, ofd
);
4657 if (FLAGS_SET(of
->flags
, OPENFILE_READ_ONLY
) && shutdown(fd
, SHUT_WR
) < 0)
4658 return log_unit_error_errno(u
, errno
, "Failed to shutdown send for socket %s: %m",
4661 log_unit_debug(u
, "socket %s opened (fd=%d)", of
->path
, fd
);
4663 int flags
= FLAGS_SET(of
->flags
, OPENFILE_READ_ONLY
) ? O_RDONLY
: O_RDWR
;
4664 if (FLAGS_SET(of
->flags
, OPENFILE_APPEND
))
4666 else if (FLAGS_SET(of
->flags
, OPENFILE_TRUNCATE
))
4669 fd
= fd_reopen(ofd
, flags
| O_CLOEXEC
);
4671 return log_unit_error_errno(u
, fd
, "Failed to open file %s: %m", of
->path
);
4673 log_unit_debug(u
, "file %s opened (fd=%d)", of
->path
, fd
);
4679 static int collect_open_file_fds(
4681 OpenFile
* open_files
,
4692 LIST_FOREACH(open_files
, of
, open_files
) {
4693 _cleanup_close_
int fd
= -EBADF
;
4695 fd
= get_open_file_fd(u
, of
);
4697 if (FLAGS_SET(of
->flags
, OPENFILE_GRACEFUL
)) {
4698 log_unit_debug_errno(u
, fd
, "Failed to get OpenFile= file descriptor for %s, ignoring: %m", of
->path
);
4705 if (!GREEDY_REALLOC(*fds
, *n_fds
+ 1))
4708 r
= strv_extend(fdnames
, of
->fdname
);
4712 (*fds
)[*n_fds
] = TAKE_FD(fd
);
4720 static void log_command_line(Unit
*unit
, const char *msg
, const char *executable
, char **argv
) {
4728 _cleanup_free_
char *cmdline
= quote_command_line(argv
, SHELL_ESCAPE_EMPTY
);
4730 log_unit_struct(unit
, LOG_DEBUG
,
4731 "EXECUTABLE=%s", executable
,
4732 LOG_UNIT_MESSAGE(unit
, "%s: %s", msg
, strnull(cmdline
)),
4733 LOG_UNIT_INVOCATION_ID(unit
));
4736 static bool exec_context_need_unprivileged_private_users(
4737 const ExecContext
*context
,
4738 const ExecParameters
*params
) {
4743 /* These options require PrivateUsers= when used in user units, as we need to be in a user namespace
4744 * to have permission to enable them when not running as root. If we have effective CAP_SYS_ADMIN
4745 * (system manager) then we have privileges and don't need this. */
4746 if (params
->runtime_scope
!= RUNTIME_SCOPE_USER
)
4749 return context
->private_users
||
4750 context
->private_tmp
||
4751 context
->private_devices
||
4752 context
->private_network
||
4753 context
->network_namespace_path
||
4754 context
->private_ipc
||
4755 context
->ipc_namespace_path
||
4756 context
->private_mounts
> 0 ||
4757 context
->mount_apivfs
||
4758 context
->n_bind_mounts
> 0 ||
4759 context
->n_temporary_filesystems
> 0 ||
4760 context
->root_directory
||
4761 !strv_isempty(context
->extension_directories
) ||
4762 context
->protect_system
!= PROTECT_SYSTEM_NO
||
4763 context
->protect_home
!= PROTECT_HOME_NO
||
4764 context
->protect_kernel_tunables
||
4765 context
->protect_kernel_modules
||
4766 context
->protect_kernel_logs
||
4767 context
->protect_control_groups
||
4768 context
->protect_clock
||
4769 context
->protect_hostname
||
4770 !strv_isempty(context
->read_write_paths
) ||
4771 !strv_isempty(context
->read_only_paths
) ||
4772 !strv_isempty(context
->inaccessible_paths
) ||
4773 !strv_isempty(context
->exec_paths
) ||
4774 !strv_isempty(context
->no_exec_paths
);
4777 static int exec_child(
4779 const ExecCommand
*command
,
4780 const ExecContext
*context
,
4781 const ExecParameters
*params
,
4782 ExecRuntime
*runtime
,
4783 const CGroupContext
*cgroup_context
,
4785 const int named_iofds
[static 3],
4787 size_t n_socket_fds
,
4788 size_t n_storage_fds
,
4793 _cleanup_strv_free_
char **our_env
= NULL
, **pass_env
= NULL
, **joined_exec_search_path
= NULL
, **accum_env
= NULL
, **replaced_argv
= NULL
;
4794 int r
, ngids
= 0, exec_fd
;
4795 _cleanup_free_ gid_t
*supplementary_gids
= NULL
;
4796 const char *username
= NULL
, *groupname
= NULL
;
4797 _cleanup_free_
char *home_buffer
= NULL
, *memory_pressure_path
= NULL
;
4798 const char *home
= NULL
, *shell
= NULL
;
4799 char **final_argv
= NULL
;
4800 dev_t journal_stream_dev
= 0;
4801 ino_t journal_stream_ino
= 0;
4802 bool userns_set_up
= false;
4803 bool needs_sandboxing
, /* Do we need to set up full sandboxing? (i.e. all namespacing, all MAC stuff, caps, yadda yadda */
4804 needs_setuid
, /* Do we need to do the actual setresuid()/setresgid() calls? */
4805 needs_mount_namespace
, /* Do we need to set up a mount namespace for this kernel? */
4806 needs_ambient_hack
; /* Do we need to apply the ambient capabilities hack? */
4808 _cleanup_free_
char *mac_selinux_context_net
= NULL
;
4809 bool use_selinux
= false;
4812 bool use_smack
= false;
4815 bool use_apparmor
= false;
4817 uid_t saved_uid
= getuid();
4818 gid_t saved_gid
= getgid();
4819 uid_t uid
= UID_INVALID
;
4820 gid_t gid
= GID_INVALID
;
4821 size_t n_fds
= n_socket_fds
+ n_storage_fds
, /* fds to pass to the child */
4822 n_keep_fds
; /* total number of fds not to close */
4824 _cleanup_free_ gid_t
*gids_after_pam
= NULL
;
4825 int ngids_after_pam
= 0;
4826 _cleanup_free_
int *fds
= NULL
;
4827 _cleanup_strv_free_
char **fdnames
= NULL
;
4833 assert(exit_status
);
4835 /* Explicitly test for CVE-2021-4034 inspired invocations */
4836 assert(command
->path
);
4837 assert(!strv_isempty(command
->argv
));
4839 rename_process_from_path(command
->path
);
4841 /* We reset exactly these signals, since they are the only ones we set to SIG_IGN in the main
4842 * daemon. All others we leave untouched because we set them to SIG_DFL or a valid handler initially,
4843 * both of which will be demoted to SIG_DFL. */
4844 (void) default_signals(SIGNALS_CRASH_HANDLER
,
4847 if (context
->ignore_sigpipe
)
4848 (void) ignore_signals(SIGPIPE
);
4850 r
= reset_signal_mask();
4852 *exit_status
= EXIT_SIGNAL_MASK
;
4853 return log_unit_error_errno(unit
, r
, "Failed to set process signal mask: %m");
4856 if (params
->idle_pipe
)
4857 do_idle_pipe_dance(params
->idle_pipe
);
4859 /* Close fds we don't need very early to make sure we don't block init reexecution because it cannot bind its
4860 * sockets. Among the fds we close are the logging fds, and we want to keep them closed, so that we don't have
4861 * any fds open we don't really want open during the transition. In order to make logging work, we switch the
4862 * log subsystem into open_when_needed mode, so that it reopens the logs on every single log call. */
4865 log_set_open_when_needed(true);
4866 log_settle_target();
4868 /* In case anything used libc syslog(), close this here, too */
4871 fds
= newdup(int, params_fds
, n_fds
);
4873 *exit_status
= EXIT_MEMORY
;
4877 fdnames
= strv_copy((char**) params
->fd_names
);
4879 *exit_status
= EXIT_MEMORY
;
4883 r
= collect_open_file_fds(unit
, params
->open_files
, &fds
, &fdnames
, &n_fds
);
4885 *exit_status
= EXIT_FDS
;
4886 return log_unit_error_errno(unit
, r
, "Failed to get OpenFile= file descriptors: %m");
4889 int keep_fds
[n_fds
+ 3];
4890 memcpy_safe(keep_fds
, fds
, n_fds
* sizeof(int));
4893 r
= add_shifted_fd(keep_fds
, ELEMENTSOF(keep_fds
), &n_keep_fds
, params
->exec_fd
, &exec_fd
);
4895 *exit_status
= EXIT_FDS
;
4896 return log_unit_error_errno(unit
, r
, "Failed to shift fd and set FD_CLOEXEC: %m");
4900 if (unit
->manager
->restrict_fs
) {
4901 int bpf_map_fd
= lsm_bpf_map_restrict_fs_fd(unit
);
4902 if (bpf_map_fd
< 0) {
4903 *exit_status
= EXIT_FDS
;
4904 return log_unit_error_errno(unit
, bpf_map_fd
, "Failed to get restrict filesystems BPF map fd: %m");
4907 r
= add_shifted_fd(keep_fds
, ELEMENTSOF(keep_fds
), &n_keep_fds
, bpf_map_fd
, &bpf_map_fd
);
4909 *exit_status
= EXIT_FDS
;
4910 return log_unit_error_errno(unit
, r
, "Failed to shift fd and set FD_CLOEXEC: %m");
4915 r
= close_remaining_fds(params
, runtime
, user_lookup_fd
, socket_fd
, keep_fds
, n_keep_fds
);
4917 *exit_status
= EXIT_FDS
;
4918 return log_unit_error_errno(unit
, r
, "Failed to close unwanted file descriptors: %m");
4921 if (!context
->same_pgrp
&&
4923 *exit_status
= EXIT_SETSID
;
4924 return log_unit_error_errno(unit
, errno
, "Failed to create new process session: %m");
4927 exec_context_tty_reset(context
, params
);
4929 if (unit_shall_confirm_spawn(unit
)) {
4930 _cleanup_free_
char *cmdline
= NULL
;
4932 cmdline
= quote_command_line(command
->argv
, SHELL_ESCAPE_EMPTY
);
4934 *exit_status
= EXIT_MEMORY
;
4938 r
= ask_for_confirmation(context
, params
->confirm_spawn
, unit
, cmdline
);
4939 if (r
!= CONFIRM_EXECUTE
) {
4940 if (r
== CONFIRM_PRETEND_SUCCESS
) {
4941 *exit_status
= EXIT_SUCCESS
;
4944 *exit_status
= EXIT_CONFIRM
;
4945 return log_unit_error_errno(unit
, SYNTHETIC_ERRNO(ECANCELED
),
4946 "Execution cancelled by the user");
4950 /* We are about to invoke NSS and PAM modules. Let's tell them what we are doing here, maybe they care. This is
4951 * used by nss-resolve to disable itself when we are about to start systemd-resolved, to avoid deadlocks. Note
4952 * that these env vars do not survive the execve(), which means they really only apply to the PAM and NSS
4953 * invocations themselves. Also note that while we'll only invoke NSS modules involved in user management they
4954 * might internally call into other NSS modules that are involved in hostname resolution, we never know. */
4955 if (setenv("SYSTEMD_ACTIVATION_UNIT", unit
->id
, true) != 0 ||
4956 setenv("SYSTEMD_ACTIVATION_SCOPE", runtime_scope_to_string(params
->runtime_scope
), true) != 0) {
4957 *exit_status
= EXIT_MEMORY
;
4958 return log_unit_error_errno(unit
, errno
, "Failed to update environment: %m");
4961 if (context
->dynamic_user
&& runtime
&& runtime
->dynamic_creds
) {
4962 _cleanup_strv_free_
char **suggested_paths
= NULL
;
4964 /* On top of that, make sure we bypass our own NSS module nss-systemd comprehensively for any NSS
4965 * checks, if DynamicUser=1 is used, as we shouldn't create a feedback loop with ourselves here. */
4966 if (putenv((char*) "SYSTEMD_NSS_DYNAMIC_BYPASS=1") != 0) {
4967 *exit_status
= EXIT_USER
;
4968 return log_unit_error_errno(unit
, errno
, "Failed to update environment: %m");
4971 r
= compile_suggested_paths(context
, params
, &suggested_paths
);
4973 *exit_status
= EXIT_MEMORY
;
4977 r
= dynamic_creds_realize(runtime
->dynamic_creds
, suggested_paths
, &uid
, &gid
);
4979 *exit_status
= EXIT_USER
;
4981 return log_unit_error_errno(unit
, SYNTHETIC_ERRNO(EOPNOTSUPP
),
4982 "Failed to update dynamic user credentials: User or group with specified name already exists.");
4983 return log_unit_error_errno(unit
, r
, "Failed to update dynamic user credentials: %m");
4986 if (!uid_is_valid(uid
)) {
4987 *exit_status
= EXIT_USER
;
4988 return log_unit_error_errno(unit
, SYNTHETIC_ERRNO(ESRCH
), "UID validation failed for \""UID_FMT
"\"", uid
);
4991 if (!gid_is_valid(gid
)) {
4992 *exit_status
= EXIT_USER
;
4993 return log_unit_error_errno(unit
, SYNTHETIC_ERRNO(ESRCH
), "GID validation failed for \""GID_FMT
"\"", gid
);
4996 if (runtime
->dynamic_creds
->user
)
4997 username
= runtime
->dynamic_creds
->user
->name
;
5000 r
= get_fixed_user(context
, &username
, &uid
, &gid
, &home
, &shell
);
5002 *exit_status
= EXIT_USER
;
5003 return log_unit_error_errno(unit
, r
, "Failed to determine user credentials: %m");
5006 r
= get_fixed_group(context
, &groupname
, &gid
);
5008 *exit_status
= EXIT_GROUP
;
5009 return log_unit_error_errno(unit
, r
, "Failed to determine group credentials: %m");
5013 /* Initialize user supplementary groups and get SupplementaryGroups= ones */
5014 r
= get_supplementary_groups(context
, username
, groupname
, gid
,
5015 &supplementary_gids
, &ngids
);
5017 *exit_status
= EXIT_GROUP
;
5018 return log_unit_error_errno(unit
, r
, "Failed to determine supplementary groups: %m");
5021 r
= send_user_lookup(unit
, user_lookup_fd
, uid
, gid
);
5023 *exit_status
= EXIT_USER
;
5024 return log_unit_error_errno(unit
, r
, "Failed to send user credentials to PID1: %m");
5027 user_lookup_fd
= safe_close(user_lookup_fd
);
5029 r
= acquire_home(context
, uid
, &home
, &home_buffer
);
5031 *exit_status
= EXIT_CHDIR
;
5032 return log_unit_error_errno(unit
, r
, "Failed to determine $HOME for user: %m");
5035 /* If a socket is connected to STDIN/STDOUT/STDERR, we must drop O_NONBLOCK */
5037 (void) fd_nonblock(socket_fd
, false);
5039 /* Journald will try to look-up our cgroup in order to populate _SYSTEMD_CGROUP and _SYSTEMD_UNIT fields.
5040 * Hence we need to migrate to the target cgroup from init.scope before connecting to journald */
5041 if (params
->cgroup_path
) {
5042 _cleanup_free_
char *p
= NULL
;
5044 r
= exec_parameters_get_cgroup_path(params
, cgroup_context
, &p
);
5046 *exit_status
= EXIT_CGROUP
;
5047 return log_unit_error_errno(unit
, r
, "Failed to acquire cgroup path: %m");
5050 r
= cg_attach_everywhere(params
->cgroup_supported
, p
, 0, NULL
, NULL
);
5051 if (r
== -EUCLEAN
) {
5052 *exit_status
= EXIT_CGROUP
;
5053 return log_unit_error_errno(unit
, r
, "Failed to attach process to cgroup %s "
5054 "because the cgroup or one of its parents or "
5055 "siblings is in the threaded mode: %m", p
);
5058 *exit_status
= EXIT_CGROUP
;
5059 return log_unit_error_errno(unit
, r
, "Failed to attach to cgroup %s: %m", p
);
5063 if (context
->network_namespace_path
&& runtime
&& runtime
->shared
&& runtime
->shared
->netns_storage_socket
[0] >= 0) {
5064 r
= open_shareable_ns_path(runtime
->shared
->netns_storage_socket
, context
->network_namespace_path
, CLONE_NEWNET
);
5066 *exit_status
= EXIT_NETWORK
;
5067 return log_unit_error_errno(unit
, r
, "Failed to open network namespace path %s: %m", context
->network_namespace_path
);
5071 if (context
->ipc_namespace_path
&& runtime
&& runtime
->shared
&& runtime
->shared
->ipcns_storage_socket
[0] >= 0) {
5072 r
= open_shareable_ns_path(runtime
->shared
->ipcns_storage_socket
, context
->ipc_namespace_path
, CLONE_NEWIPC
);
5074 *exit_status
= EXIT_NAMESPACE
;
5075 return log_unit_error_errno(unit
, r
, "Failed to open IPC namespace path %s: %m", context
->ipc_namespace_path
);
5079 r
= setup_input(context
, params
, socket_fd
, named_iofds
);
5081 *exit_status
= EXIT_STDIN
;
5082 return log_unit_error_errno(unit
, r
, "Failed to set up standard input: %m");
5085 r
= setup_output(unit
, context
, params
, STDOUT_FILENO
, socket_fd
, named_iofds
, basename(command
->path
), uid
, gid
, &journal_stream_dev
, &journal_stream_ino
);
5087 *exit_status
= EXIT_STDOUT
;
5088 return log_unit_error_errno(unit
, r
, "Failed to set up standard output: %m");
5091 r
= setup_output(unit
, context
, params
, STDERR_FILENO
, socket_fd
, named_iofds
, basename(command
->path
), uid
, gid
, &journal_stream_dev
, &journal_stream_ino
);
5093 *exit_status
= EXIT_STDERR
;
5094 return log_unit_error_errno(unit
, r
, "Failed to set up standard error output: %m");
5097 if (context
->oom_score_adjust_set
) {
5098 /* When we can't make this change due to EPERM, then let's silently skip over it. User namespaces
5099 * prohibit write access to this file, and we shouldn't trip up over that. */
5100 r
= set_oom_score_adjust(context
->oom_score_adjust
);
5101 if (ERRNO_IS_PRIVILEGE(r
))
5102 log_unit_debug_errno(unit
, r
, "Failed to adjust OOM setting, assuming containerized execution, ignoring: %m");
5104 *exit_status
= EXIT_OOM_ADJUST
;
5105 return log_unit_error_errno(unit
, r
, "Failed to adjust OOM setting: %m");
5109 if (context
->coredump_filter_set
) {
5110 r
= set_coredump_filter(context
->coredump_filter
);
5111 if (ERRNO_IS_PRIVILEGE(r
))
5112 log_unit_debug_errno(unit
, r
, "Failed to adjust coredump_filter, ignoring: %m");
5114 return log_unit_error_errno(unit
, r
, "Failed to adjust coredump_filter: %m");
5117 if (context
->nice_set
) {
5118 r
= setpriority_closest(context
->nice
);
5120 return log_unit_error_errno(unit
, r
, "Failed to set up process scheduling priority (nice level): %m");
5123 if (context
->cpu_sched_set
) {
5124 struct sched_param param
= {
5125 .sched_priority
= context
->cpu_sched_priority
,
5128 r
= sched_setscheduler(0,
5129 context
->cpu_sched_policy
|
5130 (context
->cpu_sched_reset_on_fork
?
5131 SCHED_RESET_ON_FORK
: 0),
5134 *exit_status
= EXIT_SETSCHEDULER
;
5135 return log_unit_error_errno(unit
, errno
, "Failed to set up CPU scheduling: %m");
5139 if (context
->cpu_affinity_from_numa
|| context
->cpu_set
.set
) {
5140 _cleanup_(cpu_set_reset
) CPUSet converted_cpu_set
= {};
5141 const CPUSet
*cpu_set
;
5143 if (context
->cpu_affinity_from_numa
) {
5144 r
= exec_context_cpu_affinity_from_numa(context
, &converted_cpu_set
);
5146 *exit_status
= EXIT_CPUAFFINITY
;
5147 return log_unit_error_errno(unit
, r
, "Failed to derive CPU affinity mask from NUMA mask: %m");
5150 cpu_set
= &converted_cpu_set
;
5152 cpu_set
= &context
->cpu_set
;
5154 if (sched_setaffinity(0, cpu_set
->allocated
, cpu_set
->set
) < 0) {
5155 *exit_status
= EXIT_CPUAFFINITY
;
5156 return log_unit_error_errno(unit
, errno
, "Failed to set up CPU affinity: %m");
5160 if (mpol_is_valid(numa_policy_get_type(&context
->numa_policy
))) {
5161 r
= apply_numa_policy(&context
->numa_policy
);
5163 if (ERRNO_IS_NOT_SUPPORTED(r
))
5164 log_unit_debug_errno(unit
, r
, "NUMA support not available, ignoring.");
5166 *exit_status
= EXIT_NUMA_POLICY
;
5167 return log_unit_error_errno(unit
, r
, "Failed to set NUMA memory policy: %m");
5172 if (context
->ioprio_set
)
5173 if (ioprio_set(IOPRIO_WHO_PROCESS
, 0, context
->ioprio
) < 0) {
5174 *exit_status
= EXIT_IOPRIO
;
5175 return log_unit_error_errno(unit
, errno
, "Failed to set up IO scheduling priority: %m");
5178 if (context
->timer_slack_nsec
!= NSEC_INFINITY
)
5179 if (prctl(PR_SET_TIMERSLACK
, context
->timer_slack_nsec
) < 0) {
5180 *exit_status
= EXIT_TIMERSLACK
;
5181 return log_unit_error_errno(unit
, errno
, "Failed to set up timer slack: %m");
5184 if (context
->personality
!= PERSONALITY_INVALID
) {
5185 r
= safe_personality(context
->personality
);
5187 *exit_status
= EXIT_PERSONALITY
;
5188 return log_unit_error_errno(unit
, r
, "Failed to set up execution domain (personality): %m");
5192 if (context
->utmp_id
) {
5193 const char *line
= context
->tty_path
?
5194 (path_startswith(context
->tty_path
, "/dev/") ?: context
->tty_path
) :
5196 utmp_put_init_process(context
->utmp_id
, getpid_cached(), getsid(0),
5198 context
->utmp_mode
== EXEC_UTMP_INIT
? INIT_PROCESS
:
5199 context
->utmp_mode
== EXEC_UTMP_LOGIN
? LOGIN_PROCESS
:
5204 if (uid_is_valid(uid
)) {
5205 r
= chown_terminal(STDIN_FILENO
, uid
);
5207 *exit_status
= EXIT_STDIN
;
5208 return log_unit_error_errno(unit
, r
, "Failed to change ownership of terminal: %m");
5212 if (params
->cgroup_path
) {
5213 /* If delegation is enabled we'll pass ownership of the cgroup to the user of the new process. On cgroup v1
5214 * this is only about systemd's own hierarchy, i.e. not the controller hierarchies, simply because that's not
5215 * safe. On cgroup v2 there's only one hierarchy anyway, and delegation is safe there, hence in that case only
5216 * touch a single hierarchy too. */
5218 if (params
->flags
& EXEC_CGROUP_DELEGATE
) {
5219 _cleanup_free_
char *p
= NULL
;
5221 r
= cg_set_access(SYSTEMD_CGROUP_CONTROLLER
, params
->cgroup_path
, uid
, gid
);
5223 *exit_status
= EXIT_CGROUP
;
5224 return log_unit_error_errno(unit
, r
, "Failed to adjust control group access: %m");
5227 r
= exec_parameters_get_cgroup_path(params
, cgroup_context
, &p
);
5229 *exit_status
= EXIT_CGROUP
;
5230 return log_unit_error_errno(unit
, r
, "Failed to acquire cgroup path: %m");
5233 r
= cg_set_access_recursive(SYSTEMD_CGROUP_CONTROLLER
, p
, uid
, gid
);
5235 *exit_status
= EXIT_CGROUP
;
5236 return log_unit_error_errno(unit
, r
, "Failed to adjust control subgroup access: %m");
5241 if (cgroup_context
&& cg_unified() > 0 && is_pressure_supported() > 0) {
5242 if (cgroup_context_want_memory_pressure(cgroup_context
)) {
5243 r
= cg_get_path("memory", params
->cgroup_path
, "memory.pressure", &memory_pressure_path
);
5245 *exit_status
= EXIT_MEMORY
;
5249 r
= chmod_and_chown(memory_pressure_path
, 0644, uid
, gid
);
5251 log_unit_full_errno(unit
, r
== -ENOENT
|| ERRNO_IS_PRIVILEGE(r
) ? LOG_DEBUG
: LOG_WARNING
, r
,
5252 "Failed to adjust ownership of '%s', ignoring: %m", memory_pressure_path
);
5253 memory_pressure_path
= mfree(memory_pressure_path
);
5255 } else if (cgroup_context
->memory_pressure_watch
== CGROUP_PRESSURE_WATCH_OFF
) {
5256 memory_pressure_path
= strdup("/dev/null"); /* /dev/null is explicit indicator for turning of memory pressure watch */
5257 if (!memory_pressure_path
) {
5258 *exit_status
= EXIT_MEMORY
;
5265 needs_mount_namespace
= exec_needs_mount_namespace(context
, params
, runtime
);
5267 for (ExecDirectoryType dt
= 0; dt
< _EXEC_DIRECTORY_TYPE_MAX
; dt
++) {
5268 r
= setup_exec_directory(unit
, context
, params
, uid
, gid
, dt
, needs_mount_namespace
, exit_status
);
5270 return log_unit_error_errno(unit
, r
, "Failed to set up special execution directory in %s: %m", params
->prefix
[dt
]);
5273 if (FLAGS_SET(params
->flags
, EXEC_WRITE_CREDENTIALS
)) {
5274 r
= setup_credentials(context
, params
, unit
->id
, uid
);
5276 *exit_status
= EXIT_CREDENTIALS
;
5277 return log_unit_error_errno(unit
, r
, "Failed to set up credentials: %m");
5281 r
= build_environment(
5293 memory_pressure_path
,
5296 *exit_status
= EXIT_MEMORY
;
5300 r
= build_pass_environment(context
, &pass_env
);
5302 *exit_status
= EXIT_MEMORY
;
5306 /* The $PATH variable is set to the default path in params->environment. However, this is overridden
5307 * if user-specified fields have $PATH set. The intention is to also override $PATH if the unit does
5308 * not specify PATH but the unit has ExecSearchPath. */
5309 if (!strv_isempty(context
->exec_search_path
)) {
5310 _cleanup_free_
char *joined
= NULL
;
5312 joined
= strv_join(context
->exec_search_path
, ":");
5314 *exit_status
= EXIT_MEMORY
;
5318 r
= strv_env_assign(&joined_exec_search_path
, "PATH", joined
);
5320 *exit_status
= EXIT_MEMORY
;
5325 accum_env
= strv_env_merge(params
->environment
,
5327 joined_exec_search_path
,
5329 context
->environment
,
5332 *exit_status
= EXIT_MEMORY
;
5335 accum_env
= strv_env_clean(accum_env
);
5337 (void) umask(context
->umask
);
5339 r
= setup_keyring(unit
, context
, params
, uid
, gid
);
5341 *exit_status
= EXIT_KEYRING
;
5342 return log_unit_error_errno(unit
, r
, "Failed to set up kernel keyring: %m");
5345 /* We need sandboxing if the caller asked us to apply it and the command isn't explicitly excepted
5347 needs_sandboxing
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && !(command
->flags
& EXEC_COMMAND_FULLY_PRIVILEGED
);
5349 /* We need the ambient capability hack, if the caller asked us to apply it and the command is marked
5350 * for it, and the kernel doesn't actually support ambient caps. */
5351 needs_ambient_hack
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && (command
->flags
& EXEC_COMMAND_AMBIENT_MAGIC
) && !ambient_capabilities_supported();
5353 /* We need setresuid() if the caller asked us to apply sandboxing and the command isn't explicitly
5354 * excepted from either whole sandboxing or just setresuid() itself, and the ambient hack is not
5356 if (needs_ambient_hack
)
5357 needs_setuid
= false;
5359 needs_setuid
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && !(command
->flags
& (EXEC_COMMAND_FULLY_PRIVILEGED
|EXEC_COMMAND_NO_SETUID
));
5361 uint64_t capability_ambient_set
= context
->capability_ambient_set
;
5363 if (needs_sandboxing
) {
5364 /* MAC enablement checks need to be done before a new mount ns is created, as they rely on
5365 * /sys being present. The actual MAC context application will happen later, as late as
5366 * possible, to avoid impacting our own code paths. */
5369 use_selinux
= mac_selinux_use();
5372 use_smack
= mac_smack_use();
5375 use_apparmor
= mac_apparmor_use();
5379 if (needs_sandboxing
) {
5382 /* Let's set the resource limits before we call into PAM, so that pam_limits wins over what
5383 * is set here. (See below.) */
5385 r
= setrlimit_closest_all((const struct rlimit
* const *) context
->rlimit
, &which_failed
);
5387 *exit_status
= EXIT_LIMITS
;
5388 return log_unit_error_errno(unit
, r
, "Failed to adjust resource limit RLIMIT_%s: %m", rlimit_to_string(which_failed
));
5392 if (needs_setuid
&& context
->pam_name
&& username
) {
5393 /* Let's call into PAM after we set up our own idea of resource limits to that pam_limits
5394 * wins here. (See above.) */
5396 /* All fds passed in the fds array will be closed in the pam child process. */
5397 r
= setup_pam(context
->pam_name
, username
, uid
, gid
, context
->tty_path
, &accum_env
, fds
, n_fds
);
5399 *exit_status
= EXIT_PAM
;
5400 return log_unit_error_errno(unit
, r
, "Failed to set up PAM session: %m");
5403 if (ambient_capabilities_supported()) {
5404 uint64_t ambient_after_pam
;
5406 /* PAM modules might have set some ambient caps. Query them here and merge them into
5407 * the caps we want to set in the end, so that we don't end up unsetting them. */
5408 r
= capability_get_ambient(&ambient_after_pam
);
5410 *exit_status
= EXIT_CAPABILITIES
;
5411 return log_unit_error_errno(unit
, r
, "Failed to query ambient caps: %m");
5414 capability_ambient_set
|= ambient_after_pam
;
5417 ngids_after_pam
= getgroups_alloc(&gids_after_pam
);
5418 if (ngids_after_pam
< 0) {
5419 *exit_status
= EXIT_MEMORY
;
5420 return log_unit_error_errno(unit
, ngids_after_pam
, "Failed to obtain groups after setting up PAM: %m");
5424 if (needs_sandboxing
&& exec_context_need_unprivileged_private_users(context
, params
)) {
5425 /* If we're unprivileged, set up the user namespace first to enable use of the other namespaces.
5426 * Users with CAP_SYS_ADMIN can set up user namespaces last because they will be able to
5427 * set up the all of the other namespaces (i.e. network, mount, UTS) without a user namespace. */
5429 r
= setup_private_users(saved_uid
, saved_gid
, uid
, gid
);
5430 /* If it was requested explicitly and we can't set it up, fail early. Otherwise, continue and let
5431 * the actual requested operations fail (or silently continue). */
5432 if (r
< 0 && context
->private_users
) {
5433 *exit_status
= EXIT_USER
;
5434 return log_unit_error_errno(unit
, r
, "Failed to set up user namespacing for unprivileged user: %m");
5437 log_unit_info_errno(unit
, r
, "Failed to set up user namespacing for unprivileged user, ignoring: %m");
5439 userns_set_up
= true;
5442 if (exec_needs_network_namespace(context
) && runtime
&& runtime
->shared
&& runtime
->shared
->netns_storage_socket
[0] >= 0) {
5444 /* Try to enable network namespacing if network namespacing is available and we have
5445 * CAP_NET_ADMIN. We need CAP_NET_ADMIN to be able to configure the loopback device in the
5446 * new network namespace. And if we don't have that, then we could only create a network
5447 * namespace without the ability to set up "lo". Hence gracefully skip things then. */
5448 if (ns_type_supported(NAMESPACE_NET
) && have_effective_cap(CAP_NET_ADMIN
) > 0) {
5449 r
= setup_shareable_ns(runtime
->shared
->netns_storage_socket
, CLONE_NEWNET
);
5451 if (ERRNO_IS_PRIVILEGE(r
))
5452 log_unit_notice_errno(unit
, r
,
5453 "PrivateNetwork=yes is configured, but network namespace setup not permitted, proceeding without: %m");
5455 *exit_status
= EXIT_NETWORK
;
5456 return log_unit_error_errno(unit
, r
, "Failed to set up network namespacing: %m");
5459 } else if (context
->network_namespace_path
) {
5460 *exit_status
= EXIT_NETWORK
;
5461 return log_unit_error_errno(unit
, SYNTHETIC_ERRNO(EOPNOTSUPP
),
5462 "NetworkNamespacePath= is not supported, refusing.");
5464 log_unit_notice(unit
, "PrivateNetwork=yes is configured, but the kernel does not support or we lack privileges for network namespace, proceeding without.");
5467 if (exec_needs_ipc_namespace(context
) && runtime
&& runtime
->shared
&& runtime
->shared
->ipcns_storage_socket
[0] >= 0) {
5469 if (ns_type_supported(NAMESPACE_IPC
)) {
5470 r
= setup_shareable_ns(runtime
->shared
->ipcns_storage_socket
, CLONE_NEWIPC
);
5472 log_unit_warning_errno(unit
, r
,
5473 "PrivateIPC=yes is configured, but IPC namespace setup failed, ignoring: %m");
5475 *exit_status
= EXIT_NAMESPACE
;
5476 return log_unit_error_errno(unit
, r
, "Failed to set up IPC namespacing: %m");
5478 } else if (context
->ipc_namespace_path
) {
5479 *exit_status
= EXIT_NAMESPACE
;
5480 return log_unit_error_errno(unit
, SYNTHETIC_ERRNO(EOPNOTSUPP
),
5481 "IPCNamespacePath= is not supported, refusing.");
5483 log_unit_warning(unit
, "PrivateIPC=yes is configured, but the kernel does not support IPC namespaces, ignoring.");
5486 if (needs_mount_namespace
) {
5487 _cleanup_free_
char *error_path
= NULL
;
5489 r
= apply_mount_namespace(unit
, command
->flags
, context
, params
, runtime
, memory_pressure_path
, &error_path
);
5491 *exit_status
= EXIT_NAMESPACE
;
5492 return log_unit_error_errno(unit
, r
, "Failed to set up mount namespacing%s%s: %m",
5493 error_path
? ": " : "", strempty(error_path
));
5497 if (needs_sandboxing
) {
5498 r
= apply_protect_hostname(unit
, context
, exit_status
);
5503 if (context
->memory_ksm
>= 0)
5504 if (prctl(PR_SET_MEMORY_MERGE
, context
->memory_ksm
) < 0) {
5505 if (ERRNO_IS_NOT_SUPPORTED(errno
))
5506 log_unit_debug_errno(unit
, errno
, "KSM support not available, ignoring.");
5508 *exit_status
= EXIT_KSM
;
5509 return log_unit_error_errno(unit
, errno
, "Failed to set KSM: %m");
5513 /* Drop groups as early as possible.
5514 * This needs to be done after PrivateDevices=y setup as device nodes should be owned by the host's root.
5515 * For non-root in a userns, devices will be owned by the user/group before the group change, and nobody. */
5517 _cleanup_free_ gid_t
*gids_to_enforce
= NULL
;
5518 int ngids_to_enforce
= 0;
5520 ngids_to_enforce
= merge_gid_lists(supplementary_gids
,
5525 if (ngids_to_enforce
< 0) {
5526 *exit_status
= EXIT_MEMORY
;
5527 return log_unit_error_errno(unit
,
5529 "Failed to merge group lists. Group membership might be incorrect: %m");
5532 r
= enforce_groups(gid
, gids_to_enforce
, ngids_to_enforce
);
5534 *exit_status
= EXIT_GROUP
;
5535 return log_unit_error_errno(unit
, r
, "Changing group credentials failed: %m");
5539 /* If the user namespace was not set up above, try to do it now.
5540 * It's preferred to set up the user namespace later (after all other namespaces) so as not to be
5541 * restricted by rules pertaining to combining user namespaces with other namespaces (e.g. in the
5542 * case of mount namespaces being less privileged when the mount point list is copied from a
5543 * different user namespace). */
5545 if (needs_sandboxing
&& context
->private_users
&& !userns_set_up
) {
5546 r
= setup_private_users(saved_uid
, saved_gid
, uid
, gid
);
5548 *exit_status
= EXIT_USER
;
5549 return log_unit_error_errno(unit
, r
, "Failed to set up user namespacing: %m");
5553 /* Now that the mount namespace has been set up and privileges adjusted, let's look for the thing we
5556 _cleanup_free_
char *executable
= NULL
;
5557 _cleanup_close_
int executable_fd
= -EBADF
;
5558 r
= find_executable_full(command
->path
, /* root= */ NULL
, context
->exec_search_path
, false, &executable
, &executable_fd
);
5560 if (r
!= -ENOMEM
&& (command
->flags
& EXEC_COMMAND_IGNORE_FAILURE
)) {
5561 log_unit_struct_errno(unit
, LOG_INFO
, r
,
5562 "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR
,
5563 LOG_UNIT_INVOCATION_ID(unit
),
5564 LOG_UNIT_MESSAGE(unit
, "Executable %s missing, skipping: %m",
5566 "EXECUTABLE=%s", command
->path
);
5570 *exit_status
= EXIT_EXEC
;
5572 return log_unit_struct_errno(unit
, LOG_INFO
, r
,
5573 "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR
,
5574 LOG_UNIT_INVOCATION_ID(unit
),
5575 LOG_UNIT_MESSAGE(unit
, "Failed to locate executable %s: %m",
5577 "EXECUTABLE=%s", command
->path
);
5580 r
= add_shifted_fd(keep_fds
, ELEMENTSOF(keep_fds
), &n_keep_fds
, executable_fd
, &executable_fd
);
5582 *exit_status
= EXIT_FDS
;
5583 return log_unit_error_errno(unit
, r
, "Failed to shift fd and set FD_CLOEXEC: %m");
5587 if (needs_sandboxing
&& use_selinux
&& params
->selinux_context_net
) {
5592 else if (params
->n_socket_fds
== 1)
5593 /* If stdin is not connected to a socket but we are triggered by exactly one socket unit then we
5594 * use context from that fd to compute the label. */
5595 fd
= params
->fds
[0];
5598 r
= mac_selinux_get_child_mls_label(fd
, executable
, context
->selinux_context
, &mac_selinux_context_net
);
5600 if (!context
->selinux_context_ignore
) {
5601 *exit_status
= EXIT_SELINUX_CONTEXT
;
5602 return log_unit_error_errno(unit
, r
, "Failed to determine SELinux context: %m");
5604 log_unit_debug_errno(unit
, r
, "Failed to determine SELinux context, ignoring: %m");
5610 /* We repeat the fd closing here, to make sure that nothing is leaked from the PAM modules. Note that
5611 * we are more aggressive this time, since we don't need socket_fd and the netns and ipcns fds any
5612 * more. We do keep exec_fd however, if we have it, since we need to keep it open until the final
5615 r
= close_all_fds(keep_fds
, n_keep_fds
);
5617 r
= shift_fds(fds
, n_fds
);
5619 r
= flags_fds(fds
, n_socket_fds
, n_fds
, context
->non_blocking
);
5621 *exit_status
= EXIT_FDS
;
5622 return log_unit_error_errno(unit
, r
, "Failed to adjust passed file descriptors: %m");
5625 /* At this point, the fds we want to pass to the program are all ready and set up, with O_CLOEXEC turned off
5626 * and at the right fd numbers. The are no other fds open, with one exception: the exec_fd if it is defined,
5627 * and it has O_CLOEXEC set, after all we want it to be closed by the execve(), so that our parent knows we
5630 secure_bits
= context
->secure_bits
;
5632 if (needs_sandboxing
) {
5635 /* Set the RTPRIO resource limit to 0, but only if nothing else was explicitly requested.
5636 * (Note this is placed after the general resource limit initialization, see above, in order
5637 * to take precedence.) */
5638 if (context
->restrict_realtime
&& !context
->rlimit
[RLIMIT_RTPRIO
]) {
5639 if (setrlimit(RLIMIT_RTPRIO
, &RLIMIT_MAKE_CONST(0)) < 0) {
5640 *exit_status
= EXIT_LIMITS
;
5641 return log_unit_error_errno(unit
, errno
, "Failed to adjust RLIMIT_RTPRIO resource limit: %m");
5646 /* LSM Smack needs the capability CAP_MAC_ADMIN to change the current execution security context of the
5647 * process. This is the latest place before dropping capabilities. Other MAC context are set later. */
5649 r
= setup_smack(unit
->manager
, context
, executable_fd
);
5650 if (r
< 0 && !context
->smack_process_label_ignore
) {
5651 *exit_status
= EXIT_SMACK_PROCESS_LABEL
;
5652 return log_unit_error_errno(unit
, r
, "Failed to set SMACK process label: %m");
5657 bset
= context
->capability_bounding_set
;
5658 /* If the ambient caps hack is enabled (which means the kernel can't do them, and the user asked for
5659 * our magic fallback), then let's add some extra caps, so that the service can drop privs of its own,
5660 * instead of us doing that */
5661 if (needs_ambient_hack
)
5662 bset
|= (UINT64_C(1) << CAP_SETPCAP
) |
5663 (UINT64_C(1) << CAP_SETUID
) |
5664 (UINT64_C(1) << CAP_SETGID
);
5666 if (!cap_test_all(bset
)) {
5667 r
= capability_bounding_set_drop(bset
, /* right_now= */ false);
5669 *exit_status
= EXIT_CAPABILITIES
;
5670 return log_unit_error_errno(unit
, r
, "Failed to drop capabilities: %m");
5674 /* Ambient capabilities are cleared during setresuid() (in enforce_user()) even with
5677 * To be able to raise the ambient capabilities after setresuid() they have to be added to
5678 * the inherited set and keep caps has to be set (done in enforce_user()). After setresuid()
5679 * the ambient capabilities can be raised as they are present in the permitted and
5680 * inhertiable set. However it is possible that someone wants to set ambient capabilities
5681 * without changing the user, so we also set the ambient capabilities here.
5683 * The requested ambient capabilities are raised in the inheritable set if the second
5684 * argument is true. */
5685 if (!needs_ambient_hack
) {
5686 r
= capability_ambient_set_apply(capability_ambient_set
, /* also_inherit= */ true);
5688 *exit_status
= EXIT_CAPABILITIES
;
5689 return log_unit_error_errno(unit
, r
, "Failed to apply ambient capabilities (before UID change): %m");
5694 /* chroot to root directory first, before we lose the ability to chroot */
5695 r
= apply_root_directory(context
, params
, runtime
, needs_mount_namespace
, exit_status
);
5697 return log_unit_error_errno(unit
, r
, "Chrooting to the requested root directory failed: %m");
5700 if (uid_is_valid(uid
)) {
5701 r
= enforce_user(context
, uid
, capability_ambient_set
);
5703 *exit_status
= EXIT_USER
;
5704 return log_unit_error_errno(unit
, r
, "Failed to change UID to " UID_FMT
": %m", uid
);
5707 if (!needs_ambient_hack
&& capability_ambient_set
!= 0) {
5709 /* Raise the ambient capabilities after user change. */
5710 r
= capability_ambient_set_apply(capability_ambient_set
, /* also_inherit= */ false);
5712 *exit_status
= EXIT_CAPABILITIES
;
5713 return log_unit_error_errno(unit
, r
, "Failed to apply ambient capabilities (after UID change): %m");
5719 /* Apply working directory here, because the working directory might be on NFS and only the user running
5720 * this service might have the correct privilege to change to the working directory */
5721 r
= apply_working_directory(context
, params
, runtime
, home
, exit_status
);
5723 return log_unit_error_errno(unit
, r
, "Changing to the requested working directory failed: %m");
5725 if (needs_sandboxing
) {
5726 /* Apply other MAC contexts late, but before seccomp syscall filtering, as those should really be last to
5727 * influence our own codepaths as little as possible. Moreover, applying MAC contexts usually requires
5728 * syscalls that are subject to seccomp filtering, hence should probably be applied before the syscalls
5729 * are restricted. */
5733 char *exec_context
= mac_selinux_context_net
?: context
->selinux_context
;
5736 r
= setexeccon(exec_context
);
5738 if (!context
->selinux_context_ignore
) {
5739 *exit_status
= EXIT_SELINUX_CONTEXT
;
5740 return log_unit_error_errno(unit
, r
, "Failed to change SELinux context to %s: %m", exec_context
);
5742 log_unit_debug_errno(unit
, r
, "Failed to change SELinux context to %s, ignoring: %m", exec_context
);
5749 if (use_apparmor
&& context
->apparmor_profile
) {
5750 r
= aa_change_onexec(context
->apparmor_profile
);
5751 if (r
< 0 && !context
->apparmor_profile_ignore
) {
5752 *exit_status
= EXIT_APPARMOR_PROFILE
;
5753 return log_unit_error_errno(unit
, errno
, "Failed to prepare AppArmor profile change to %s: %m", context
->apparmor_profile
);
5758 /* PR_GET_SECUREBITS is not privileged, while PR_SET_SECUREBITS is. So to suppress potential
5759 * EPERMs we'll try not to call PR_SET_SECUREBITS unless necessary. Setting securebits
5760 * requires CAP_SETPCAP. */
5761 if (prctl(PR_GET_SECUREBITS
) != secure_bits
) {
5762 /* CAP_SETPCAP is required to set securebits. This capability is raised into the
5763 * effective set here.
5765 * The effective set is overwritten during execve() with the following values:
5767 * - ambient set (for non-root processes)
5769 * - (inheritable | bounding) set for root processes)
5771 * Hence there is no security impact to raise it in the effective set before execve
5773 r
= capability_gain_cap_setpcap(/* return_caps= */ NULL
);
5775 *exit_status
= EXIT_CAPABILITIES
;
5776 return log_unit_error_errno(unit
, r
, "Failed to gain CAP_SETPCAP for setting secure bits");
5778 if (prctl(PR_SET_SECUREBITS
, secure_bits
) < 0) {
5779 *exit_status
= EXIT_SECUREBITS
;
5780 return log_unit_error_errno(unit
, errno
, "Failed to set process secure bits: %m");
5784 if (context_has_no_new_privileges(context
))
5785 if (prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0) < 0) {
5786 *exit_status
= EXIT_NO_NEW_PRIVILEGES
;
5787 return log_unit_error_errno(unit
, errno
, "Failed to disable new privileges: %m");
5791 r
= apply_address_families(unit
, context
);
5793 *exit_status
= EXIT_ADDRESS_FAMILIES
;
5794 return log_unit_error_errno(unit
, r
, "Failed to restrict address families: %m");
5797 r
= apply_memory_deny_write_execute(unit
, context
);
5799 *exit_status
= EXIT_SECCOMP
;
5800 return log_unit_error_errno(unit
, r
, "Failed to disable writing to executable memory: %m");
5803 r
= apply_restrict_realtime(unit
, context
);
5805 *exit_status
= EXIT_SECCOMP
;
5806 return log_unit_error_errno(unit
, r
, "Failed to apply realtime restrictions: %m");
5809 r
= apply_restrict_suid_sgid(unit
, context
);
5811 *exit_status
= EXIT_SECCOMP
;
5812 return log_unit_error_errno(unit
, r
, "Failed to apply SUID/SGID restrictions: %m");
5815 r
= apply_restrict_namespaces(unit
, context
);
5817 *exit_status
= EXIT_SECCOMP
;
5818 return log_unit_error_errno(unit
, r
, "Failed to apply namespace restrictions: %m");
5821 r
= apply_protect_sysctl(unit
, context
);
5823 *exit_status
= EXIT_SECCOMP
;
5824 return log_unit_error_errno(unit
, r
, "Failed to apply sysctl restrictions: %m");
5827 r
= apply_protect_kernel_modules(unit
, context
);
5829 *exit_status
= EXIT_SECCOMP
;
5830 return log_unit_error_errno(unit
, r
, "Failed to apply module loading restrictions: %m");
5833 r
= apply_protect_kernel_logs(unit
, context
);
5835 *exit_status
= EXIT_SECCOMP
;
5836 return log_unit_error_errno(unit
, r
, "Failed to apply kernel log restrictions: %m");
5839 r
= apply_protect_clock(unit
, context
);
5841 *exit_status
= EXIT_SECCOMP
;
5842 return log_unit_error_errno(unit
, r
, "Failed to apply clock restrictions: %m");
5845 r
= apply_private_devices(unit
, context
);
5847 *exit_status
= EXIT_SECCOMP
;
5848 return log_unit_error_errno(unit
, r
, "Failed to set up private devices: %m");
5851 r
= apply_syscall_archs(unit
, context
);
5853 *exit_status
= EXIT_SECCOMP
;
5854 return log_unit_error_errno(unit
, r
, "Failed to apply syscall architecture restrictions: %m");
5857 r
= apply_lock_personality(unit
, context
);
5859 *exit_status
= EXIT_SECCOMP
;
5860 return log_unit_error_errno(unit
, r
, "Failed to lock personalities: %m");
5863 r
= apply_syscall_log(unit
, context
);
5865 *exit_status
= EXIT_SECCOMP
;
5866 return log_unit_error_errno(unit
, r
, "Failed to apply system call log filters: %m");
5869 /* This really should remain the last step before the execve(), to make sure our own code is unaffected
5870 * by the filter as little as possible. */
5871 r
= apply_syscall_filter(unit
, context
, needs_ambient_hack
);
5873 *exit_status
= EXIT_SECCOMP
;
5874 return log_unit_error_errno(unit
, r
, "Failed to apply system call filters: %m");
5879 r
= apply_restrict_filesystems(unit
, context
);
5881 *exit_status
= EXIT_BPF
;
5882 return log_unit_error_errno(unit
, r
, "Failed to restrict filesystems: %m");
5888 if (!strv_isempty(context
->unset_environment
)) {
5891 ee
= strv_env_delete(accum_env
, 1, context
->unset_environment
);
5893 *exit_status
= EXIT_MEMORY
;
5897 strv_free_and_replace(accum_env
, ee
);
5900 if (!FLAGS_SET(command
->flags
, EXEC_COMMAND_NO_ENV_EXPAND
)) {
5901 _cleanup_strv_free_
char **unset_variables
= NULL
, **bad_variables
= NULL
;
5903 r
= replace_env_argv(command
->argv
, accum_env
, &replaced_argv
, &unset_variables
, &bad_variables
);
5905 *exit_status
= EXIT_MEMORY
;
5906 return log_unit_error_errno(unit
, r
, "Failed to replace environment variables: %m");
5908 final_argv
= replaced_argv
;
5910 if (!strv_isempty(unset_variables
)) {
5911 _cleanup_free_
char *ju
= strv_join(unset_variables
, ", ");
5912 log_unit_warning(unit
, "Referenced but unset environment variable evaluates to an empty string: %s", strna(ju
));
5915 if (!strv_isempty(bad_variables
)) {
5916 _cleanup_free_
char *jb
= strv_join(bad_variables
, ", ");
5917 log_unit_warning(unit
, "Invalid environment variable name evaluates to an empty string: %s", strna(jb
));;
5920 final_argv
= command
->argv
;
5922 log_command_line(unit
, "Executing", executable
, final_argv
);
5927 /* We have finished with all our initializations. Let's now let the manager know that. From this point
5928 * on, if the manager sees POLLHUP on the exec_fd, then execve() was successful. */
5930 if (write(exec_fd
, &hot
, sizeof(hot
)) < 0) {
5931 *exit_status
= EXIT_EXEC
;
5932 return log_unit_error_errno(unit
, errno
, "Failed to enable exec_fd: %m");
5936 r
= fexecve_or_execve(executable_fd
, executable
, final_argv
, accum_env
);
5941 /* The execve() failed. This means the exec_fd is still open. Which means we need to tell the manager
5942 * that POLLHUP on it no longer means execve() succeeded. */
5944 if (write(exec_fd
, &hot
, sizeof(hot
)) < 0) {
5945 *exit_status
= EXIT_EXEC
;
5946 return log_unit_error_errno(unit
, errno
, "Failed to disable exec_fd: %m");
5950 *exit_status
= EXIT_EXEC
;
5951 return log_unit_error_errno(unit
, r
, "Failed to execute %s: %m", executable
);
5954 static int exec_context_load_environment(const Unit
*unit
, const ExecContext
*c
, char ***l
);
5955 static int exec_context_named_iofds(const ExecContext
*c
, const ExecParameters
*p
, int named_iofds
[static 3]);
5957 int exec_spawn(Unit
*unit
,
5958 ExecCommand
*command
,
5959 const ExecContext
*context
,
5960 const ExecParameters
*params
,
5961 ExecRuntime
*runtime
,
5962 const CGroupContext
*cgroup_context
,
5965 int socket_fd
, r
, named_iofds
[3] = { -1, -1, -1 }, *fds
= NULL
;
5966 _cleanup_free_
char *subcgroup_path
= NULL
;
5967 _cleanup_strv_free_
char **files_env
= NULL
;
5968 size_t n_storage_fds
= 0, n_socket_fds
= 0;
5976 assert(params
->fds
|| (params
->n_socket_fds
+ params
->n_storage_fds
<= 0));
5978 LOG_CONTEXT_PUSH_UNIT(unit
);
5980 if (context
->std_input
== EXEC_INPUT_SOCKET
||
5981 context
->std_output
== EXEC_OUTPUT_SOCKET
||
5982 context
->std_error
== EXEC_OUTPUT_SOCKET
) {
5984 if (params
->n_socket_fds
> 1)
5985 return log_unit_error_errno(unit
, SYNTHETIC_ERRNO(EINVAL
), "Got more than one socket.");
5987 if (params
->n_socket_fds
== 0)
5988 return log_unit_error_errno(unit
, SYNTHETIC_ERRNO(EINVAL
), "Got no socket.");
5990 socket_fd
= params
->fds
[0];
5994 n_socket_fds
= params
->n_socket_fds
;
5995 n_storage_fds
= params
->n_storage_fds
;
5998 r
= exec_context_named_iofds(context
, params
, named_iofds
);
6000 return log_unit_error_errno(unit
, r
, "Failed to load a named file descriptor: %m");
6002 r
= exec_context_load_environment(unit
, context
, &files_env
);
6004 return log_unit_error_errno(unit
, r
, "Failed to load environment files: %m");
6006 /* Fork with up-to-date SELinux label database, so the child inherits the up-to-date db
6007 and, until the next SELinux policy changes, we save further reloads in future children. */
6008 mac_selinux_maybe_reload();
6010 /* We won't know the real executable path until we create the mount namespace in the child, but we
6011 want to log from the parent, so we use the possibly inaccurate path here. */
6012 log_command_line(unit
, "About to execute", command
->path
, command
->argv
);
6014 if (params
->cgroup_path
) {
6015 r
= exec_parameters_get_cgroup_path(params
, cgroup_context
, &subcgroup_path
);
6017 return log_unit_error_errno(unit
, r
, "Failed to acquire subcgroup path: %m");
6019 /* If there's a subcgroup, then let's create it here now (the main cgroup was already
6020 * realized by the unit logic) */
6022 r
= cg_create(SYSTEMD_CGROUP_CONTROLLER
, subcgroup_path
);
6024 return log_unit_error_errno(unit
, r
, "Failed to create subcgroup '%s': %m", subcgroup_path
);
6030 return log_unit_error_errno(unit
, errno
, "Failed to fork: %m");
6033 int exit_status
= EXIT_SUCCESS
;
6035 r
= exec_child(unit
,
6047 unit
->manager
->user_lookup_fds
[1],
6051 const char *status
=
6052 exit_status_to_string(exit_status
,
6053 EXIT_STATUS_LIBC
| EXIT_STATUS_SYSTEMD
);
6055 log_unit_struct_errno(unit
, LOG_ERR
, r
,
6056 "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR
,
6057 LOG_UNIT_INVOCATION_ID(unit
),
6058 LOG_UNIT_MESSAGE(unit
, "Failed at step %s spawning %s: %m",
6059 status
, command
->path
),
6060 "EXECUTABLE=%s", command
->path
);
6066 log_unit_debug(unit
, "Forked %s as "PID_FMT
, command
->path
, pid
);
6068 /* We add the new process to the cgroup both in the child (so that we can be sure that no user code is ever
6069 * executed outside of the cgroup) and in the parent (so that we can be sure that when we kill the cgroup the
6070 * process will be killed too). */
6072 (void) cg_attach(SYSTEMD_CGROUP_CONTROLLER
, subcgroup_path
, pid
);
6074 exec_status_start(&command
->exec_status
, pid
);
6080 void exec_context_init(ExecContext
*c
) {
6084 c
->ioprio
= IOPRIO_DEFAULT_CLASS_AND_PRIO
;
6085 c
->cpu_sched_policy
= SCHED_OTHER
;
6086 c
->syslog_priority
= LOG_DAEMON
|LOG_INFO
;
6087 c
->syslog_level_prefix
= true;
6088 c
->ignore_sigpipe
= true;
6089 c
->timer_slack_nsec
= NSEC_INFINITY
;
6090 c
->personality
= PERSONALITY_INVALID
;
6091 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++)
6092 c
->directories
[t
].mode
= 0755;
6093 c
->timeout_clean_usec
= USEC_INFINITY
;
6094 c
->capability_bounding_set
= CAP_MASK_UNSET
;
6095 assert_cc(NAMESPACE_FLAGS_INITIAL
!= NAMESPACE_FLAGS_ALL
);
6096 c
->restrict_namespaces
= NAMESPACE_FLAGS_INITIAL
;
6097 c
->log_level_max
= -1;
6099 c
->syscall_errno
= SECCOMP_ERROR_NUMBER_KILL
;
6101 c
->tty_rows
= UINT_MAX
;
6102 c
->tty_cols
= UINT_MAX
;
6103 numa_policy_reset(&c
->numa_policy
);
6104 c
->private_mounts
= -1;
6108 void exec_context_done(ExecContext
*c
) {
6111 c
->environment
= strv_free(c
->environment
);
6112 c
->environment_files
= strv_free(c
->environment_files
);
6113 c
->pass_environment
= strv_free(c
->pass_environment
);
6114 c
->unset_environment
= strv_free(c
->unset_environment
);
6116 rlimit_free_all(c
->rlimit
);
6118 for (size_t l
= 0; l
< 3; l
++) {
6119 c
->stdio_fdname
[l
] = mfree(c
->stdio_fdname
[l
]);
6120 c
->stdio_file
[l
] = mfree(c
->stdio_file
[l
]);
6123 c
->working_directory
= mfree(c
->working_directory
);
6124 c
->root_directory
= mfree(c
->root_directory
);
6125 c
->root_image
= mfree(c
->root_image
);
6126 c
->root_image_options
= mount_options_free_all(c
->root_image_options
);
6127 c
->root_hash
= mfree(c
->root_hash
);
6128 c
->root_hash_size
= 0;
6129 c
->root_hash_path
= mfree(c
->root_hash_path
);
6130 c
->root_hash_sig
= mfree(c
->root_hash_sig
);
6131 c
->root_hash_sig_size
= 0;
6132 c
->root_hash_sig_path
= mfree(c
->root_hash_sig_path
);
6133 c
->root_verity
= mfree(c
->root_verity
);
6134 c
->extension_images
= mount_image_free_many(c
->extension_images
, &c
->n_extension_images
);
6135 c
->extension_directories
= strv_free(c
->extension_directories
);
6136 c
->tty_path
= mfree(c
->tty_path
);
6137 c
->syslog_identifier
= mfree(c
->syslog_identifier
);
6138 c
->user
= mfree(c
->user
);
6139 c
->group
= mfree(c
->group
);
6141 c
->supplementary_groups
= strv_free(c
->supplementary_groups
);
6143 c
->pam_name
= mfree(c
->pam_name
);
6145 c
->read_only_paths
= strv_free(c
->read_only_paths
);
6146 c
->read_write_paths
= strv_free(c
->read_write_paths
);
6147 c
->inaccessible_paths
= strv_free(c
->inaccessible_paths
);
6148 c
->exec_paths
= strv_free(c
->exec_paths
);
6149 c
->no_exec_paths
= strv_free(c
->no_exec_paths
);
6150 c
->exec_search_path
= strv_free(c
->exec_search_path
);
6152 bind_mount_free_many(c
->bind_mounts
, c
->n_bind_mounts
);
6153 c
->bind_mounts
= NULL
;
6154 c
->n_bind_mounts
= 0;
6155 temporary_filesystem_free_many(c
->temporary_filesystems
, c
->n_temporary_filesystems
);
6156 c
->temporary_filesystems
= NULL
;
6157 c
->n_temporary_filesystems
= 0;
6158 c
->mount_images
= mount_image_free_many(c
->mount_images
, &c
->n_mount_images
);
6160 cpu_set_reset(&c
->cpu_set
);
6161 numa_policy_reset(&c
->numa_policy
);
6163 c
->utmp_id
= mfree(c
->utmp_id
);
6164 c
->selinux_context
= mfree(c
->selinux_context
);
6165 c
->apparmor_profile
= mfree(c
->apparmor_profile
);
6166 c
->smack_process_label
= mfree(c
->smack_process_label
);
6168 c
->restrict_filesystems
= set_free(c
->restrict_filesystems
);
6170 c
->syscall_filter
= hashmap_free(c
->syscall_filter
);
6171 c
->syscall_archs
= set_free(c
->syscall_archs
);
6172 c
->address_families
= set_free(c
->address_families
);
6174 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++)
6175 exec_directory_done(&c
->directories
[t
]);
6177 c
->log_level_max
= -1;
6179 exec_context_free_log_extra_fields(c
);
6180 c
->log_filter_allowed_patterns
= set_free(c
->log_filter_allowed_patterns
);
6181 c
->log_filter_denied_patterns
= set_free(c
->log_filter_denied_patterns
);
6183 c
->log_ratelimit_interval_usec
= 0;
6184 c
->log_ratelimit_burst
= 0;
6186 c
->stdin_data
= mfree(c
->stdin_data
);
6187 c
->stdin_data_size
= 0;
6189 c
->network_namespace_path
= mfree(c
->network_namespace_path
);
6190 c
->ipc_namespace_path
= mfree(c
->ipc_namespace_path
);
6192 c
->log_namespace
= mfree(c
->log_namespace
);
6194 c
->load_credentials
= hashmap_free(c
->load_credentials
);
6195 c
->set_credentials
= hashmap_free(c
->set_credentials
);
6196 c
->import_credentials
= set_free(c
->import_credentials
);
6198 c
->root_image_policy
= image_policy_free(c
->root_image_policy
);
6199 c
->mount_image_policy
= image_policy_free(c
->mount_image_policy
);
6200 c
->extension_image_policy
= image_policy_free(c
->extension_image_policy
);
6203 int exec_context_destroy_runtime_directory(const ExecContext
*c
, const char *runtime_prefix
) {
6206 if (!runtime_prefix
)
6209 for (size_t i
= 0; i
< c
->directories
[EXEC_DIRECTORY_RUNTIME
].n_items
; i
++) {
6210 _cleanup_free_
char *p
= NULL
;
6212 if (exec_directory_is_private(c
, EXEC_DIRECTORY_RUNTIME
))
6213 p
= path_join(runtime_prefix
, "private", c
->directories
[EXEC_DIRECTORY_RUNTIME
].items
[i
].path
);
6215 p
= path_join(runtime_prefix
, c
->directories
[EXEC_DIRECTORY_RUNTIME
].items
[i
].path
);
6219 /* We execute this synchronously, since we need to be sure this is gone when we start the
6221 (void) rm_rf(p
, REMOVE_ROOT
);
6223 STRV_FOREACH(symlink
, c
->directories
[EXEC_DIRECTORY_RUNTIME
].items
[i
].symlinks
) {
6224 _cleanup_free_
char *symlink_abs
= NULL
;
6226 if (exec_directory_is_private(c
, EXEC_DIRECTORY_RUNTIME
))
6227 symlink_abs
= path_join(runtime_prefix
, "private", *symlink
);
6229 symlink_abs
= path_join(runtime_prefix
, *symlink
);
6233 (void) unlink(symlink_abs
);
6240 int exec_context_destroy_credentials(const ExecContext
*c
, const char *runtime_prefix
, const char *unit
) {
6241 _cleanup_free_
char *p
= NULL
;
6245 if (!runtime_prefix
|| !unit
)
6248 p
= path_join(runtime_prefix
, "credentials", unit
);
6252 /* This is either a tmpfs/ramfs of its own, or a plain directory. Either way, let's first try to
6253 * unmount it, and afterwards remove the mount point */
6254 (void) umount2(p
, MNT_DETACH
|UMOUNT_NOFOLLOW
);
6255 (void) rm_rf(p
, REMOVE_ROOT
|REMOVE_CHMOD
);
6260 int exec_context_destroy_mount_ns_dir(Unit
*u
) {
6261 _cleanup_free_
char *p
= NULL
;
6263 if (!u
|| !MANAGER_IS_SYSTEM(u
->manager
))
6266 p
= path_join("/run/systemd/propagate/", u
->id
);
6270 /* This is only filled transiently (see mount_in_namespace()), should be empty or even non-existent*/
6271 if (rmdir(p
) < 0 && errno
!= ENOENT
)
6272 log_unit_debug_errno(u
, errno
, "Unable to remove propagation dir '%s', ignoring: %m", p
);
6277 static void exec_command_done(ExecCommand
*c
) {
6280 c
->path
= mfree(c
->path
);
6281 c
->argv
= strv_free(c
->argv
);
6284 void exec_command_done_array(ExecCommand
*c
, size_t n
) {
6285 for (size_t i
= 0; i
< n
; i
++)
6286 exec_command_done(c
+i
);
6289 ExecCommand
* exec_command_free_list(ExecCommand
*c
) {
6293 LIST_REMOVE(command
, c
, i
);
6294 exec_command_done(i
);
6301 void exec_command_free_array(ExecCommand
**c
, size_t n
) {
6302 for (size_t i
= 0; i
< n
; i
++)
6303 c
[i
] = exec_command_free_list(c
[i
]);
6306 void exec_command_reset_status_array(ExecCommand
*c
, size_t n
) {
6307 for (size_t i
= 0; i
< n
; i
++)
6308 exec_status_reset(&c
[i
].exec_status
);
6311 void exec_command_reset_status_list_array(ExecCommand
**c
, size_t n
) {
6312 for (size_t i
= 0; i
< n
; i
++)
6313 LIST_FOREACH(command
, z
, c
[i
])
6314 exec_status_reset(&z
->exec_status
);
6317 typedef struct InvalidEnvInfo
{
6322 static void invalid_env(const char *p
, void *userdata
) {
6323 InvalidEnvInfo
*info
= userdata
;
6325 log_unit_error(info
->unit
, "Ignoring invalid environment assignment '%s': %s", p
, info
->path
);
6328 const char* exec_context_fdname(const ExecContext
*c
, int fd_index
) {
6334 if (c
->std_input
!= EXEC_INPUT_NAMED_FD
)
6337 return c
->stdio_fdname
[STDIN_FILENO
] ?: "stdin";
6340 if (c
->std_output
!= EXEC_OUTPUT_NAMED_FD
)
6343 return c
->stdio_fdname
[STDOUT_FILENO
] ?: "stdout";
6346 if (c
->std_error
!= EXEC_OUTPUT_NAMED_FD
)
6349 return c
->stdio_fdname
[STDERR_FILENO
] ?: "stderr";
6356 static int exec_context_named_iofds(
6357 const ExecContext
*c
,
6358 const ExecParameters
*p
,
6359 int named_iofds
[static 3]) {
6362 const char* stdio_fdname
[3];
6367 assert(named_iofds
);
6369 targets
= (c
->std_input
== EXEC_INPUT_NAMED_FD
) +
6370 (c
->std_output
== EXEC_OUTPUT_NAMED_FD
) +
6371 (c
->std_error
== EXEC_OUTPUT_NAMED_FD
);
6373 for (size_t i
= 0; i
< 3; i
++)
6374 stdio_fdname
[i
] = exec_context_fdname(c
, i
);
6376 n_fds
= p
->n_storage_fds
+ p
->n_socket_fds
;
6378 for (size_t i
= 0; i
< n_fds
&& targets
> 0; i
++)
6379 if (named_iofds
[STDIN_FILENO
] < 0 &&
6380 c
->std_input
== EXEC_INPUT_NAMED_FD
&&
6381 stdio_fdname
[STDIN_FILENO
] &&
6382 streq(p
->fd_names
[i
], stdio_fdname
[STDIN_FILENO
])) {
6384 named_iofds
[STDIN_FILENO
] = p
->fds
[i
];
6387 } else if (named_iofds
[STDOUT_FILENO
] < 0 &&
6388 c
->std_output
== EXEC_OUTPUT_NAMED_FD
&&
6389 stdio_fdname
[STDOUT_FILENO
] &&
6390 streq(p
->fd_names
[i
], stdio_fdname
[STDOUT_FILENO
])) {
6392 named_iofds
[STDOUT_FILENO
] = p
->fds
[i
];
6395 } else if (named_iofds
[STDERR_FILENO
] < 0 &&
6396 c
->std_error
== EXEC_OUTPUT_NAMED_FD
&&
6397 stdio_fdname
[STDERR_FILENO
] &&
6398 streq(p
->fd_names
[i
], stdio_fdname
[STDERR_FILENO
])) {
6400 named_iofds
[STDERR_FILENO
] = p
->fds
[i
];
6404 return targets
== 0 ? 0 : -ENOENT
;
6407 static int exec_context_load_environment(const Unit
*unit
, const ExecContext
*c
, char ***ret
) {
6408 _cleanup_strv_free_
char **v
= NULL
;
6414 STRV_FOREACH(i
, c
->environment_files
) {
6415 _cleanup_globfree_ glob_t pglob
= {};
6416 bool ignore
= false;
6424 if (!path_is_absolute(fn
)) {
6430 /* Filename supports globbing, take all matching files */
6431 r
= safe_glob(fn
, 0, &pglob
);
6438 /* When we don't match anything, -ENOENT should be returned */
6439 assert(pglob
.gl_pathc
> 0);
6441 for (size_t n
= 0; n
< pglob
.gl_pathc
; n
++) {
6442 _cleanup_strv_free_
char **p
= NULL
;
6444 r
= load_env_file(NULL
, pglob
.gl_pathv
[n
], &p
);
6451 /* Log invalid environment variables with filename */
6453 InvalidEnvInfo info
= {
6455 .path
= pglob
.gl_pathv
[n
]
6458 p
= strv_env_clean_with_callback(p
, invalid_env
, &info
);
6464 char **m
= strv_env_merge(v
, p
);
6468 strv_free_and_replace(v
, m
);
6478 static bool tty_may_match_dev_console(const char *tty
) {
6479 _cleanup_free_
char *resolved
= NULL
;
6484 tty
= skip_dev_prefix(tty
);
6486 /* trivial identity? */
6487 if (streq(tty
, "console"))
6490 if (resolve_dev_console(&resolved
) < 0)
6491 return true; /* if we could not resolve, assume it may */
6493 /* "tty0" means the active VC, so it may be the same sometimes */
6494 return path_equal(resolved
, tty
) || (streq(resolved
, "tty0") && tty_is_vc(tty
));
6497 static bool exec_context_may_touch_tty(const ExecContext
*ec
) {
6500 return ec
->tty_reset
||
6502 ec
->tty_vt_disallocate
||
6503 is_terminal_input(ec
->std_input
) ||
6504 is_terminal_output(ec
->std_output
) ||
6505 is_terminal_output(ec
->std_error
);
6508 bool exec_context_may_touch_console(const ExecContext
*ec
) {
6510 return exec_context_may_touch_tty(ec
) &&
6511 tty_may_match_dev_console(exec_context_tty_path(ec
));
6514 static void strv_fprintf(FILE *f
, char **l
) {
6518 fprintf(f
, " %s", *g
);
6521 static void strv_dump(FILE* f
, const char *prefix
, const char *name
, char **strv
) {
6526 if (!strv_isempty(strv
)) {
6527 fprintf(f
, "%s%s:", prefix
, name
);
6528 strv_fprintf(f
, strv
);
6533 void exec_context_dump(const ExecContext
*c
, FILE* f
, const char *prefix
) {
6539 prefix
= strempty(prefix
);
6543 "%sWorkingDirectory: %s\n"
6544 "%sRootDirectory: %s\n"
6545 "%sRootEphemeral: %s\n"
6546 "%sNonBlocking: %s\n"
6547 "%sPrivateTmp: %s\n"
6548 "%sPrivateDevices: %s\n"
6549 "%sProtectKernelTunables: %s\n"
6550 "%sProtectKernelModules: %s\n"
6551 "%sProtectKernelLogs: %s\n"
6552 "%sProtectClock: %s\n"
6553 "%sProtectControlGroups: %s\n"
6554 "%sPrivateNetwork: %s\n"
6555 "%sPrivateUsers: %s\n"
6556 "%sProtectHome: %s\n"
6557 "%sProtectSystem: %s\n"
6558 "%sMountAPIVFS: %s\n"
6559 "%sIgnoreSIGPIPE: %s\n"
6560 "%sMemoryDenyWriteExecute: %s\n"
6561 "%sRestrictRealtime: %s\n"
6562 "%sRestrictSUIDSGID: %s\n"
6563 "%sKeyringMode: %s\n"
6564 "%sProtectHostname: %s\n"
6565 "%sProtectProc: %s\n"
6566 "%sProcSubset: %s\n",
6568 prefix
, empty_to_root(c
->working_directory
),
6569 prefix
, empty_to_root(c
->root_directory
),
6570 prefix
, yes_no(c
->root_ephemeral
),
6571 prefix
, yes_no(c
->non_blocking
),
6572 prefix
, yes_no(c
->private_tmp
),
6573 prefix
, yes_no(c
->private_devices
),
6574 prefix
, yes_no(c
->protect_kernel_tunables
),
6575 prefix
, yes_no(c
->protect_kernel_modules
),
6576 prefix
, yes_no(c
->protect_kernel_logs
),
6577 prefix
, yes_no(c
->protect_clock
),
6578 prefix
, yes_no(c
->protect_control_groups
),
6579 prefix
, yes_no(c
->private_network
),
6580 prefix
, yes_no(c
->private_users
),
6581 prefix
, protect_home_to_string(c
->protect_home
),
6582 prefix
, protect_system_to_string(c
->protect_system
),
6583 prefix
, yes_no(exec_context_get_effective_mount_apivfs(c
)),
6584 prefix
, yes_no(c
->ignore_sigpipe
),
6585 prefix
, yes_no(c
->memory_deny_write_execute
),
6586 prefix
, yes_no(c
->restrict_realtime
),
6587 prefix
, yes_no(c
->restrict_suid_sgid
),
6588 prefix
, exec_keyring_mode_to_string(c
->keyring_mode
),
6589 prefix
, yes_no(c
->protect_hostname
),
6590 prefix
, protect_proc_to_string(c
->protect_proc
),
6591 prefix
, proc_subset_to_string(c
->proc_subset
));
6594 fprintf(f
, "%sRootImage: %s\n", prefix
, c
->root_image
);
6596 if (c
->root_image_options
) {
6597 fprintf(f
, "%sRootImageOptions:", prefix
);
6598 LIST_FOREACH(mount_options
, o
, c
->root_image_options
)
6599 if (!isempty(o
->options
))
6600 fprintf(f
, " %s:%s",
6601 partition_designator_to_string(o
->partition_designator
),
6607 _cleanup_free_
char *encoded
= NULL
;
6608 encoded
= hexmem(c
->root_hash
, c
->root_hash_size
);
6610 fprintf(f
, "%sRootHash: %s\n", prefix
, encoded
);
6613 if (c
->root_hash_path
)
6614 fprintf(f
, "%sRootHash: %s\n", prefix
, c
->root_hash_path
);
6616 if (c
->root_hash_sig
) {
6617 _cleanup_free_
char *encoded
= NULL
;
6619 len
= base64mem(c
->root_hash_sig
, c
->root_hash_sig_size
, &encoded
);
6621 fprintf(f
, "%sRootHashSignature: base64:%s\n", prefix
, encoded
);
6624 if (c
->root_hash_sig_path
)
6625 fprintf(f
, "%sRootHashSignature: %s\n", prefix
, c
->root_hash_sig_path
);
6628 fprintf(f
, "%sRootVerity: %s\n", prefix
, c
->root_verity
);
6630 STRV_FOREACH(e
, c
->environment
)
6631 fprintf(f
, "%sEnvironment: %s\n", prefix
, *e
);
6633 STRV_FOREACH(e
, c
->environment_files
)
6634 fprintf(f
, "%sEnvironmentFile: %s\n", prefix
, *e
);
6636 STRV_FOREACH(e
, c
->pass_environment
)
6637 fprintf(f
, "%sPassEnvironment: %s\n", prefix
, *e
);
6639 STRV_FOREACH(e
, c
->unset_environment
)
6640 fprintf(f
, "%sUnsetEnvironment: %s\n", prefix
, *e
);
6642 fprintf(f
, "%sRuntimeDirectoryPreserve: %s\n", prefix
, exec_preserve_mode_to_string(c
->runtime_directory_preserve_mode
));
6644 for (ExecDirectoryType dt
= 0; dt
< _EXEC_DIRECTORY_TYPE_MAX
; dt
++) {
6645 fprintf(f
, "%s%sMode: %04o\n", prefix
, exec_directory_type_to_string(dt
), c
->directories
[dt
].mode
);
6647 for (size_t i
= 0; i
< c
->directories
[dt
].n_items
; i
++) {
6648 fprintf(f
, "%s%s: %s\n", prefix
, exec_directory_type_to_string(dt
), c
->directories
[dt
].items
[i
].path
);
6650 STRV_FOREACH(d
, c
->directories
[dt
].items
[i
].symlinks
)
6651 fprintf(f
, "%s%s: %s:%s\n", prefix
, exec_directory_type_symlink_to_string(dt
), c
->directories
[dt
].items
[i
].path
, *d
);
6655 fprintf(f
, "%sTimeoutCleanSec: %s\n", prefix
, FORMAT_TIMESPAN(c
->timeout_clean_usec
, USEC_PER_SEC
));
6658 fprintf(f
, "%sNice: %i\n", prefix
, c
->nice
);
6660 if (c
->oom_score_adjust_set
)
6661 fprintf(f
, "%sOOMScoreAdjust: %i\n", prefix
, c
->oom_score_adjust
);
6663 if (c
->coredump_filter_set
)
6664 fprintf(f
, "%sCoredumpFilter: 0x%"PRIx64
"\n", prefix
, c
->coredump_filter
);
6666 for (unsigned i
= 0; i
< RLIM_NLIMITS
; i
++)
6668 fprintf(f
, "%sLimit%s: " RLIM_FMT
"\n",
6669 prefix
, rlimit_to_string(i
), c
->rlimit
[i
]->rlim_max
);
6670 fprintf(f
, "%sLimit%sSoft: " RLIM_FMT
"\n",
6671 prefix
, rlimit_to_string(i
), c
->rlimit
[i
]->rlim_cur
);
6674 if (c
->ioprio_set
) {
6675 _cleanup_free_
char *class_str
= NULL
;
6677 r
= ioprio_class_to_string_alloc(ioprio_prio_class(c
->ioprio
), &class_str
);
6679 fprintf(f
, "%sIOSchedulingClass: %s\n", prefix
, class_str
);
6681 fprintf(f
, "%sIOPriority: %d\n", prefix
, ioprio_prio_data(c
->ioprio
));
6684 if (c
->cpu_sched_set
) {
6685 _cleanup_free_
char *policy_str
= NULL
;
6687 r
= sched_policy_to_string_alloc(c
->cpu_sched_policy
, &policy_str
);
6689 fprintf(f
, "%sCPUSchedulingPolicy: %s\n", prefix
, policy_str
);
6692 "%sCPUSchedulingPriority: %i\n"
6693 "%sCPUSchedulingResetOnFork: %s\n",
6694 prefix
, c
->cpu_sched_priority
,
6695 prefix
, yes_no(c
->cpu_sched_reset_on_fork
));
6698 if (c
->cpu_set
.set
) {
6699 _cleanup_free_
char *affinity
= NULL
;
6701 affinity
= cpu_set_to_range_string(&c
->cpu_set
);
6702 fprintf(f
, "%sCPUAffinity: %s\n", prefix
, affinity
);
6705 if (mpol_is_valid(numa_policy_get_type(&c
->numa_policy
))) {
6706 _cleanup_free_
char *nodes
= NULL
;
6708 nodes
= cpu_set_to_range_string(&c
->numa_policy
.nodes
);
6709 fprintf(f
, "%sNUMAPolicy: %s\n", prefix
, mpol_to_string(numa_policy_get_type(&c
->numa_policy
)));
6710 fprintf(f
, "%sNUMAMask: %s\n", prefix
, strnull(nodes
));
6713 if (c
->timer_slack_nsec
!= NSEC_INFINITY
)
6714 fprintf(f
, "%sTimerSlackNSec: "NSEC_FMT
"\n", prefix
, c
->timer_slack_nsec
);
6717 "%sStandardInput: %s\n"
6718 "%sStandardOutput: %s\n"
6719 "%sStandardError: %s\n",
6720 prefix
, exec_input_to_string(c
->std_input
),
6721 prefix
, exec_output_to_string(c
->std_output
),
6722 prefix
, exec_output_to_string(c
->std_error
));
6724 if (c
->std_input
== EXEC_INPUT_NAMED_FD
)
6725 fprintf(f
, "%sStandardInputFileDescriptorName: %s\n", prefix
, c
->stdio_fdname
[STDIN_FILENO
]);
6726 if (c
->std_output
== EXEC_OUTPUT_NAMED_FD
)
6727 fprintf(f
, "%sStandardOutputFileDescriptorName: %s\n", prefix
, c
->stdio_fdname
[STDOUT_FILENO
]);
6728 if (c
->std_error
== EXEC_OUTPUT_NAMED_FD
)
6729 fprintf(f
, "%sStandardErrorFileDescriptorName: %s\n", prefix
, c
->stdio_fdname
[STDERR_FILENO
]);
6731 if (c
->std_input
== EXEC_INPUT_FILE
)
6732 fprintf(f
, "%sStandardInputFile: %s\n", prefix
, c
->stdio_file
[STDIN_FILENO
]);
6733 if (c
->std_output
== EXEC_OUTPUT_FILE
)
6734 fprintf(f
, "%sStandardOutputFile: %s\n", prefix
, c
->stdio_file
[STDOUT_FILENO
]);
6735 if (c
->std_output
== EXEC_OUTPUT_FILE_APPEND
)
6736 fprintf(f
, "%sStandardOutputFileToAppend: %s\n", prefix
, c
->stdio_file
[STDOUT_FILENO
]);
6737 if (c
->std_output
== EXEC_OUTPUT_FILE_TRUNCATE
)
6738 fprintf(f
, "%sStandardOutputFileToTruncate: %s\n", prefix
, c
->stdio_file
[STDOUT_FILENO
]);
6739 if (c
->std_error
== EXEC_OUTPUT_FILE
)
6740 fprintf(f
, "%sStandardErrorFile: %s\n", prefix
, c
->stdio_file
[STDERR_FILENO
]);
6741 if (c
->std_error
== EXEC_OUTPUT_FILE_APPEND
)
6742 fprintf(f
, "%sStandardErrorFileToAppend: %s\n", prefix
, c
->stdio_file
[STDERR_FILENO
]);
6743 if (c
->std_error
== EXEC_OUTPUT_FILE_TRUNCATE
)
6744 fprintf(f
, "%sStandardErrorFileToTruncate: %s\n", prefix
, c
->stdio_file
[STDERR_FILENO
]);
6750 "%sTTYVHangup: %s\n"
6751 "%sTTYVTDisallocate: %s\n"
6753 "%sTTYColumns: %u\n",
6754 prefix
, c
->tty_path
,
6755 prefix
, yes_no(c
->tty_reset
),
6756 prefix
, yes_no(c
->tty_vhangup
),
6757 prefix
, yes_no(c
->tty_vt_disallocate
),
6758 prefix
, c
->tty_rows
,
6759 prefix
, c
->tty_cols
);
6761 if (IN_SET(c
->std_output
,
6763 EXEC_OUTPUT_JOURNAL
,
6764 EXEC_OUTPUT_KMSG_AND_CONSOLE
,
6765 EXEC_OUTPUT_JOURNAL_AND_CONSOLE
) ||
6766 IN_SET(c
->std_error
,
6768 EXEC_OUTPUT_JOURNAL
,
6769 EXEC_OUTPUT_KMSG_AND_CONSOLE
,
6770 EXEC_OUTPUT_JOURNAL_AND_CONSOLE
)) {
6772 _cleanup_free_
char *fac_str
= NULL
, *lvl_str
= NULL
;
6774 r
= log_facility_unshifted_to_string_alloc(c
->syslog_priority
>> 3, &fac_str
);
6776 fprintf(f
, "%sSyslogFacility: %s\n", prefix
, fac_str
);
6778 r
= log_level_to_string_alloc(LOG_PRI(c
->syslog_priority
), &lvl_str
);
6780 fprintf(f
, "%sSyslogLevel: %s\n", prefix
, lvl_str
);
6783 if (c
->log_level_max
>= 0) {
6784 _cleanup_free_
char *t
= NULL
;
6786 (void) log_level_to_string_alloc(c
->log_level_max
, &t
);
6788 fprintf(f
, "%sLogLevelMax: %s\n", prefix
, strna(t
));
6791 if (c
->log_ratelimit_interval_usec
> 0)
6793 "%sLogRateLimitIntervalSec: %s\n",
6794 prefix
, FORMAT_TIMESPAN(c
->log_ratelimit_interval_usec
, USEC_PER_SEC
));
6796 if (c
->log_ratelimit_burst
> 0)
6797 fprintf(f
, "%sLogRateLimitBurst: %u\n", prefix
, c
->log_ratelimit_burst
);
6799 if (!set_isempty(c
->log_filter_allowed_patterns
) || !set_isempty(c
->log_filter_denied_patterns
)) {
6800 fprintf(f
, "%sLogFilterPatterns:", prefix
);
6803 SET_FOREACH(pattern
, c
->log_filter_allowed_patterns
)
6804 fprintf(f
, " %s", pattern
);
6805 SET_FOREACH(pattern
, c
->log_filter_denied_patterns
)
6806 fprintf(f
, " ~%s", pattern
);
6810 for (size_t j
= 0; j
< c
->n_log_extra_fields
; j
++) {
6811 fprintf(f
, "%sLogExtraFields: ", prefix
);
6812 fwrite(c
->log_extra_fields
[j
].iov_base
,
6813 1, c
->log_extra_fields
[j
].iov_len
,
6818 if (c
->log_namespace
)
6819 fprintf(f
, "%sLogNamespace: %s\n", prefix
, c
->log_namespace
);
6821 if (c
->secure_bits
) {
6822 _cleanup_free_
char *str
= NULL
;
6824 r
= secure_bits_to_string_alloc(c
->secure_bits
, &str
);
6826 fprintf(f
, "%sSecure Bits: %s\n", prefix
, str
);
6829 if (c
->capability_bounding_set
!= CAP_MASK_UNSET
) {
6830 _cleanup_free_
char *str
= NULL
;
6832 r
= capability_set_to_string(c
->capability_bounding_set
, &str
);
6834 fprintf(f
, "%sCapabilityBoundingSet: %s\n", prefix
, str
);
6837 if (c
->capability_ambient_set
!= 0) {
6838 _cleanup_free_
char *str
= NULL
;
6840 r
= capability_set_to_string(c
->capability_ambient_set
, &str
);
6842 fprintf(f
, "%sAmbientCapabilities: %s\n", prefix
, str
);
6846 fprintf(f
, "%sUser: %s\n", prefix
, c
->user
);
6848 fprintf(f
, "%sGroup: %s\n", prefix
, c
->group
);
6850 fprintf(f
, "%sDynamicUser: %s\n", prefix
, yes_no(c
->dynamic_user
));
6852 strv_dump(f
, prefix
, "SupplementaryGroups", c
->supplementary_groups
);
6855 fprintf(f
, "%sPAMName: %s\n", prefix
, c
->pam_name
);
6857 strv_dump(f
, prefix
, "ReadWritePaths", c
->read_write_paths
);
6858 strv_dump(f
, prefix
, "ReadOnlyPaths", c
->read_only_paths
);
6859 strv_dump(f
, prefix
, "InaccessiblePaths", c
->inaccessible_paths
);
6860 strv_dump(f
, prefix
, "ExecPaths", c
->exec_paths
);
6861 strv_dump(f
, prefix
, "NoExecPaths", c
->no_exec_paths
);
6862 strv_dump(f
, prefix
, "ExecSearchPath", c
->exec_search_path
);
6864 for (size_t i
= 0; i
< c
->n_bind_mounts
; i
++)
6865 fprintf(f
, "%s%s: %s%s:%s:%s\n", prefix
,
6866 c
->bind_mounts
[i
].read_only
? "BindReadOnlyPaths" : "BindPaths",
6867 c
->bind_mounts
[i
].ignore_enoent
? "-": "",
6868 c
->bind_mounts
[i
].source
,
6869 c
->bind_mounts
[i
].destination
,
6870 c
->bind_mounts
[i
].recursive
? "rbind" : "norbind");
6872 for (size_t i
= 0; i
< c
->n_temporary_filesystems
; i
++) {
6873 const TemporaryFileSystem
*t
= c
->temporary_filesystems
+ i
;
6875 fprintf(f
, "%sTemporaryFileSystem: %s%s%s\n", prefix
,
6877 isempty(t
->options
) ? "" : ":",
6878 strempty(t
->options
));
6883 "%sUtmpIdentifier: %s\n",
6884 prefix
, c
->utmp_id
);
6886 if (c
->selinux_context
)
6888 "%sSELinuxContext: %s%s\n",
6889 prefix
, c
->selinux_context_ignore
? "-" : "", c
->selinux_context
);
6891 if (c
->apparmor_profile
)
6893 "%sAppArmorProfile: %s%s\n",
6894 prefix
, c
->apparmor_profile_ignore
? "-" : "", c
->apparmor_profile
);
6896 if (c
->smack_process_label
)
6898 "%sSmackProcessLabel: %s%s\n",
6899 prefix
, c
->smack_process_label_ignore
? "-" : "", c
->smack_process_label
);
6901 if (c
->personality
!= PERSONALITY_INVALID
)
6903 "%sPersonality: %s\n",
6904 prefix
, strna(personality_to_string(c
->personality
)));
6907 "%sLockPersonality: %s\n",
6908 prefix
, yes_no(c
->lock_personality
));
6910 if (c
->syscall_filter
) {
6912 "%sSystemCallFilter: ",
6915 if (!c
->syscall_allow_list
)
6921 HASHMAP_FOREACH_KEY(val
, id
, c
->syscall_filter
) {
6922 _cleanup_free_
char *name
= NULL
;
6923 const char *errno_name
= NULL
;
6924 int num
= PTR_TO_INT(val
);
6931 name
= seccomp_syscall_resolve_num_arch(SCMP_ARCH_NATIVE
, PTR_TO_INT(id
) - 1);
6932 fputs(strna(name
), f
);
6935 errno_name
= seccomp_errno_or_action_to_string(num
);
6937 fprintf(f
, ":%s", errno_name
);
6939 fprintf(f
, ":%d", num
);
6947 if (c
->syscall_archs
) {
6949 "%sSystemCallArchitectures:",
6954 SET_FOREACH(id
, c
->syscall_archs
)
6955 fprintf(f
, " %s", strna(seccomp_arch_to_string(PTR_TO_UINT32(id
) - 1)));
6960 if (exec_context_restrict_namespaces_set(c
)) {
6961 _cleanup_free_
char *s
= NULL
;
6963 r
= namespace_flags_to_string(c
->restrict_namespaces
, &s
);
6965 fprintf(f
, "%sRestrictNamespaces: %s\n",
6970 if (exec_context_restrict_filesystems_set(c
)) {
6972 SET_FOREACH(fs
, c
->restrict_filesystems
)
6973 fprintf(f
, "%sRestrictFileSystems: %s\n", prefix
, fs
);
6977 if (c
->network_namespace_path
)
6979 "%sNetworkNamespacePath: %s\n",
6980 prefix
, c
->network_namespace_path
);
6982 if (c
->syscall_errno
> 0) {
6983 fprintf(f
, "%sSystemCallErrorNumber: ", prefix
);
6986 const char *errno_name
= seccomp_errno_or_action_to_string(c
->syscall_errno
);
6988 fputs(errno_name
, f
);
6990 fprintf(f
, "%d", c
->syscall_errno
);
6995 for (size_t i
= 0; i
< c
->n_mount_images
; i
++) {
6996 fprintf(f
, "%sMountImages: %s%s:%s", prefix
,
6997 c
->mount_images
[i
].ignore_enoent
? "-": "",
6998 c
->mount_images
[i
].source
,
6999 c
->mount_images
[i
].destination
);
7000 LIST_FOREACH(mount_options
, o
, c
->mount_images
[i
].mount_options
)
7001 fprintf(f
, ":%s:%s",
7002 partition_designator_to_string(o
->partition_designator
),
7003 strempty(o
->options
));
7007 for (size_t i
= 0; i
< c
->n_extension_images
; i
++) {
7008 fprintf(f
, "%sExtensionImages: %s%s", prefix
,
7009 c
->extension_images
[i
].ignore_enoent
? "-": "",
7010 c
->extension_images
[i
].source
);
7011 LIST_FOREACH(mount_options
, o
, c
->extension_images
[i
].mount_options
)
7012 fprintf(f
, ":%s:%s",
7013 partition_designator_to_string(o
->partition_designator
),
7014 strempty(o
->options
));
7018 strv_dump(f
, prefix
, "ExtensionDirectories", c
->extension_directories
);
7021 bool exec_context_maintains_privileges(const ExecContext
*c
) {
7024 /* Returns true if the process forked off would run under
7025 * an unchanged UID or as root. */
7030 if (streq(c
->user
, "root") || streq(c
->user
, "0"))
7036 int exec_context_get_effective_ioprio(const ExecContext
*c
) {
7044 p
= ioprio_get(IOPRIO_WHO_PROCESS
, 0);
7046 return IOPRIO_DEFAULT_CLASS_AND_PRIO
;
7048 return ioprio_normalize(p
);
7051 bool exec_context_get_effective_mount_apivfs(const ExecContext
*c
) {
7054 /* Explicit setting wins */
7055 if (c
->mount_apivfs_set
)
7056 return c
->mount_apivfs
;
7058 /* Default to "yes" if root directory or image are specified */
7059 if (exec_context_with_rootfs(c
))
7065 void exec_context_free_log_extra_fields(ExecContext
*c
) {
7068 for (size_t l
= 0; l
< c
->n_log_extra_fields
; l
++)
7069 free(c
->log_extra_fields
[l
].iov_base
);
7070 c
->log_extra_fields
= mfree(c
->log_extra_fields
);
7071 c
->n_log_extra_fields
= 0;
7074 void exec_context_revert_tty(ExecContext
*c
) {
7075 _cleanup_close_
int fd
= -EBADF
;
7082 /* First, reset the TTY (possibly kicking everybody else from the TTY) */
7083 exec_context_tty_reset(c
, NULL
);
7085 /* And then undo what chown_terminal() did earlier. Note that we only do this if we have a path
7086 * configured. If the TTY was passed to us as file descriptor we assume the TTY is opened and managed
7087 * by whoever passed it to us and thus knows better when and how to chmod()/chown() it back. */
7088 if (!exec_context_may_touch_tty(c
))
7091 path
= exec_context_tty_path(c
);
7095 fd
= open(path
, O_PATH
|O_CLOEXEC
);
7097 return (void) log_full_errno(errno
== ENOENT
? LOG_DEBUG
: LOG_WARNING
, errno
,
7098 "Failed to open TTY inode of '%s' to adjust ownership/access mode, ignoring: %m",
7101 if (fstat(fd
, &st
) < 0)
7102 return (void) log_warning_errno(errno
, "Failed to stat TTY '%s', ignoring: %m", path
);
7104 /* Let's add a superficial check that we only do this for stuff that looks like a TTY. We only check
7105 * if things are a character device, since a proper check either means we'd have to open the TTY and
7106 * use isatty(), but we'd rather not do that since opening TTYs comes with all kinds of side-effects
7107 * and is slow. Or we'd have to hardcode dev_t major information, which we'd rather avoid. Why bother
7108 * with this at all? → https://github.com/systemd/systemd/issues/19213 */
7109 if (!S_ISCHR(st
.st_mode
))
7110 return log_warning("Configured TTY '%s' is not actually a character device, ignoring.", path
);
7112 r
= fchmod_and_chown(fd
, TTY_MODE
, 0, TTY_GID
);
7114 log_warning_errno(r
, "Failed to reset TTY ownership/access mode of %s, ignoring: %m", path
);
7117 int exec_context_get_clean_directories(
7123 _cleanup_strv_free_
char **l
= NULL
;
7130 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
7131 if (!FLAGS_SET(mask
, 1U << t
))
7137 for (size_t i
= 0; i
< c
->directories
[t
].n_items
; i
++) {
7140 j
= path_join(prefix
[t
], c
->directories
[t
].items
[i
].path
);
7144 r
= strv_consume(&l
, j
);
7148 /* Also remove private directories unconditionally. */
7149 if (t
!= EXEC_DIRECTORY_CONFIGURATION
) {
7150 j
= path_join(prefix
[t
], "private", c
->directories
[t
].items
[i
].path
);
7154 r
= strv_consume(&l
, j
);
7159 STRV_FOREACH(symlink
, c
->directories
[t
].items
[i
].symlinks
) {
7160 j
= path_join(prefix
[t
], *symlink
);
7164 r
= strv_consume(&l
, j
);
7175 int exec_context_get_clean_mask(ExecContext
*c
, ExecCleanMask
*ret
) {
7176 ExecCleanMask mask
= 0;
7181 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++)
7182 if (c
->directories
[t
].n_items
> 0)
7189 bool exec_context_has_encrypted_credentials(ExecContext
*c
) {
7190 ExecLoadCredential
*load_cred
;
7191 ExecSetCredential
*set_cred
;
7195 HASHMAP_FOREACH(load_cred
, c
->load_credentials
)
7196 if (load_cred
->encrypted
)
7199 HASHMAP_FOREACH(set_cred
, c
->set_credentials
)
7200 if (set_cred
->encrypted
)
7206 void exec_status_start(ExecStatus
*s
, pid_t pid
) {
7213 dual_timestamp_get(&s
->start_timestamp
);
7216 void exec_status_exit(ExecStatus
*s
, const ExecContext
*context
, pid_t pid
, int code
, int status
) {
7224 dual_timestamp_get(&s
->exit_timestamp
);
7229 if (context
&& context
->utmp_id
)
7230 (void) utmp_put_dead_process(context
->utmp_id
, pid
, code
, status
);
7233 void exec_status_reset(ExecStatus
*s
) {
7236 *s
= (ExecStatus
) {};
7239 void exec_status_dump(const ExecStatus
*s
, FILE *f
, const char *prefix
) {
7246 prefix
= strempty(prefix
);
7249 "%sPID: "PID_FMT
"\n",
7252 if (dual_timestamp_is_set(&s
->start_timestamp
))
7254 "%sStart Timestamp: %s\n",
7255 prefix
, FORMAT_TIMESTAMP(s
->start_timestamp
.realtime
));
7257 if (dual_timestamp_is_set(&s
->exit_timestamp
))
7259 "%sExit Timestamp: %s\n"
7261 "%sExit Status: %i\n",
7262 prefix
, FORMAT_TIMESTAMP(s
->exit_timestamp
.realtime
),
7263 prefix
, sigchld_code_to_string(s
->code
),
7267 static void exec_command_dump(ExecCommand
*c
, FILE *f
, const char *prefix
) {
7268 _cleanup_free_
char *cmd
= NULL
;
7269 const char *prefix2
;
7274 prefix
= strempty(prefix
);
7275 prefix2
= strjoina(prefix
, "\t");
7277 cmd
= quote_command_line(c
->argv
, SHELL_ESCAPE_EMPTY
);
7280 "%sCommand Line: %s\n",
7281 prefix
, strnull(cmd
));
7283 exec_status_dump(&c
->exec_status
, f
, prefix2
);
7286 void exec_command_dump_list(ExecCommand
*c
, FILE *f
, const char *prefix
) {
7289 prefix
= strempty(prefix
);
7291 LIST_FOREACH(command
, i
, c
)
7292 exec_command_dump(i
, f
, prefix
);
7295 void exec_command_append_list(ExecCommand
**l
, ExecCommand
*e
) {
7302 /* It's kind of important, that we keep the order here */
7303 end
= LIST_FIND_TAIL(command
, *l
);
7304 LIST_INSERT_AFTER(command
, *l
, end
, e
);
7309 int exec_command_set(ExecCommand
*c
, const char *path
, ...) {
7317 l
= strv_new_ap(path
, ap
);
7329 free_and_replace(c
->path
, p
);
7331 return strv_free_and_replace(c
->argv
, l
);
7334 int exec_command_append(ExecCommand
*c
, const char *path
, ...) {
7335 _cleanup_strv_free_
char **l
= NULL
;
7343 l
= strv_new_ap(path
, ap
);
7349 r
= strv_extend_strv(&c
->argv
, l
, false);
7356 static char *destroy_tree(char *path
) {
7360 if (!path_equal(path
, RUN_SYSTEMD_EMPTY
)) {
7361 log_debug("Spawning process to nuke '%s'", path
);
7363 (void) asynchronous_rm_rf(path
, REMOVE_ROOT
|REMOVE_SUBVOLUME
|REMOVE_PHYSICAL
);
7369 static ExecSharedRuntime
* exec_shared_runtime_free(ExecSharedRuntime
*rt
) {
7374 (void) hashmap_remove(rt
->manager
->exec_shared_runtime_by_id
, rt
->id
);
7376 rt
->id
= mfree(rt
->id
);
7377 rt
->tmp_dir
= mfree(rt
->tmp_dir
);
7378 rt
->var_tmp_dir
= mfree(rt
->var_tmp_dir
);
7379 safe_close_pair(rt
->netns_storage_socket
);
7380 safe_close_pair(rt
->ipcns_storage_socket
);
7384 DEFINE_TRIVIAL_UNREF_FUNC(ExecSharedRuntime
, exec_shared_runtime
, exec_shared_runtime_free
);
7385 DEFINE_TRIVIAL_CLEANUP_FUNC(ExecSharedRuntime
*, exec_shared_runtime_free
);
7387 ExecSharedRuntime
* exec_shared_runtime_destroy(ExecSharedRuntime
*rt
) {
7391 assert(rt
->n_ref
> 0);
7397 rt
->tmp_dir
= destroy_tree(rt
->tmp_dir
);
7398 rt
->var_tmp_dir
= destroy_tree(rt
->var_tmp_dir
);
7400 return exec_shared_runtime_free(rt
);
7403 static int exec_shared_runtime_allocate(ExecSharedRuntime
**ret
, const char *id
) {
7404 _cleanup_free_
char *id_copy
= NULL
;
7405 ExecSharedRuntime
*n
;
7409 id_copy
= strdup(id
);
7413 n
= new(ExecSharedRuntime
, 1);
7417 *n
= (ExecSharedRuntime
) {
7418 .id
= TAKE_PTR(id_copy
),
7419 .netns_storage_socket
= PIPE_EBADF
,
7420 .ipcns_storage_socket
= PIPE_EBADF
,
7427 static int exec_shared_runtime_add(
7432 int netns_storage_socket
[2],
7433 int ipcns_storage_socket
[2],
7434 ExecSharedRuntime
**ret
) {
7436 _cleanup_(exec_shared_runtime_freep
) ExecSharedRuntime
*rt
= NULL
;
7442 /* tmp_dir, var_tmp_dir, {net,ipc}ns_storage_socket fds are donated on success */
7444 r
= exec_shared_runtime_allocate(&rt
, id
);
7448 r
= hashmap_ensure_put(&m
->exec_shared_runtime_by_id
, &string_hash_ops
, rt
->id
, rt
);
7452 assert(!!rt
->tmp_dir
== !!rt
->var_tmp_dir
); /* We require both to be set together */
7453 rt
->tmp_dir
= TAKE_PTR(*tmp_dir
);
7454 rt
->var_tmp_dir
= TAKE_PTR(*var_tmp_dir
);
7456 if (netns_storage_socket
) {
7457 rt
->netns_storage_socket
[0] = TAKE_FD(netns_storage_socket
[0]);
7458 rt
->netns_storage_socket
[1] = TAKE_FD(netns_storage_socket
[1]);
7461 if (ipcns_storage_socket
) {
7462 rt
->ipcns_storage_socket
[0] = TAKE_FD(ipcns_storage_socket
[0]);
7463 rt
->ipcns_storage_socket
[1] = TAKE_FD(ipcns_storage_socket
[1]);
7470 /* do not remove created ExecSharedRuntime object when the operation succeeds. */
7475 static int exec_shared_runtime_make(
7477 const ExecContext
*c
,
7479 ExecSharedRuntime
**ret
) {
7481 _cleanup_(namespace_cleanup_tmpdirp
) char *tmp_dir
= NULL
, *var_tmp_dir
= NULL
;
7482 _cleanup_close_pair_
int netns_storage_socket
[2] = PIPE_EBADF
, ipcns_storage_socket
[2] = PIPE_EBADF
;
7489 /* It is not necessary to create ExecSharedRuntime object. */
7490 if (!exec_needs_network_namespace(c
) && !exec_needs_ipc_namespace(c
) && !c
->private_tmp
) {
7495 if (c
->private_tmp
&&
7496 !(prefixed_path_strv_contains(c
->inaccessible_paths
, "/tmp") &&
7497 (prefixed_path_strv_contains(c
->inaccessible_paths
, "/var/tmp") ||
7498 prefixed_path_strv_contains(c
->inaccessible_paths
, "/var")))) {
7499 r
= setup_tmp_dirs(id
, &tmp_dir
, &var_tmp_dir
);
7504 if (exec_needs_network_namespace(c
)) {
7505 if (socketpair(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
, 0, netns_storage_socket
) < 0)
7509 if (exec_needs_ipc_namespace(c
)) {
7510 if (socketpair(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
, 0, ipcns_storage_socket
) < 0)
7514 r
= exec_shared_runtime_add(m
, id
, &tmp_dir
, &var_tmp_dir
, netns_storage_socket
, ipcns_storage_socket
, ret
);
7521 int exec_shared_runtime_acquire(Manager
*m
, const ExecContext
*c
, const char *id
, bool create
, ExecSharedRuntime
**ret
) {
7522 ExecSharedRuntime
*rt
;
7529 rt
= hashmap_get(m
->exec_shared_runtime_by_id
, id
);
7531 /* We already have an ExecSharedRuntime object, let's increase the ref count and reuse it */
7539 /* If not found, then create a new object. */
7540 r
= exec_shared_runtime_make(m
, c
, id
, &rt
);
7544 /* When r == 0, it is not necessary to create ExecSharedRuntime object. */
7550 /* increment reference counter. */
7556 int exec_shared_runtime_serialize(const Manager
*m
, FILE *f
, FDSet
*fds
) {
7557 ExecSharedRuntime
*rt
;
7563 HASHMAP_FOREACH(rt
, m
->exec_shared_runtime_by_id
) {
7564 fprintf(f
, "exec-runtime=%s", rt
->id
);
7567 fprintf(f
, " tmp-dir=%s", rt
->tmp_dir
);
7569 if (rt
->var_tmp_dir
)
7570 fprintf(f
, " var-tmp-dir=%s", rt
->var_tmp_dir
);
7572 if (rt
->netns_storage_socket
[0] >= 0) {
7575 copy
= fdset_put_dup(fds
, rt
->netns_storage_socket
[0]);
7579 fprintf(f
, " netns-socket-0=%i", copy
);
7582 if (rt
->netns_storage_socket
[1] >= 0) {
7585 copy
= fdset_put_dup(fds
, rt
->netns_storage_socket
[1]);
7589 fprintf(f
, " netns-socket-1=%i", copy
);
7592 if (rt
->ipcns_storage_socket
[0] >= 0) {
7595 copy
= fdset_put_dup(fds
, rt
->ipcns_storage_socket
[0]);
7599 fprintf(f
, " ipcns-socket-0=%i", copy
);
7602 if (rt
->ipcns_storage_socket
[1] >= 0) {
7605 copy
= fdset_put_dup(fds
, rt
->ipcns_storage_socket
[1]);
7609 fprintf(f
, " ipcns-socket-1=%i", copy
);
7618 int exec_shared_runtime_deserialize_compat(Unit
*u
, const char *key
, const char *value
, FDSet
*fds
) {
7619 _cleanup_(exec_shared_runtime_freep
) ExecSharedRuntime
*rt_create
= NULL
;
7620 ExecSharedRuntime
*rt
;
7623 /* This is for the migration from old (v237 or earlier) deserialization text.
7624 * Due to the bug #7790, this may not work with the units that use JoinsNamespaceOf=.
7625 * Even if the ExecSharedRuntime object originally created by the other unit, we cannot judge
7626 * so or not from the serialized text, then we always creates a new object owned by this. */
7632 /* Manager manages ExecSharedRuntime objects by the unit id.
7633 * So, we omit the serialized text when the unit does not have id (yet?)... */
7634 if (isempty(u
->id
)) {
7635 log_unit_debug(u
, "Invocation ID not found. Dropping runtime parameter.");
7639 if (hashmap_ensure_allocated(&u
->manager
->exec_shared_runtime_by_id
, &string_hash_ops
) < 0)
7642 rt
= hashmap_get(u
->manager
->exec_shared_runtime_by_id
, u
->id
);
7644 if (exec_shared_runtime_allocate(&rt_create
, u
->id
) < 0)
7650 if (streq(key
, "tmp-dir")) {
7651 if (free_and_strdup_warn(&rt
->tmp_dir
, value
) < 0)
7654 } else if (streq(key
, "var-tmp-dir")) {
7655 if (free_and_strdup_warn(&rt
->var_tmp_dir
, value
) < 0)
7658 } else if (streq(key
, "netns-socket-0")) {
7661 if ((fd
= parse_fd(value
)) < 0 || !fdset_contains(fds
, fd
)) {
7662 log_unit_debug(u
, "Failed to parse netns socket value: %s", value
);
7666 safe_close(rt
->netns_storage_socket
[0]);
7667 rt
->netns_storage_socket
[0] = fdset_remove(fds
, fd
);
7669 } else if (streq(key
, "netns-socket-1")) {
7672 if ((fd
= parse_fd(value
)) < 0 || !fdset_contains(fds
, fd
)) {
7673 log_unit_debug(u
, "Failed to parse netns socket value: %s", value
);
7677 safe_close(rt
->netns_storage_socket
[1]);
7678 rt
->netns_storage_socket
[1] = fdset_remove(fds
, fd
);
7683 /* If the object is newly created, then put it to the hashmap which manages ExecSharedRuntime objects. */
7685 r
= hashmap_put(u
->manager
->exec_shared_runtime_by_id
, rt_create
->id
, rt_create
);
7687 log_unit_debug_errno(u
, r
, "Failed to put runtime parameter to manager's storage: %m");
7691 rt_create
->manager
= u
->manager
;
7694 TAKE_PTR(rt_create
);
7700 int exec_shared_runtime_deserialize_one(Manager
*m
, const char *value
, FDSet
*fds
) {
7701 _cleanup_free_
char *tmp_dir
= NULL
, *var_tmp_dir
= NULL
;
7703 int r
, netns_fdpair
[] = {-1, -1}, ipcns_fdpair
[] = {-1, -1};
7704 const char *p
, *v
= ASSERT_PTR(value
);
7710 n
= strcspn(v
, " ");
7711 id
= strndupa_safe(v
, n
);
7716 v
= startswith(p
, "tmp-dir=");
7718 n
= strcspn(v
, " ");
7719 tmp_dir
= strndup(v
, n
);
7727 v
= startswith(p
, "var-tmp-dir=");
7729 n
= strcspn(v
, " ");
7730 var_tmp_dir
= strndup(v
, n
);
7738 v
= startswith(p
, "netns-socket-0=");
7742 n
= strcspn(v
, " ");
7743 buf
= strndupa_safe(v
, n
);
7745 netns_fdpair
[0] = parse_fd(buf
);
7746 if (netns_fdpair
[0] < 0)
7747 return log_debug_errno(netns_fdpair
[0], "Unable to parse exec-runtime specification netns-socket-0=%s: %m", buf
);
7748 if (!fdset_contains(fds
, netns_fdpair
[0]))
7749 return log_debug_errno(SYNTHETIC_ERRNO(EBADF
),
7750 "exec-runtime specification netns-socket-0= refers to unknown fd %d: %m", netns_fdpair
[0]);
7751 netns_fdpair
[0] = fdset_remove(fds
, netns_fdpair
[0]);
7757 v
= startswith(p
, "netns-socket-1=");
7761 n
= strcspn(v
, " ");
7762 buf
= strndupa_safe(v
, n
);
7764 netns_fdpair
[1] = parse_fd(buf
);
7765 if (netns_fdpair
[1] < 0)
7766 return log_debug_errno(netns_fdpair
[1], "Unable to parse exec-runtime specification netns-socket-1=%s: %m", buf
);
7767 if (!fdset_contains(fds
, netns_fdpair
[1]))
7768 return log_debug_errno(SYNTHETIC_ERRNO(EBADF
),
7769 "exec-runtime specification netns-socket-1= refers to unknown fd %d: %m", netns_fdpair
[1]);
7770 netns_fdpair
[1] = fdset_remove(fds
, netns_fdpair
[1]);
7776 v
= startswith(p
, "ipcns-socket-0=");
7780 n
= strcspn(v
, " ");
7781 buf
= strndupa_safe(v
, n
);
7783 ipcns_fdpair
[0] = parse_fd(buf
);
7784 if (ipcns_fdpair
[0] < 0)
7785 return log_debug_errno(ipcns_fdpair
[0], "Unable to parse exec-runtime specification ipcns-socket-0=%s: %m", buf
);
7786 if (!fdset_contains(fds
, ipcns_fdpair
[0]))
7787 return log_debug_errno(SYNTHETIC_ERRNO(EBADF
),
7788 "exec-runtime specification ipcns-socket-0= refers to unknown fd %d: %m", ipcns_fdpair
[0]);
7789 ipcns_fdpair
[0] = fdset_remove(fds
, ipcns_fdpair
[0]);
7795 v
= startswith(p
, "ipcns-socket-1=");
7799 n
= strcspn(v
, " ");
7800 buf
= strndupa_safe(v
, n
);
7802 ipcns_fdpair
[1] = parse_fd(buf
);
7803 if (ipcns_fdpair
[1] < 0)
7804 return log_debug_errno(ipcns_fdpair
[1], "Unable to parse exec-runtime specification ipcns-socket-1=%s: %m", buf
);
7805 if (!fdset_contains(fds
, ipcns_fdpair
[1]))
7806 return log_debug_errno(SYNTHETIC_ERRNO(EBADF
),
7807 "exec-runtime specification ipcns-socket-1= refers to unknown fd %d: %m", ipcns_fdpair
[1]);
7808 ipcns_fdpair
[1] = fdset_remove(fds
, ipcns_fdpair
[1]);
7812 r
= exec_shared_runtime_add(m
, id
, &tmp_dir
, &var_tmp_dir
, netns_fdpair
, ipcns_fdpair
, NULL
);
7814 return log_debug_errno(r
, "Failed to add exec-runtime: %m");
7818 void exec_shared_runtime_vacuum(Manager
*m
) {
7819 ExecSharedRuntime
*rt
;
7823 /* Free unreferenced ExecSharedRuntime objects. This is used after manager deserialization process. */
7825 HASHMAP_FOREACH(rt
, m
->exec_shared_runtime_by_id
) {
7829 (void) exec_shared_runtime_free(rt
);
7833 int exec_runtime_make(
7835 const ExecContext
*context
,
7836 ExecSharedRuntime
*shared
,
7837 DynamicCreds
*creds
,
7838 ExecRuntime
**ret
) {
7839 _cleanup_close_pair_
int ephemeral_storage_socket
[2] = PIPE_EBADF
;
7840 _cleanup_free_
char *ephemeral
= NULL
;
7841 _cleanup_(exec_runtime_freep
) ExecRuntime
*rt
= NULL
;
7848 if (!shared
&& !creds
&& !exec_needs_ephemeral(context
)) {
7853 if (exec_needs_ephemeral(context
)) {
7854 r
= mkdir_p("/var/lib/systemd/ephemeral-trees", 0755);
7858 r
= tempfn_random_child("/var/lib/systemd/ephemeral-trees", unit
->id
, &ephemeral
);
7862 if (socketpair(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
, 0, ephemeral_storage_socket
) < 0)
7866 rt
= new(ExecRuntime
, 1);
7870 *rt
= (ExecRuntime
) {
7872 .dynamic_creds
= creds
,
7873 .ephemeral_copy
= TAKE_PTR(ephemeral
),
7874 .ephemeral_storage_socket
[0] = TAKE_FD(ephemeral_storage_socket
[0]),
7875 .ephemeral_storage_socket
[1] = TAKE_FD(ephemeral_storage_socket
[1]),
7878 *ret
= TAKE_PTR(rt
);
7882 ExecRuntime
* exec_runtime_free(ExecRuntime
*rt
) {
7886 exec_shared_runtime_unref(rt
->shared
);
7887 dynamic_creds_unref(rt
->dynamic_creds
);
7889 rt
->ephemeral_copy
= destroy_tree(rt
->ephemeral_copy
);
7891 safe_close_pair(rt
->ephemeral_storage_socket
);
7895 ExecRuntime
* exec_runtime_destroy(ExecRuntime
*rt
) {
7899 rt
->shared
= exec_shared_runtime_destroy(rt
->shared
);
7900 rt
->dynamic_creds
= dynamic_creds_destroy(rt
->dynamic_creds
);
7901 return exec_runtime_free(rt
);
7904 void exec_params_clear(ExecParameters
*p
) {
7908 p
->environment
= strv_free(p
->environment
);
7909 p
->fd_names
= strv_free(p
->fd_names
);
7910 p
->fds
= mfree(p
->fds
);
7911 p
->exec_fd
= safe_close(p
->exec_fd
);
7914 ExecSetCredential
*exec_set_credential_free(ExecSetCredential
*sc
) {
7923 ExecLoadCredential
*exec_load_credential_free(ExecLoadCredential
*lc
) {
7932 void exec_directory_done(ExecDirectory
*d
) {
7936 for (size_t i
= 0; i
< d
->n_items
; i
++) {
7937 free(d
->items
[i
].path
);
7938 strv_free(d
->items
[i
].symlinks
);
7941 d
->items
= mfree(d
->items
);
7946 static ExecDirectoryItem
*exec_directory_find(ExecDirectory
*d
, const char *path
) {
7950 for (size_t i
= 0; i
< d
->n_items
; i
++)
7951 if (path_equal(d
->items
[i
].path
, path
))
7952 return &d
->items
[i
];
7957 int exec_directory_add(ExecDirectory
*d
, const char *path
, const char *symlink
) {
7958 _cleanup_strv_free_
char **s
= NULL
;
7959 _cleanup_free_
char *p
= NULL
;
7960 ExecDirectoryItem
*existing
;
7966 existing
= exec_directory_find(d
, path
);
7968 r
= strv_extend(&existing
->symlinks
, symlink
);
7972 return 0; /* existing item is updated */
7980 s
= strv_new(symlink
);
7985 if (!GREEDY_REALLOC(d
->items
, d
->n_items
+ 1))
7988 d
->items
[d
->n_items
++] = (ExecDirectoryItem
) {
7989 .path
= TAKE_PTR(p
),
7990 .symlinks
= TAKE_PTR(s
),
7993 return 1; /* new item is added */
7996 static int exec_directory_item_compare_func(const ExecDirectoryItem
*a
, const ExecDirectoryItem
*b
) {
8000 return path_compare(a
->path
, b
->path
);
8003 void exec_directory_sort(ExecDirectory
*d
) {
8006 /* Sort the exec directories to make always parent directories processed at first in
8007 * setup_exec_directory(), e.g., even if StateDirectory=foo/bar foo, we need to create foo at first,
8008 * then foo/bar. Also, set .only_create flag if one of the parent directories is contained in the
8009 * list. See also comments in setup_exec_directory() and issue #24783. */
8011 if (d
->n_items
<= 1)
8014 typesafe_qsort(d
->items
, d
->n_items
, exec_directory_item_compare_func
);
8016 for (size_t i
= 1; i
< d
->n_items
; i
++)
8017 for (size_t j
= 0; j
< i
; j
++)
8018 if (path_startswith(d
->items
[i
].path
, d
->items
[j
].path
)) {
8019 d
->items
[i
].only_create
= true;
8024 ExecCleanMask
exec_clean_mask_from_string(const char *s
) {
8025 ExecDirectoryType t
;
8029 if (streq(s
, "all"))
8030 return EXEC_CLEAN_ALL
;
8031 if (streq(s
, "fdstore"))
8032 return EXEC_CLEAN_FDSTORE
;
8034 t
= exec_resource_type_from_string(s
);
8036 return (ExecCleanMask
) t
;
8041 DEFINE_HASH_OPS_WITH_VALUE_DESTRUCTOR(exec_set_credential_hash_ops
, char, string_hash_func
, string_compare_func
, ExecSetCredential
, exec_set_credential_free
);
8042 DEFINE_HASH_OPS_WITH_VALUE_DESTRUCTOR(exec_load_credential_hash_ops
, char, string_hash_func
, string_compare_func
, ExecLoadCredential
, exec_load_credential_free
);
8044 static const char* const exec_input_table
[_EXEC_INPUT_MAX
] = {
8045 [EXEC_INPUT_NULL
] = "null",
8046 [EXEC_INPUT_TTY
] = "tty",
8047 [EXEC_INPUT_TTY_FORCE
] = "tty-force",
8048 [EXEC_INPUT_TTY_FAIL
] = "tty-fail",
8049 [EXEC_INPUT_SOCKET
] = "socket",
8050 [EXEC_INPUT_NAMED_FD
] = "fd",
8051 [EXEC_INPUT_DATA
] = "data",
8052 [EXEC_INPUT_FILE
] = "file",
8055 DEFINE_STRING_TABLE_LOOKUP(exec_input
, ExecInput
);
8057 static const char* const exec_output_table
[_EXEC_OUTPUT_MAX
] = {
8058 [EXEC_OUTPUT_INHERIT
] = "inherit",
8059 [EXEC_OUTPUT_NULL
] = "null",
8060 [EXEC_OUTPUT_TTY
] = "tty",
8061 [EXEC_OUTPUT_KMSG
] = "kmsg",
8062 [EXEC_OUTPUT_KMSG_AND_CONSOLE
] = "kmsg+console",
8063 [EXEC_OUTPUT_JOURNAL
] = "journal",
8064 [EXEC_OUTPUT_JOURNAL_AND_CONSOLE
] = "journal+console",
8065 [EXEC_OUTPUT_SOCKET
] = "socket",
8066 [EXEC_OUTPUT_NAMED_FD
] = "fd",
8067 [EXEC_OUTPUT_FILE
] = "file",
8068 [EXEC_OUTPUT_FILE_APPEND
] = "append",
8069 [EXEC_OUTPUT_FILE_TRUNCATE
] = "truncate",
8072 DEFINE_STRING_TABLE_LOOKUP(exec_output
, ExecOutput
);
8074 static const char* const exec_utmp_mode_table
[_EXEC_UTMP_MODE_MAX
] = {
8075 [EXEC_UTMP_INIT
] = "init",
8076 [EXEC_UTMP_LOGIN
] = "login",
8077 [EXEC_UTMP_USER
] = "user",
8080 DEFINE_STRING_TABLE_LOOKUP(exec_utmp_mode
, ExecUtmpMode
);
8082 static const char* const exec_preserve_mode_table
[_EXEC_PRESERVE_MODE_MAX
] = {
8083 [EXEC_PRESERVE_NO
] = "no",
8084 [EXEC_PRESERVE_YES
] = "yes",
8085 [EXEC_PRESERVE_RESTART
] = "restart",
8088 DEFINE_STRING_TABLE_LOOKUP_WITH_BOOLEAN(exec_preserve_mode
, ExecPreserveMode
, EXEC_PRESERVE_YES
);
8090 /* This table maps ExecDirectoryType to the setting it is configured with in the unit */
8091 static const char* const exec_directory_type_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
8092 [EXEC_DIRECTORY_RUNTIME
] = "RuntimeDirectory",
8093 [EXEC_DIRECTORY_STATE
] = "StateDirectory",
8094 [EXEC_DIRECTORY_CACHE
] = "CacheDirectory",
8095 [EXEC_DIRECTORY_LOGS
] = "LogsDirectory",
8096 [EXEC_DIRECTORY_CONFIGURATION
] = "ConfigurationDirectory",
8099 DEFINE_STRING_TABLE_LOOKUP(exec_directory_type
, ExecDirectoryType
);
8101 /* This table maps ExecDirectoryType to the symlink setting it is configured with in the unit */
8102 static const char* const exec_directory_type_symlink_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
8103 [EXEC_DIRECTORY_RUNTIME
] = "RuntimeDirectorySymlink",
8104 [EXEC_DIRECTORY_STATE
] = "StateDirectorySymlink",
8105 [EXEC_DIRECTORY_CACHE
] = "CacheDirectorySymlink",
8106 [EXEC_DIRECTORY_LOGS
] = "LogsDirectorySymlink",
8107 [EXEC_DIRECTORY_CONFIGURATION
] = "ConfigurationDirectorySymlink",
8110 DEFINE_STRING_TABLE_LOOKUP(exec_directory_type_symlink
, ExecDirectoryType
);
8112 /* And this table maps ExecDirectoryType too, but to a generic term identifying the type of resource. This
8113 * one is supposed to be generic enough to be used for unit types that don't use ExecContext and per-unit
8114 * directories, specifically .timer units with their timestamp touch file. */
8115 static const char* const exec_resource_type_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
8116 [EXEC_DIRECTORY_RUNTIME
] = "runtime",
8117 [EXEC_DIRECTORY_STATE
] = "state",
8118 [EXEC_DIRECTORY_CACHE
] = "cache",
8119 [EXEC_DIRECTORY_LOGS
] = "logs",
8120 [EXEC_DIRECTORY_CONFIGURATION
] = "configuration",
8123 DEFINE_STRING_TABLE_LOOKUP(exec_resource_type
, ExecDirectoryType
);
8125 /* And this table also maps ExecDirectoryType, to the environment variable we pass the selected directory to
8126 * the service payload in. */
8127 static const char* const exec_directory_env_name_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
8128 [EXEC_DIRECTORY_RUNTIME
] = "RUNTIME_DIRECTORY",
8129 [EXEC_DIRECTORY_STATE
] = "STATE_DIRECTORY",
8130 [EXEC_DIRECTORY_CACHE
] = "CACHE_DIRECTORY",
8131 [EXEC_DIRECTORY_LOGS
] = "LOGS_DIRECTORY",
8132 [EXEC_DIRECTORY_CONFIGURATION
] = "CONFIGURATION_DIRECTORY",
8135 DEFINE_PRIVATE_STRING_TABLE_LOOKUP_TO_STRING(exec_directory_env_name
, ExecDirectoryType
);
8137 static const char* const exec_keyring_mode_table
[_EXEC_KEYRING_MODE_MAX
] = {
8138 [EXEC_KEYRING_INHERIT
] = "inherit",
8139 [EXEC_KEYRING_PRIVATE
] = "private",
8140 [EXEC_KEYRING_SHARED
] = "shared",
8143 DEFINE_STRING_TABLE_LOOKUP(exec_keyring_mode
, ExecKeyringMode
);