1 /* SPDX-License-Identifier: LGPL-2.1+ */
10 #include <sys/capability.h>
11 #include <sys/eventfd.h>
13 #include <sys/personality.h>
14 #include <sys/prctl.h>
16 #include <sys/socket.h>
18 #include <sys/types.h>
24 #include <security/pam_appl.h>
28 #include <selinux/selinux.h>
36 #include <sys/apparmor.h>
39 #include "sd-messages.h"
42 #include "alloc-util.h"
44 #include "apparmor-util.h"
49 #include "capability-util.h"
50 #include "chown-recursive.h"
51 #include "cpu-set-util.h"
55 #include "errno-list.h"
57 #include "exit-status.h"
59 #include "format-util.h"
61 #include "glob-util.h"
68 #include "memory-util.h"
71 #include "namespace.h"
72 #include "parse-util.h"
73 #include "path-util.h"
74 #include "process-util.h"
75 #include "rlimit-util.h"
78 #include "seccomp-util.h"
80 #include "securebits-util.h"
81 #include "selinux-util.h"
82 #include "signal-util.h"
83 #include "smack-util.h"
84 #include "socket-util.h"
86 #include "stat-util.h"
87 #include "string-table.h"
88 #include "string-util.h"
90 #include "syslog-util.h"
91 #include "terminal-util.h"
92 #include "umask-util.h"
94 #include "user-util.h"
95 #include "utmp-wtmp.h"
97 #define IDLE_TIMEOUT_USEC (5*USEC_PER_SEC)
98 #define IDLE_TIMEOUT2_USEC (1*USEC_PER_SEC)
100 /* This assumes there is a 'tty' group */
101 #define TTY_MODE 0620
103 #define SNDBUF_SIZE (8*1024*1024)
105 static int shift_fds(int fds
[], size_t n_fds
) {
106 int start
, restart_from
;
111 /* Modifies the fds array! (sorts it) */
121 for (i
= start
; i
< (int) n_fds
; i
++) {
124 /* Already at right index? */
128 nfd
= fcntl(fds
[i
], F_DUPFD
, i
+ 3);
135 /* Hmm, the fd we wanted isn't free? Then
136 * let's remember that and try again from here */
137 if (nfd
!= i
+3 && restart_from
< 0)
141 if (restart_from
< 0)
144 start
= restart_from
;
150 static int flags_fds(const int fds
[], size_t n_socket_fds
, size_t n_storage_fds
, bool nonblock
) {
154 n_fds
= n_socket_fds
+ n_storage_fds
;
160 /* Drops/Sets O_NONBLOCK and FD_CLOEXEC from the file flags.
161 * O_NONBLOCK only applies to socket activation though. */
163 for (i
= 0; i
< n_fds
; i
++) {
165 if (i
< n_socket_fds
) {
166 r
= fd_nonblock(fds
[i
], nonblock
);
171 /* We unconditionally drop FD_CLOEXEC from the fds,
172 * since after all we want to pass these fds to our
175 r
= fd_cloexec(fds
[i
], false);
183 static const char *exec_context_tty_path(const ExecContext
*context
) {
186 if (context
->stdio_as_fds
)
189 if (context
->tty_path
)
190 return context
->tty_path
;
192 return "/dev/console";
195 static void exec_context_tty_reset(const ExecContext
*context
, const ExecParameters
*p
) {
200 path
= exec_context_tty_path(context
);
202 if (context
->tty_vhangup
) {
203 if (p
&& p
->stdin_fd
>= 0)
204 (void) terminal_vhangup_fd(p
->stdin_fd
);
206 (void) terminal_vhangup(path
);
209 if (context
->tty_reset
) {
210 if (p
&& p
->stdin_fd
>= 0)
211 (void) reset_terminal_fd(p
->stdin_fd
, true);
213 (void) reset_terminal(path
);
216 if (context
->tty_vt_disallocate
&& path
)
217 (void) vt_disallocate(path
);
220 static bool is_terminal_input(ExecInput i
) {
223 EXEC_INPUT_TTY_FORCE
,
224 EXEC_INPUT_TTY_FAIL
);
227 static bool is_terminal_output(ExecOutput o
) {
230 EXEC_OUTPUT_SYSLOG_AND_CONSOLE
,
231 EXEC_OUTPUT_KMSG_AND_CONSOLE
,
232 EXEC_OUTPUT_JOURNAL_AND_CONSOLE
);
235 static bool is_syslog_output(ExecOutput o
) {
238 EXEC_OUTPUT_SYSLOG_AND_CONSOLE
);
241 static bool is_kmsg_output(ExecOutput o
) {
244 EXEC_OUTPUT_KMSG_AND_CONSOLE
);
247 static bool exec_context_needs_term(const ExecContext
*c
) {
250 /* Return true if the execution context suggests we should set $TERM to something useful. */
252 if (is_terminal_input(c
->std_input
))
255 if (is_terminal_output(c
->std_output
))
258 if (is_terminal_output(c
->std_error
))
261 return !!c
->tty_path
;
264 static int open_null_as(int flags
, int nfd
) {
269 fd
= open("/dev/null", flags
|O_NOCTTY
);
273 return move_fd(fd
, nfd
, false);
276 static int connect_journal_socket(int fd
, uid_t uid
, gid_t gid
) {
277 static const union sockaddr_union sa
= {
278 .un
.sun_family
= AF_UNIX
,
279 .un
.sun_path
= "/run/systemd/journal/stdout",
281 uid_t olduid
= UID_INVALID
;
282 gid_t oldgid
= GID_INVALID
;
285 if (gid_is_valid(gid
)) {
288 if (setegid(gid
) < 0)
292 if (uid_is_valid(uid
)) {
295 if (seteuid(uid
) < 0) {
301 r
= connect(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
)) < 0 ? -errno
: 0;
303 /* If we fail to restore the uid or gid, things will likely
304 fail later on. This should only happen if an LSM interferes. */
306 if (uid_is_valid(uid
))
307 (void) seteuid(olduid
);
310 if (gid_is_valid(gid
))
311 (void) setegid(oldgid
);
316 static int connect_logger_as(
318 const ExecContext
*context
,
319 const ExecParameters
*params
,
326 _cleanup_close_
int fd
= -1;
331 assert(output
< _EXEC_OUTPUT_MAX
);
335 fd
= socket(AF_UNIX
, SOCK_STREAM
, 0);
339 r
= connect_journal_socket(fd
, uid
, gid
);
343 if (shutdown(fd
, SHUT_RD
) < 0)
346 (void) fd_inc_sndbuf(fd
, SNDBUF_SIZE
);
356 context
->syslog_identifier
?: ident
,
357 params
->flags
& EXEC_PASS_LOG_UNIT
? unit
->id
: "",
358 context
->syslog_priority
,
359 !!context
->syslog_level_prefix
,
360 is_syslog_output(output
),
361 is_kmsg_output(output
),
362 is_terminal_output(output
)) < 0)
365 return move_fd(TAKE_FD(fd
), nfd
, false);
368 static int open_terminal_as(const char *path
, int flags
, int nfd
) {
374 fd
= open_terminal(path
, flags
| O_NOCTTY
);
378 return move_fd(fd
, nfd
, false);
381 static int acquire_path(const char *path
, int flags
, mode_t mode
) {
382 union sockaddr_union sa
= {};
383 _cleanup_close_
int fd
= -1;
388 if (IN_SET(flags
& O_ACCMODE
, O_WRONLY
, O_RDWR
))
391 fd
= open(path
, flags
|O_NOCTTY
, mode
);
395 if (errno
!= ENXIO
) /* ENXIO is returned when we try to open() an AF_UNIX file system socket on Linux */
397 if (strlen(path
) >= sizeof(sa
.un
.sun_path
)) /* Too long, can't be a UNIX socket */
400 /* So, it appears the specified path could be an AF_UNIX socket. Let's see if we can connect to it. */
402 fd
= socket(AF_UNIX
, SOCK_STREAM
, 0);
406 salen
= sockaddr_un_set_path(&sa
.un
, path
);
410 if (connect(fd
, &sa
.sa
, salen
) < 0)
411 return errno
== EINVAL
? -ENXIO
: -errno
; /* Propagate initial error if we get EINVAL, i.e. we have
412 * indication that his wasn't an AF_UNIX socket after all */
414 if ((flags
& O_ACCMODE
) == O_RDONLY
)
415 r
= shutdown(fd
, SHUT_WR
);
416 else if ((flags
& O_ACCMODE
) == O_WRONLY
)
417 r
= shutdown(fd
, SHUT_RD
);
426 static int fixup_input(
427 const ExecContext
*context
,
429 bool apply_tty_stdin
) {
435 std_input
= context
->std_input
;
437 if (is_terminal_input(std_input
) && !apply_tty_stdin
)
438 return EXEC_INPUT_NULL
;
440 if (std_input
== EXEC_INPUT_SOCKET
&& socket_fd
< 0)
441 return EXEC_INPUT_NULL
;
443 if (std_input
== EXEC_INPUT_DATA
&& context
->stdin_data_size
== 0)
444 return EXEC_INPUT_NULL
;
449 static int fixup_output(ExecOutput std_output
, int socket_fd
) {
451 if (std_output
== EXEC_OUTPUT_SOCKET
&& socket_fd
< 0)
452 return EXEC_OUTPUT_INHERIT
;
457 static int setup_input(
458 const ExecContext
*context
,
459 const ExecParameters
*params
,
461 int named_iofds
[3]) {
468 if (params
->stdin_fd
>= 0) {
469 if (dup2(params
->stdin_fd
, STDIN_FILENO
) < 0)
472 /* Try to make this the controlling tty, if it is a tty, and reset it */
473 if (isatty(STDIN_FILENO
)) {
474 (void) ioctl(STDIN_FILENO
, TIOCSCTTY
, context
->std_input
== EXEC_INPUT_TTY_FORCE
);
475 (void) reset_terminal_fd(STDIN_FILENO
, true);
481 i
= fixup_input(context
, socket_fd
, params
->flags
& EXEC_APPLY_TTY_STDIN
);
485 case EXEC_INPUT_NULL
:
486 return open_null_as(O_RDONLY
, STDIN_FILENO
);
489 case EXEC_INPUT_TTY_FORCE
:
490 case EXEC_INPUT_TTY_FAIL
: {
493 fd
= acquire_terminal(exec_context_tty_path(context
),
494 i
== EXEC_INPUT_TTY_FAIL
? ACQUIRE_TERMINAL_TRY
:
495 i
== EXEC_INPUT_TTY_FORCE
? ACQUIRE_TERMINAL_FORCE
:
496 ACQUIRE_TERMINAL_WAIT
,
501 return move_fd(fd
, STDIN_FILENO
, false);
504 case EXEC_INPUT_SOCKET
:
505 assert(socket_fd
>= 0);
507 return dup2(socket_fd
, STDIN_FILENO
) < 0 ? -errno
: STDIN_FILENO
;
509 case EXEC_INPUT_NAMED_FD
:
510 assert(named_iofds
[STDIN_FILENO
] >= 0);
512 (void) fd_nonblock(named_iofds
[STDIN_FILENO
], false);
513 return dup2(named_iofds
[STDIN_FILENO
], STDIN_FILENO
) < 0 ? -errno
: STDIN_FILENO
;
515 case EXEC_INPUT_DATA
: {
518 fd
= acquire_data_fd(context
->stdin_data
, context
->stdin_data_size
, 0);
522 return move_fd(fd
, STDIN_FILENO
, false);
525 case EXEC_INPUT_FILE
: {
529 assert(context
->stdio_file
[STDIN_FILENO
]);
531 rw
= (context
->std_output
== EXEC_OUTPUT_FILE
&& streq_ptr(context
->stdio_file
[STDIN_FILENO
], context
->stdio_file
[STDOUT_FILENO
])) ||
532 (context
->std_error
== EXEC_OUTPUT_FILE
&& streq_ptr(context
->stdio_file
[STDIN_FILENO
], context
->stdio_file
[STDERR_FILENO
]));
534 fd
= acquire_path(context
->stdio_file
[STDIN_FILENO
], rw
? O_RDWR
: O_RDONLY
, 0666 & ~context
->umask
);
538 return move_fd(fd
, STDIN_FILENO
, false);
542 assert_not_reached("Unknown input type");
546 static bool can_inherit_stderr_from_stdout(
547 const ExecContext
*context
,
553 /* Returns true, if given the specified STDERR and STDOUT output we can directly dup() the stdout fd to the
556 if (e
== EXEC_OUTPUT_INHERIT
)
561 if (e
== EXEC_OUTPUT_NAMED_FD
)
562 return streq_ptr(context
->stdio_fdname
[STDOUT_FILENO
], context
->stdio_fdname
[STDERR_FILENO
]);
564 if (IN_SET(e
, EXEC_OUTPUT_FILE
, EXEC_OUTPUT_FILE_APPEND
))
565 return streq_ptr(context
->stdio_file
[STDOUT_FILENO
], context
->stdio_file
[STDERR_FILENO
]);
570 static int setup_output(
572 const ExecContext
*context
,
573 const ExecParameters
*params
,
580 dev_t
*journal_stream_dev
,
581 ino_t
*journal_stream_ino
) {
591 assert(journal_stream_dev
);
592 assert(journal_stream_ino
);
594 if (fileno
== STDOUT_FILENO
&& params
->stdout_fd
>= 0) {
596 if (dup2(params
->stdout_fd
, STDOUT_FILENO
) < 0)
599 return STDOUT_FILENO
;
602 if (fileno
== STDERR_FILENO
&& params
->stderr_fd
>= 0) {
603 if (dup2(params
->stderr_fd
, STDERR_FILENO
) < 0)
606 return STDERR_FILENO
;
609 i
= fixup_input(context
, socket_fd
, params
->flags
& EXEC_APPLY_TTY_STDIN
);
610 o
= fixup_output(context
->std_output
, socket_fd
);
612 if (fileno
== STDERR_FILENO
) {
614 e
= fixup_output(context
->std_error
, socket_fd
);
616 /* This expects the input and output are already set up */
618 /* Don't change the stderr file descriptor if we inherit all
619 * the way and are not on a tty */
620 if (e
== EXEC_OUTPUT_INHERIT
&&
621 o
== EXEC_OUTPUT_INHERIT
&&
622 i
== EXEC_INPUT_NULL
&&
623 !is_terminal_input(context
->std_input
) &&
627 /* Duplicate from stdout if possible */
628 if (can_inherit_stderr_from_stdout(context
, o
, e
))
629 return dup2(STDOUT_FILENO
, fileno
) < 0 ? -errno
: fileno
;
633 } else if (o
== EXEC_OUTPUT_INHERIT
) {
634 /* If input got downgraded, inherit the original value */
635 if (i
== EXEC_INPUT_NULL
&& is_terminal_input(context
->std_input
))
636 return open_terminal_as(exec_context_tty_path(context
), O_WRONLY
, fileno
);
638 /* If the input is connected to anything that's not a /dev/null or a data fd, inherit that... */
639 if (!IN_SET(i
, EXEC_INPUT_NULL
, EXEC_INPUT_DATA
))
640 return dup2(STDIN_FILENO
, fileno
) < 0 ? -errno
: fileno
;
642 /* If we are not started from PID 1 we just inherit STDOUT from our parent process. */
646 /* We need to open /dev/null here anew, to get the right access mode. */
647 return open_null_as(O_WRONLY
, fileno
);
652 case EXEC_OUTPUT_NULL
:
653 return open_null_as(O_WRONLY
, fileno
);
655 case EXEC_OUTPUT_TTY
:
656 if (is_terminal_input(i
))
657 return dup2(STDIN_FILENO
, fileno
) < 0 ? -errno
: fileno
;
659 /* We don't reset the terminal if this is just about output */
660 return open_terminal_as(exec_context_tty_path(context
), O_WRONLY
, fileno
);
662 case EXEC_OUTPUT_SYSLOG
:
663 case EXEC_OUTPUT_SYSLOG_AND_CONSOLE
:
664 case EXEC_OUTPUT_KMSG
:
665 case EXEC_OUTPUT_KMSG_AND_CONSOLE
:
666 case EXEC_OUTPUT_JOURNAL
:
667 case EXEC_OUTPUT_JOURNAL_AND_CONSOLE
:
668 r
= connect_logger_as(unit
, context
, params
, o
, ident
, fileno
, uid
, gid
);
670 log_unit_warning_errno(unit
, r
, "Failed to connect %s to the journal socket, ignoring: %m", fileno
== STDOUT_FILENO
? "stdout" : "stderr");
671 r
= open_null_as(O_WRONLY
, fileno
);
675 /* If we connected this fd to the journal via a stream, patch the device/inode into the passed
676 * parameters, but only then. This is useful so that we can set $JOURNAL_STREAM that permits
677 * services to detect whether they are connected to the journal or not.
679 * If both stdout and stderr are connected to a stream then let's make sure to store the data
680 * about STDERR as that's usually the best way to do logging. */
682 if (fstat(fileno
, &st
) >= 0 &&
683 (*journal_stream_ino
== 0 || fileno
== STDERR_FILENO
)) {
684 *journal_stream_dev
= st
.st_dev
;
685 *journal_stream_ino
= st
.st_ino
;
690 case EXEC_OUTPUT_SOCKET
:
691 assert(socket_fd
>= 0);
693 return dup2(socket_fd
, fileno
) < 0 ? -errno
: fileno
;
695 case EXEC_OUTPUT_NAMED_FD
:
696 assert(named_iofds
[fileno
] >= 0);
698 (void) fd_nonblock(named_iofds
[fileno
], false);
699 return dup2(named_iofds
[fileno
], fileno
) < 0 ? -errno
: fileno
;
701 case EXEC_OUTPUT_FILE
:
702 case EXEC_OUTPUT_FILE_APPEND
: {
706 assert(context
->stdio_file
[fileno
]);
708 rw
= context
->std_input
== EXEC_INPUT_FILE
&&
709 streq_ptr(context
->stdio_file
[fileno
], context
->stdio_file
[STDIN_FILENO
]);
712 return dup2(STDIN_FILENO
, fileno
) < 0 ? -errno
: fileno
;
715 if (o
== EXEC_OUTPUT_FILE_APPEND
)
718 fd
= acquire_path(context
->stdio_file
[fileno
], flags
, 0666 & ~context
->umask
);
722 return move_fd(fd
, fileno
, 0);
726 assert_not_reached("Unknown error type");
730 static int chown_terminal(int fd
, uid_t uid
) {
735 /* Before we chown/chmod the TTY, let's ensure this is actually a tty */
736 if (isatty(fd
) < 1) {
737 if (IN_SET(errno
, EINVAL
, ENOTTY
))
738 return 0; /* not a tty */
743 /* This might fail. What matters are the results. */
744 r
= fchmod_and_chown(fd
, TTY_MODE
, uid
, -1);
751 static int setup_confirm_stdio(const char *vc
, int *_saved_stdin
, int *_saved_stdout
) {
752 _cleanup_close_
int fd
= -1, saved_stdin
= -1, saved_stdout
= -1;
755 assert(_saved_stdin
);
756 assert(_saved_stdout
);
758 saved_stdin
= fcntl(STDIN_FILENO
, F_DUPFD
, 3);
762 saved_stdout
= fcntl(STDOUT_FILENO
, F_DUPFD
, 3);
763 if (saved_stdout
< 0)
766 fd
= acquire_terminal(vc
, ACQUIRE_TERMINAL_WAIT
, DEFAULT_CONFIRM_USEC
);
770 r
= chown_terminal(fd
, getuid());
774 r
= reset_terminal_fd(fd
, true);
778 r
= rearrange_stdio(fd
, fd
, STDERR_FILENO
);
783 *_saved_stdin
= saved_stdin
;
784 *_saved_stdout
= saved_stdout
;
786 saved_stdin
= saved_stdout
= -1;
791 static void write_confirm_error_fd(int err
, int fd
, const Unit
*u
) {
794 if (err
== -ETIMEDOUT
)
795 dprintf(fd
, "Confirmation question timed out for %s, assuming positive response.\n", u
->id
);
798 dprintf(fd
, "Couldn't ask confirmation for %s: %m, assuming positive response.\n", u
->id
);
802 static void write_confirm_error(int err
, const char *vc
, const Unit
*u
) {
803 _cleanup_close_
int fd
= -1;
807 fd
= open_terminal(vc
, O_WRONLY
|O_NOCTTY
|O_CLOEXEC
);
811 write_confirm_error_fd(err
, fd
, u
);
814 static int restore_confirm_stdio(int *saved_stdin
, int *saved_stdout
) {
818 assert(saved_stdout
);
822 if (*saved_stdin
>= 0)
823 if (dup2(*saved_stdin
, STDIN_FILENO
) < 0)
826 if (*saved_stdout
>= 0)
827 if (dup2(*saved_stdout
, STDOUT_FILENO
) < 0)
830 *saved_stdin
= safe_close(*saved_stdin
);
831 *saved_stdout
= safe_close(*saved_stdout
);
837 CONFIRM_PRETEND_FAILURE
= -1,
838 CONFIRM_PRETEND_SUCCESS
= 0,
842 static int ask_for_confirmation(const char *vc
, Unit
*u
, const char *cmdline
) {
843 int saved_stdout
= -1, saved_stdin
= -1, r
;
844 _cleanup_free_
char *e
= NULL
;
847 /* For any internal errors, assume a positive response. */
848 r
= setup_confirm_stdio(vc
, &saved_stdin
, &saved_stdout
);
850 write_confirm_error(r
, vc
, u
);
851 return CONFIRM_EXECUTE
;
854 /* confirm_spawn might have been disabled while we were sleeping. */
855 if (manager_is_confirm_spawn_disabled(u
->manager
)) {
860 e
= ellipsize(cmdline
, 60, 100);
868 r
= ask_char(&c
, "yfshiDjcn", "Execute %s? [y, f, s – h for help] ", e
);
870 write_confirm_error_fd(r
, STDOUT_FILENO
, u
);
877 printf("Resuming normal execution.\n");
878 manager_disable_confirm_spawn();
882 unit_dump(u
, stdout
, " ");
883 continue; /* ask again */
885 printf("Failing execution.\n");
886 r
= CONFIRM_PRETEND_FAILURE
;
889 printf(" c - continue, proceed without asking anymore\n"
890 " D - dump, show the state of the unit\n"
891 " f - fail, don't execute the command and pretend it failed\n"
893 " i - info, show a short summary of the unit\n"
894 " j - jobs, show jobs that are in progress\n"
895 " s - skip, don't execute the command and pretend it succeeded\n"
896 " y - yes, execute the command\n");
897 continue; /* ask again */
899 printf(" Description: %s\n"
902 u
->id
, u
->description
, cmdline
);
903 continue; /* ask again */
905 manager_dump_jobs(u
->manager
, stdout
, " ");
906 continue; /* ask again */
908 /* 'n' was removed in favor of 'f'. */
909 printf("Didn't understand 'n', did you mean 'f'?\n");
910 continue; /* ask again */
912 printf("Skipping execution.\n");
913 r
= CONFIRM_PRETEND_SUCCESS
;
919 assert_not_reached("Unhandled choice");
925 restore_confirm_stdio(&saved_stdin
, &saved_stdout
);
929 static int get_fixed_user(const ExecContext
*c
, const char **user
,
930 uid_t
*uid
, gid_t
*gid
,
931 const char **home
, const char **shell
) {
940 /* Note that we don't set $HOME or $SHELL if they are not particularly enlightening anyway
941 * (i.e. are "/" or "/bin/nologin"). */
944 r
= get_user_creds(&name
, uid
, gid
, home
, shell
, USER_CREDS_CLEAN
);
952 static int get_fixed_group(const ExecContext
*c
, const char **group
, gid_t
*gid
) {
962 r
= get_group_creds(&name
, gid
, 0);
970 static int get_supplementary_groups(const ExecContext
*c
, const char *user
,
971 const char *group
, gid_t gid
,
972 gid_t
**supplementary_gids
, int *ngids
) {
976 bool keep_groups
= false;
977 gid_t
*groups
= NULL
;
978 _cleanup_free_ gid_t
*l_gids
= NULL
;
983 * If user is given, then lookup GID and supplementary groups list.
984 * We avoid NSS lookups for gid=0. Also we have to initialize groups
985 * here and as early as possible so we keep the list of supplementary
986 * groups of the caller.
988 if (user
&& gid_is_valid(gid
) && gid
!= 0) {
989 /* First step, initialize groups from /etc/groups */
990 if (initgroups(user
, gid
) < 0)
996 if (strv_isempty(c
->supplementary_groups
))
1000 * If SupplementaryGroups= was passed then NGROUPS_MAX has to
1001 * be positive, otherwise fail.
1004 ngroups_max
= (int) sysconf(_SC_NGROUPS_MAX
);
1005 if (ngroups_max
<= 0) {
1009 return -EOPNOTSUPP
; /* For all other values */
1012 l_gids
= new(gid_t
, ngroups_max
);
1018 * Lookup the list of groups that the user belongs to, we
1019 * avoid NSS lookups here too for gid=0.
1022 if (getgrouplist(user
, gid
, l_gids
, &k
) < 0)
1027 STRV_FOREACH(i
, c
->supplementary_groups
) {
1030 if (k
>= ngroups_max
)
1034 r
= get_group_creds(&g
, l_gids
+k
, 0);
1042 * Sets ngids to zero to drop all supplementary groups, happens
1043 * when we are under root and SupplementaryGroups= is empty.
1050 /* Otherwise get the final list of supplementary groups */
1051 groups
= memdup(l_gids
, sizeof(gid_t
) * k
);
1055 *supplementary_gids
= groups
;
1063 static int enforce_groups(gid_t gid
, const gid_t
*supplementary_gids
, int ngids
) {
1066 /* Handle SupplementaryGroups= if it is not empty */
1068 r
= maybe_setgroups(ngids
, supplementary_gids
);
1073 if (gid_is_valid(gid
)) {
1074 /* Then set our gids */
1075 if (setresgid(gid
, gid
, gid
) < 0)
1082 static int enforce_user(const ExecContext
*context
, uid_t uid
) {
1085 if (!uid_is_valid(uid
))
1088 /* Sets (but doesn't look up) the uid and make sure we keep the
1089 * capabilities while doing so. */
1091 if (context
->capability_ambient_set
!= 0) {
1093 /* First step: If we need to keep capabilities but
1094 * drop privileges we need to make sure we keep our
1095 * caps, while we drop privileges. */
1097 int sb
= context
->secure_bits
| 1<<SECURE_KEEP_CAPS
;
1099 if (prctl(PR_GET_SECUREBITS
) != sb
)
1100 if (prctl(PR_SET_SECUREBITS
, sb
) < 0)
1105 /* Second step: actually set the uids */
1106 if (setresuid(uid
, uid
, uid
) < 0)
1109 /* At this point we should have all necessary capabilities but
1110 are otherwise a normal user. However, the caps might got
1111 corrupted due to the setresuid() so we need clean them up
1112 later. This is done outside of this call. */
1119 static int null_conv(
1121 const struct pam_message
**msg
,
1122 struct pam_response
**resp
,
1123 void *appdata_ptr
) {
1125 /* We don't support conversations */
1127 return PAM_CONV_ERR
;
1132 static int setup_pam(
1139 int fds
[], size_t n_fds
) {
1143 static const struct pam_conv conv
= {
1148 _cleanup_(barrier_destroy
) Barrier barrier
= BARRIER_NULL
;
1149 pam_handle_t
*handle
= NULL
;
1151 int pam_code
= PAM_SUCCESS
, r
;
1152 char **nv
, **e
= NULL
;
1153 bool close_session
= false;
1154 pid_t pam_pid
= 0, parent_pid
;
1161 /* We set up PAM in the parent process, then fork. The child
1162 * will then stay around until killed via PR_GET_PDEATHSIG or
1163 * systemd via the cgroup logic. It will then remove the PAM
1164 * session again. The parent process will exec() the actual
1165 * daemon. We do things this way to ensure that the main PID
1166 * of the daemon is the one we initially fork()ed. */
1168 r
= barrier_create(&barrier
);
1172 if (log_get_max_level() < LOG_DEBUG
)
1173 flags
|= PAM_SILENT
;
1175 pam_code
= pam_start(name
, user
, &conv
, &handle
);
1176 if (pam_code
!= PAM_SUCCESS
) {
1182 _cleanup_free_
char *q
= NULL
;
1184 /* Hmm, so no TTY was explicitly passed, but an fd passed to us directly might be a TTY. Let's figure
1185 * out if that's the case, and read the TTY off it. */
1187 if (getttyname_malloc(STDIN_FILENO
, &q
) >= 0)
1188 tty
= strjoina("/dev/", q
);
1192 pam_code
= pam_set_item(handle
, PAM_TTY
, tty
);
1193 if (pam_code
!= PAM_SUCCESS
)
1197 STRV_FOREACH(nv
, *env
) {
1198 pam_code
= pam_putenv(handle
, *nv
);
1199 if (pam_code
!= PAM_SUCCESS
)
1203 pam_code
= pam_acct_mgmt(handle
, flags
);
1204 if (pam_code
!= PAM_SUCCESS
)
1207 pam_code
= pam_open_session(handle
, flags
);
1208 if (pam_code
!= PAM_SUCCESS
)
1211 close_session
= true;
1213 e
= pam_getenvlist(handle
);
1215 pam_code
= PAM_BUF_ERR
;
1219 /* Block SIGTERM, so that we know that it won't get lost in
1222 assert_se(sigprocmask_many(SIG_BLOCK
, &old_ss
, SIGTERM
, -1) >= 0);
1224 parent_pid
= getpid_cached();
1226 r
= safe_fork("(sd-pam)", 0, &pam_pid
);
1230 int sig
, ret
= EXIT_PAM
;
1232 /* The child's job is to reset the PAM session on
1234 barrier_set_role(&barrier
, BARRIER_CHILD
);
1236 /* Make sure we don't keep open the passed fds in this child. We assume that otherwise only those fds
1237 * are open here that have been opened by PAM. */
1238 (void) close_many(fds
, n_fds
);
1240 /* Drop privileges - we don't need any to pam_close_session
1241 * and this will make PR_SET_PDEATHSIG work in most cases.
1242 * If this fails, ignore the error - but expect sd-pam threads
1243 * to fail to exit normally */
1245 r
= maybe_setgroups(0, NULL
);
1247 log_warning_errno(r
, "Failed to setgroups() in sd-pam: %m");
1248 if (setresgid(gid
, gid
, gid
) < 0)
1249 log_warning_errno(errno
, "Failed to setresgid() in sd-pam: %m");
1250 if (setresuid(uid
, uid
, uid
) < 0)
1251 log_warning_errno(errno
, "Failed to setresuid() in sd-pam: %m");
1253 (void) ignore_signals(SIGPIPE
, -1);
1255 /* Wait until our parent died. This will only work if
1256 * the above setresuid() succeeds, otherwise the kernel
1257 * will not allow unprivileged parents kill their privileged
1258 * children this way. We rely on the control groups kill logic
1259 * to do the rest for us. */
1260 if (prctl(PR_SET_PDEATHSIG
, SIGTERM
) < 0)
1263 /* Tell the parent that our setup is done. This is especially
1264 * important regarding dropping privileges. Otherwise, unit
1265 * setup might race against our setresuid(2) call.
1267 * If the parent aborted, we'll detect this below, hence ignore
1268 * return failure here. */
1269 (void) barrier_place(&barrier
);
1271 /* Check if our parent process might already have died? */
1272 if (getppid() == parent_pid
) {
1275 assert_se(sigemptyset(&ss
) >= 0);
1276 assert_se(sigaddset(&ss
, SIGTERM
) >= 0);
1279 if (sigwait(&ss
, &sig
) < 0) {
1286 assert(sig
== SIGTERM
);
1291 /* If our parent died we'll end the session */
1292 if (getppid() != parent_pid
) {
1293 pam_code
= pam_close_session(handle
, flags
);
1294 if (pam_code
!= PAM_SUCCESS
)
1301 pam_end(handle
, pam_code
| flags
);
1305 barrier_set_role(&barrier
, BARRIER_PARENT
);
1307 /* If the child was forked off successfully it will do all the
1308 * cleanups, so forget about the handle here. */
1311 /* Unblock SIGTERM again in the parent */
1312 assert_se(sigprocmask(SIG_SETMASK
, &old_ss
, NULL
) >= 0);
1314 /* We close the log explicitly here, since the PAM modules
1315 * might have opened it, but we don't want this fd around. */
1318 /* Synchronously wait for the child to initialize. We don't care for
1319 * errors as we cannot recover. However, warn loudly if it happens. */
1320 if (!barrier_place_and_sync(&barrier
))
1321 log_error("PAM initialization failed");
1323 return strv_free_and_replace(*env
, e
);
1326 if (pam_code
!= PAM_SUCCESS
) {
1327 log_error("PAM failed: %s", pam_strerror(handle
, pam_code
));
1328 r
= -EPERM
; /* PAM errors do not map to errno */
1330 log_error_errno(r
, "PAM failed: %m");
1334 pam_code
= pam_close_session(handle
, flags
);
1336 pam_end(handle
, pam_code
| flags
);
1348 static void rename_process_from_path(const char *path
) {
1349 char process_name
[11];
1353 /* This resulting string must fit in 10 chars (i.e. the length
1354 * of "/sbin/init") to look pretty in /bin/ps */
1358 rename_process("(...)");
1364 /* The end of the process name is usually more
1365 * interesting, since the first bit might just be
1371 process_name
[0] = '(';
1372 memcpy(process_name
+1, p
, l
);
1373 process_name
[1+l
] = ')';
1374 process_name
[1+l
+1] = 0;
1376 rename_process(process_name
);
1379 static bool context_has_address_families(const ExecContext
*c
) {
1382 return c
->address_families_whitelist
||
1383 !set_isempty(c
->address_families
);
1386 static bool context_has_syscall_filters(const ExecContext
*c
) {
1389 return c
->syscall_whitelist
||
1390 !hashmap_isempty(c
->syscall_filter
);
1393 static bool context_has_no_new_privileges(const ExecContext
*c
) {
1396 if (c
->no_new_privileges
)
1399 if (have_effective_cap(CAP_SYS_ADMIN
)) /* if we are privileged, we don't need NNP */
1402 /* We need NNP if we have any form of seccomp and are unprivileged */
1403 return context_has_address_families(c
) ||
1404 c
->memory_deny_write_execute
||
1405 c
->restrict_realtime
||
1406 c
->restrict_suid_sgid
||
1407 exec_context_restrict_namespaces_set(c
) ||
1408 c
->protect_kernel_tunables
||
1409 c
->protect_kernel_modules
||
1410 c
->private_devices
||
1411 context_has_syscall_filters(c
) ||
1412 !set_isempty(c
->syscall_archs
) ||
1413 c
->lock_personality
||
1414 c
->protect_hostname
;
1419 static bool skip_seccomp_unavailable(const Unit
* u
, const char* msg
) {
1421 if (is_seccomp_available())
1424 log_unit_debug(u
, "SECCOMP features not detected in the kernel, skipping %s", msg
);
1428 static int apply_syscall_filter(const Unit
* u
, const ExecContext
*c
, bool needs_ambient_hack
) {
1429 uint32_t negative_action
, default_action
, action
;
1435 if (!context_has_syscall_filters(c
))
1438 if (skip_seccomp_unavailable(u
, "SystemCallFilter="))
1441 negative_action
= c
->syscall_errno
== 0 ? scmp_act_kill_process() : SCMP_ACT_ERRNO(c
->syscall_errno
);
1443 if (c
->syscall_whitelist
) {
1444 default_action
= negative_action
;
1445 action
= SCMP_ACT_ALLOW
;
1447 default_action
= SCMP_ACT_ALLOW
;
1448 action
= negative_action
;
1451 if (needs_ambient_hack
) {
1452 r
= seccomp_filter_set_add(c
->syscall_filter
, c
->syscall_whitelist
, syscall_filter_sets
+ SYSCALL_FILTER_SET_SETUID
);
1457 return seccomp_load_syscall_filter_set_raw(default_action
, c
->syscall_filter
, action
, false);
1460 static int apply_syscall_archs(const Unit
*u
, const ExecContext
*c
) {
1464 if (set_isempty(c
->syscall_archs
))
1467 if (skip_seccomp_unavailable(u
, "SystemCallArchitectures="))
1470 return seccomp_restrict_archs(c
->syscall_archs
);
1473 static int apply_address_families(const Unit
* u
, const ExecContext
*c
) {
1477 if (!context_has_address_families(c
))
1480 if (skip_seccomp_unavailable(u
, "RestrictAddressFamilies="))
1483 return seccomp_restrict_address_families(c
->address_families
, c
->address_families_whitelist
);
1486 static int apply_memory_deny_write_execute(const Unit
* u
, const ExecContext
*c
) {
1490 if (!c
->memory_deny_write_execute
)
1493 if (skip_seccomp_unavailable(u
, "MemoryDenyWriteExecute="))
1496 return seccomp_memory_deny_write_execute();
1499 static int apply_restrict_realtime(const Unit
* u
, const ExecContext
*c
) {
1503 if (!c
->restrict_realtime
)
1506 if (skip_seccomp_unavailable(u
, "RestrictRealtime="))
1509 return seccomp_restrict_realtime();
1512 static int apply_restrict_suid_sgid(const Unit
* u
, const ExecContext
*c
) {
1516 if (!c
->restrict_suid_sgid
)
1519 if (skip_seccomp_unavailable(u
, "RestrictSUIDSGID="))
1522 return seccomp_restrict_suid_sgid();
1525 static int apply_protect_sysctl(const Unit
*u
, const ExecContext
*c
) {
1529 /* Turn off the legacy sysctl() system call. Many distributions turn this off while building the kernel, but
1530 * let's protect even those systems where this is left on in the kernel. */
1532 if (!c
->protect_kernel_tunables
)
1535 if (skip_seccomp_unavailable(u
, "ProtectKernelTunables="))
1538 return seccomp_protect_sysctl();
1541 static int apply_protect_kernel_modules(const Unit
*u
, const ExecContext
*c
) {
1545 /* Turn off module syscalls on ProtectKernelModules=yes */
1547 if (!c
->protect_kernel_modules
)
1550 if (skip_seccomp_unavailable(u
, "ProtectKernelModules="))
1553 return seccomp_load_syscall_filter_set(SCMP_ACT_ALLOW
, syscall_filter_sets
+ SYSCALL_FILTER_SET_MODULE
, SCMP_ACT_ERRNO(EPERM
), false);
1556 static int apply_private_devices(const Unit
*u
, const ExecContext
*c
) {
1560 /* If PrivateDevices= is set, also turn off iopl and all @raw-io syscalls. */
1562 if (!c
->private_devices
)
1565 if (skip_seccomp_unavailable(u
, "PrivateDevices="))
1568 return seccomp_load_syscall_filter_set(SCMP_ACT_ALLOW
, syscall_filter_sets
+ SYSCALL_FILTER_SET_RAW_IO
, SCMP_ACT_ERRNO(EPERM
), false);
1571 static int apply_restrict_namespaces(const Unit
*u
, const ExecContext
*c
) {
1575 if (!exec_context_restrict_namespaces_set(c
))
1578 if (skip_seccomp_unavailable(u
, "RestrictNamespaces="))
1581 return seccomp_restrict_namespaces(c
->restrict_namespaces
);
1584 static int apply_lock_personality(const Unit
* u
, const ExecContext
*c
) {
1585 unsigned long personality
;
1591 if (!c
->lock_personality
)
1594 if (skip_seccomp_unavailable(u
, "LockPersonality="))
1597 personality
= c
->personality
;
1599 /* If personality is not specified, use either PER_LINUX or PER_LINUX32 depending on what is currently set. */
1600 if (personality
== PERSONALITY_INVALID
) {
1602 r
= opinionated_personality(&personality
);
1607 return seccomp_lock_personality(personality
);
1612 static void do_idle_pipe_dance(int idle_pipe
[static 4]) {
1615 idle_pipe
[1] = safe_close(idle_pipe
[1]);
1616 idle_pipe
[2] = safe_close(idle_pipe
[2]);
1618 if (idle_pipe
[0] >= 0) {
1621 r
= fd_wait_for_event(idle_pipe
[0], POLLHUP
, IDLE_TIMEOUT_USEC
);
1623 if (idle_pipe
[3] >= 0 && r
== 0 /* timeout */) {
1626 /* Signal systemd that we are bored and want to continue. */
1627 n
= write(idle_pipe
[3], "x", 1);
1629 /* Wait for systemd to react to the signal above. */
1630 fd_wait_for_event(idle_pipe
[0], POLLHUP
, IDLE_TIMEOUT2_USEC
);
1633 idle_pipe
[0] = safe_close(idle_pipe
[0]);
1637 idle_pipe
[3] = safe_close(idle_pipe
[3]);
1640 static const char *exec_directory_env_name_to_string(ExecDirectoryType t
);
1642 static int build_environment(
1644 const ExecContext
*c
,
1645 const ExecParameters
*p
,
1648 const char *username
,
1650 dev_t journal_stream_dev
,
1651 ino_t journal_stream_ino
,
1654 _cleanup_strv_free_
char **our_env
= NULL
;
1655 ExecDirectoryType t
;
1664 our_env
= new0(char*, 14 + _EXEC_DIRECTORY_TYPE_MAX
);
1669 _cleanup_free_
char *joined
= NULL
;
1671 if (asprintf(&x
, "LISTEN_PID="PID_FMT
, getpid_cached()) < 0)
1673 our_env
[n_env
++] = x
;
1675 if (asprintf(&x
, "LISTEN_FDS=%zu", n_fds
) < 0)
1677 our_env
[n_env
++] = x
;
1679 joined
= strv_join(p
->fd_names
, ":");
1683 x
= strjoin("LISTEN_FDNAMES=", joined
);
1686 our_env
[n_env
++] = x
;
1689 if ((p
->flags
& EXEC_SET_WATCHDOG
) && p
->watchdog_usec
> 0) {
1690 if (asprintf(&x
, "WATCHDOG_PID="PID_FMT
, getpid_cached()) < 0)
1692 our_env
[n_env
++] = x
;
1694 if (asprintf(&x
, "WATCHDOG_USEC="USEC_FMT
, p
->watchdog_usec
) < 0)
1696 our_env
[n_env
++] = x
;
1699 /* If this is D-Bus, tell the nss-systemd module, since it relies on being able to use D-Bus look up dynamic
1700 * users via PID 1, possibly dead-locking the dbus daemon. This way it will not use D-Bus to resolve names, but
1701 * check the database directly. */
1702 if (p
->flags
& EXEC_NSS_BYPASS_BUS
) {
1703 x
= strdup("SYSTEMD_NSS_BYPASS_BUS=1");
1706 our_env
[n_env
++] = x
;
1710 x
= strappend("HOME=", home
);
1714 path_simplify(x
+ 5, true);
1715 our_env
[n_env
++] = x
;
1719 x
= strappend("LOGNAME=", username
);
1722 our_env
[n_env
++] = x
;
1724 x
= strappend("USER=", username
);
1727 our_env
[n_env
++] = x
;
1731 x
= strappend("SHELL=", shell
);
1735 path_simplify(x
+ 6, true);
1736 our_env
[n_env
++] = x
;
1739 if (!sd_id128_is_null(u
->invocation_id
)) {
1740 if (asprintf(&x
, "INVOCATION_ID=" SD_ID128_FORMAT_STR
, SD_ID128_FORMAT_VAL(u
->invocation_id
)) < 0)
1743 our_env
[n_env
++] = x
;
1746 if (exec_context_needs_term(c
)) {
1747 const char *tty_path
, *term
= NULL
;
1749 tty_path
= exec_context_tty_path(c
);
1751 /* If we are forked off PID 1 and we are supposed to operate on /dev/console, then let's try to inherit
1752 * the $TERM set for PID 1. This is useful for containers so that the $TERM the container manager
1753 * passes to PID 1 ends up all the way in the console login shown. */
1755 if (path_equal(tty_path
, "/dev/console") && getppid() == 1)
1756 term
= getenv("TERM");
1758 term
= default_term_for_tty(tty_path
);
1760 x
= strappend("TERM=", term
);
1763 our_env
[n_env
++] = x
;
1766 if (journal_stream_dev
!= 0 && journal_stream_ino
!= 0) {
1767 if (asprintf(&x
, "JOURNAL_STREAM=" DEV_FMT
":" INO_FMT
, journal_stream_dev
, journal_stream_ino
) < 0)
1770 our_env
[n_env
++] = x
;
1773 for (t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
1774 _cleanup_free_
char *pre
= NULL
, *joined
= NULL
;
1780 if (strv_isempty(c
->directories
[t
].paths
))
1783 n
= exec_directory_env_name_to_string(t
);
1787 pre
= strjoin(p
->prefix
[t
], "/");
1791 joined
= strv_join_prefix(c
->directories
[t
].paths
, ":", pre
);
1795 x
= strjoin(n
, "=", joined
);
1799 our_env
[n_env
++] = x
;
1802 our_env
[n_env
++] = NULL
;
1803 assert(n_env
<= 14 + _EXEC_DIRECTORY_TYPE_MAX
);
1805 *ret
= TAKE_PTR(our_env
);
1810 static int build_pass_environment(const ExecContext
*c
, char ***ret
) {
1811 _cleanup_strv_free_
char **pass_env
= NULL
;
1812 size_t n_env
= 0, n_bufsize
= 0;
1815 STRV_FOREACH(i
, c
->pass_environment
) {
1816 _cleanup_free_
char *x
= NULL
;
1822 x
= strjoin(*i
, "=", v
);
1826 if (!GREEDY_REALLOC(pass_env
, n_bufsize
, n_env
+ 2))
1829 pass_env
[n_env
++] = TAKE_PTR(x
);
1830 pass_env
[n_env
] = NULL
;
1833 *ret
= TAKE_PTR(pass_env
);
1838 static bool exec_needs_mount_namespace(
1839 const ExecContext
*context
,
1840 const ExecParameters
*params
,
1841 const ExecRuntime
*runtime
) {
1846 if (context
->root_image
)
1849 if (!strv_isempty(context
->read_write_paths
) ||
1850 !strv_isempty(context
->read_only_paths
) ||
1851 !strv_isempty(context
->inaccessible_paths
))
1854 if (context
->n_bind_mounts
> 0)
1857 if (context
->n_temporary_filesystems
> 0)
1860 if (!IN_SET(context
->mount_flags
, 0, MS_SHARED
))
1863 if (context
->private_tmp
&& runtime
&& (runtime
->tmp_dir
|| runtime
->var_tmp_dir
))
1866 if (context
->private_devices
||
1867 context
->private_mounts
||
1868 context
->protect_system
!= PROTECT_SYSTEM_NO
||
1869 context
->protect_home
!= PROTECT_HOME_NO
||
1870 context
->protect_kernel_tunables
||
1871 context
->protect_kernel_modules
||
1872 context
->protect_control_groups
)
1875 if (context
->root_directory
) {
1876 ExecDirectoryType t
;
1878 if (context
->mount_apivfs
)
1881 for (t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
1882 if (!params
->prefix
[t
])
1885 if (!strv_isempty(context
->directories
[t
].paths
))
1890 if (context
->dynamic_user
&&
1891 (!strv_isempty(context
->directories
[EXEC_DIRECTORY_STATE
].paths
) ||
1892 !strv_isempty(context
->directories
[EXEC_DIRECTORY_CACHE
].paths
) ||
1893 !strv_isempty(context
->directories
[EXEC_DIRECTORY_LOGS
].paths
)))
1899 static int setup_private_users(uid_t uid
, gid_t gid
) {
1900 _cleanup_free_
char *uid_map
= NULL
, *gid_map
= NULL
;
1901 _cleanup_close_pair_
int errno_pipe
[2] = { -1, -1 };
1902 _cleanup_close_
int unshare_ready_fd
= -1;
1903 _cleanup_(sigkill_waitp
) pid_t pid
= 0;
1908 /* Set up a user namespace and map root to root, the selected UID/GID to itself, and everything else to
1909 * nobody. In order to be able to write this mapping we need CAP_SETUID in the original user namespace, which
1910 * we however lack after opening the user namespace. To work around this we fork() a temporary child process,
1911 * which waits for the parent to create the new user namespace while staying in the original namespace. The
1912 * child then writes the UID mapping, under full privileges. The parent waits for the child to finish and
1913 * continues execution normally. */
1915 if (uid
!= 0 && uid_is_valid(uid
)) {
1916 r
= asprintf(&uid_map
,
1917 "0 0 1\n" /* Map root → root */
1918 UID_FMT
" " UID_FMT
" 1\n", /* Map $UID → $UID */
1923 uid_map
= strdup("0 0 1\n"); /* The case where the above is the same */
1928 if (gid
!= 0 && gid_is_valid(gid
)) {
1929 r
= asprintf(&gid_map
,
1930 "0 0 1\n" /* Map root → root */
1931 GID_FMT
" " GID_FMT
" 1\n", /* Map $GID → $GID */
1936 gid_map
= strdup("0 0 1\n"); /* The case where the above is the same */
1941 /* Create a communication channel so that the parent can tell the child when it finished creating the user
1943 unshare_ready_fd
= eventfd(0, EFD_CLOEXEC
);
1944 if (unshare_ready_fd
< 0)
1947 /* Create a communication channel so that the child can tell the parent a proper error code in case it
1949 if (pipe2(errno_pipe
, O_CLOEXEC
) < 0)
1952 r
= safe_fork("(sd-userns)", FORK_RESET_SIGNALS
|FORK_DEATHSIG
, &pid
);
1956 _cleanup_close_
int fd
= -1;
1960 /* Child process, running in the original user namespace. Let's update the parent's UID/GID map from
1961 * here, after the parent opened its own user namespace. */
1964 errno_pipe
[0] = safe_close(errno_pipe
[0]);
1966 /* Wait until the parent unshared the user namespace */
1967 if (read(unshare_ready_fd
, &c
, sizeof(c
)) < 0) {
1972 /* Disable the setgroups() system call in the child user namespace, for good. */
1973 a
= procfs_file_alloca(ppid
, "setgroups");
1974 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
1976 if (errno
!= ENOENT
) {
1981 /* If the file is missing the kernel is too old, let's continue anyway. */
1983 if (write(fd
, "deny\n", 5) < 0) {
1988 fd
= safe_close(fd
);
1991 /* First write the GID map */
1992 a
= procfs_file_alloca(ppid
, "gid_map");
1993 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
1998 if (write(fd
, gid_map
, strlen(gid_map
)) < 0) {
2002 fd
= safe_close(fd
);
2004 /* The write the UID map */
2005 a
= procfs_file_alloca(ppid
, "uid_map");
2006 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
2011 if (write(fd
, uid_map
, strlen(uid_map
)) < 0) {
2016 _exit(EXIT_SUCCESS
);
2019 (void) write(errno_pipe
[1], &r
, sizeof(r
));
2020 _exit(EXIT_FAILURE
);
2023 errno_pipe
[1] = safe_close(errno_pipe
[1]);
2025 if (unshare(CLONE_NEWUSER
) < 0)
2028 /* Let the child know that the namespace is ready now */
2029 if (write(unshare_ready_fd
, &c
, sizeof(c
)) < 0)
2032 /* Try to read an error code from the child */
2033 n
= read(errno_pipe
[0], &r
, sizeof(r
));
2036 if (n
== sizeof(r
)) { /* an error code was sent to us */
2041 if (n
!= 0) /* on success we should have read 0 bytes */
2044 r
= wait_for_terminate_and_check("(sd-userns)", pid
, 0);
2048 if (r
!= EXIT_SUCCESS
) /* If something strange happened with the child, let's consider this fatal, too */
2054 static int setup_exec_directory(
2055 const ExecContext
*context
,
2056 const ExecParameters
*params
,
2059 ExecDirectoryType type
,
2062 static const int exit_status_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
2063 [EXEC_DIRECTORY_RUNTIME
] = EXIT_RUNTIME_DIRECTORY
,
2064 [EXEC_DIRECTORY_STATE
] = EXIT_STATE_DIRECTORY
,
2065 [EXEC_DIRECTORY_CACHE
] = EXIT_CACHE_DIRECTORY
,
2066 [EXEC_DIRECTORY_LOGS
] = EXIT_LOGS_DIRECTORY
,
2067 [EXEC_DIRECTORY_CONFIGURATION
] = EXIT_CONFIGURATION_DIRECTORY
,
2074 assert(type
>= 0 && type
< _EXEC_DIRECTORY_TYPE_MAX
);
2075 assert(exit_status
);
2077 if (!params
->prefix
[type
])
2080 if (params
->flags
& EXEC_CHOWN_DIRECTORIES
) {
2081 if (!uid_is_valid(uid
))
2083 if (!gid_is_valid(gid
))
2087 STRV_FOREACH(rt
, context
->directories
[type
].paths
) {
2088 _cleanup_free_
char *p
= NULL
, *pp
= NULL
;
2090 p
= path_join(params
->prefix
[type
], *rt
);
2096 r
= mkdir_parents_label(p
, 0755);
2100 if (context
->dynamic_user
&&
2101 (!IN_SET(type
, EXEC_DIRECTORY_RUNTIME
, EXEC_DIRECTORY_CONFIGURATION
) ||
2102 (type
== EXEC_DIRECTORY_RUNTIME
&& context
->runtime_directory_preserve_mode
!= EXEC_PRESERVE_NO
))) {
2103 _cleanup_free_
char *private_root
= NULL
;
2105 /* So, here's one extra complication when dealing with DynamicUser=1 units. In that case we
2106 * want to avoid leaving a directory around fully accessible that is owned by a dynamic user
2107 * whose UID is later on reused. To lock this down we use the same trick used by container
2108 * managers to prohibit host users to get access to files of the same UID in containers: we
2109 * place everything inside a directory that has an access mode of 0700 and is owned root:root,
2110 * so that it acts as security boundary for unprivileged host code. We then use fs namespacing
2111 * to make this directory permeable for the service itself.
2113 * Specifically: for a service which wants a special directory "foo/" we first create a
2114 * directory "private/" with access mode 0700 owned by root:root. Then we place "foo" inside of
2115 * that directory (i.e. "private/foo/"), and make "foo" a symlink to "private/foo". This way,
2116 * privileged host users can access "foo/" as usual, but unprivileged host users can't look
2117 * into it. Inside of the namespaceof the container "private/" is replaced by a more liberally
2118 * accessible tmpfs, into which the host's "private/foo/" is mounted under the same name, thus
2119 * disabling the access boundary for the service and making sure it only gets access to the
2120 * dirs it needs but no others. Tricky? Yes, absolutely, but it works!
2122 * Note that we don't do this for EXEC_DIRECTORY_CONFIGURATION as that's assumed not to be
2123 * owned by the service itself.
2124 * Also, note that we don't do this for EXEC_DIRECTORY_RUNTIME as that's often used for sharing
2125 * files or sockets with other services. */
2127 private_root
= path_join(params
->prefix
[type
], "private");
2128 if (!private_root
) {
2133 /* First set up private root if it doesn't exist yet, with access mode 0700 and owned by root:root */
2134 r
= mkdir_safe_label(private_root
, 0700, 0, 0, MKDIR_WARN_MODE
);
2138 pp
= path_join(private_root
, *rt
);
2144 /* Create all directories between the configured directory and this private root, and mark them 0755 */
2145 r
= mkdir_parents_label(pp
, 0755);
2149 if (is_dir(p
, false) > 0 &&
2150 (laccess(pp
, F_OK
) < 0 && errno
== ENOENT
)) {
2152 /* Hmm, the private directory doesn't exist yet, but the normal one exists? If so, move
2153 * it over. Most likely the service has been upgraded from one that didn't use
2154 * DynamicUser=1, to one that does. */
2156 if (rename(p
, pp
) < 0) {
2161 /* Otherwise, create the actual directory for the service */
2163 r
= mkdir_label(pp
, context
->directories
[type
].mode
);
2164 if (r
< 0 && r
!= -EEXIST
)
2168 /* And link it up from the original place */
2169 r
= symlink_idempotent(pp
, p
, true);
2174 r
= mkdir_label(p
, context
->directories
[type
].mode
);
2179 if (type
== EXEC_DIRECTORY_CONFIGURATION
) {
2182 /* Don't change the owner/access mode of the configuration directory,
2183 * as in the common case it is not written to by a service, and shall
2184 * not be writable. */
2186 if (stat(p
, &st
) < 0) {
2191 /* Still complain if the access mode doesn't match */
2192 if (((st
.st_mode
^ context
->directories
[type
].mode
) & 07777) != 0)
2193 log_warning("%s \'%s\' already exists but the mode is different. "
2194 "(File system: %o %sMode: %o)",
2195 exec_directory_type_to_string(type
), *rt
,
2196 st
.st_mode
& 07777, exec_directory_type_to_string(type
), context
->directories
[type
].mode
& 07777);
2203 /* Lock down the access mode (we use chmod_and_chown() to make this idempotent. We don't
2204 * specify UID/GID here, so that path_chown_recursive() can optimize things depending on the
2205 * current UID/GID ownership.) */
2206 r
= chmod_and_chown(pp
?: p
, context
->directories
[type
].mode
, UID_INVALID
, GID_INVALID
);
2210 /* Then, change the ownership of the whole tree, if necessary. When dynamic users are used we
2211 * drop the suid/sgid bits, since we really don't want SUID/SGID files for dynamic UID/GID
2212 * assignments to exist.*/
2213 r
= path_chown_recursive(pp
?: p
, uid
, gid
, context
->dynamic_user
? 01777 : 07777);
2221 *exit_status
= exit_status_table
[type
];
2226 static int setup_smack(
2227 const ExecContext
*context
,
2228 const ExecCommand
*command
) {
2235 if (context
->smack_process_label
) {
2236 r
= mac_smack_apply_pid(0, context
->smack_process_label
);
2240 #ifdef SMACK_DEFAULT_PROCESS_LABEL
2242 _cleanup_free_
char *exec_label
= NULL
;
2244 r
= mac_smack_read(command
->path
, SMACK_ATTR_EXEC
, &exec_label
);
2245 if (r
< 0 && !IN_SET(r
, -ENODATA
, -EOPNOTSUPP
))
2248 r
= mac_smack_apply_pid(0, exec_label
? : SMACK_DEFAULT_PROCESS_LABEL
);
2258 static int compile_bind_mounts(
2259 const ExecContext
*context
,
2260 const ExecParameters
*params
,
2261 BindMount
**ret_bind_mounts
,
2262 size_t *ret_n_bind_mounts
,
2263 char ***ret_empty_directories
) {
2265 _cleanup_strv_free_
char **empty_directories
= NULL
;
2266 BindMount
*bind_mounts
;
2268 ExecDirectoryType t
;
2273 assert(ret_bind_mounts
);
2274 assert(ret_n_bind_mounts
);
2275 assert(ret_empty_directories
);
2277 n
= context
->n_bind_mounts
;
2278 for (t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
2279 if (!params
->prefix
[t
])
2282 n
+= strv_length(context
->directories
[t
].paths
);
2286 *ret_bind_mounts
= NULL
;
2287 *ret_n_bind_mounts
= 0;
2288 *ret_empty_directories
= NULL
;
2292 bind_mounts
= new(BindMount
, n
);
2296 for (i
= 0; i
< context
->n_bind_mounts
; i
++) {
2297 BindMount
*item
= context
->bind_mounts
+ i
;
2300 s
= strdup(item
->source
);
2306 d
= strdup(item
->destination
);
2313 bind_mounts
[h
++] = (BindMount
) {
2316 .read_only
= item
->read_only
,
2317 .recursive
= item
->recursive
,
2318 .ignore_enoent
= item
->ignore_enoent
,
2322 for (t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
2325 if (!params
->prefix
[t
])
2328 if (strv_isempty(context
->directories
[t
].paths
))
2331 if (context
->dynamic_user
&&
2332 !IN_SET(t
, EXEC_DIRECTORY_RUNTIME
, EXEC_DIRECTORY_CONFIGURATION
) &&
2333 !(context
->root_directory
|| context
->root_image
)) {
2336 /* So this is for a dynamic user, and we need to make sure the process can access its own
2337 * directory. For that we overmount the usually inaccessible "private" subdirectory with a
2338 * tmpfs that makes it accessible and is empty except for the submounts we do this for. */
2340 private_root
= strjoin(params
->prefix
[t
], "/private");
2341 if (!private_root
) {
2346 r
= strv_consume(&empty_directories
, private_root
);
2351 STRV_FOREACH(suffix
, context
->directories
[t
].paths
) {
2354 if (context
->dynamic_user
&&
2355 !IN_SET(t
, EXEC_DIRECTORY_RUNTIME
, EXEC_DIRECTORY_CONFIGURATION
))
2356 s
= strjoin(params
->prefix
[t
], "/private/", *suffix
);
2358 s
= strjoin(params
->prefix
[t
], "/", *suffix
);
2364 if (context
->dynamic_user
&&
2365 !IN_SET(t
, EXEC_DIRECTORY_RUNTIME
, EXEC_DIRECTORY_CONFIGURATION
) &&
2366 (context
->root_directory
|| context
->root_image
))
2367 /* When RootDirectory= or RootImage= are set, then the symbolic link to the private
2368 * directory is not created on the root directory. So, let's bind-mount the directory
2369 * on the 'non-private' place. */
2370 d
= strjoin(params
->prefix
[t
], "/", *suffix
);
2379 bind_mounts
[h
++] = (BindMount
) {
2383 .nosuid
= context
->dynamic_user
, /* don't allow suid/sgid when DynamicUser= is on */
2385 .ignore_enoent
= false,
2392 *ret_bind_mounts
= bind_mounts
;
2393 *ret_n_bind_mounts
= n
;
2394 *ret_empty_directories
= TAKE_PTR(empty_directories
);
2399 bind_mount_free_many(bind_mounts
, h
);
2403 static int apply_mount_namespace(
2405 const ExecCommand
*command
,
2406 const ExecContext
*context
,
2407 const ExecParameters
*params
,
2408 const ExecRuntime
*runtime
,
2409 char **error_path
) {
2411 _cleanup_strv_free_
char **empty_directories
= NULL
;
2412 char *tmp
= NULL
, *var
= NULL
;
2413 const char *root_dir
= NULL
, *root_image
= NULL
;
2414 NamespaceInfo ns_info
;
2415 bool needs_sandboxing
;
2416 BindMount
*bind_mounts
= NULL
;
2417 size_t n_bind_mounts
= 0;
2422 /* The runtime struct only contains the parent of the private /tmp,
2423 * which is non-accessible to world users. Inside of it there's a /tmp
2424 * that is sticky, and that's the one we want to use here. */
2426 if (context
->private_tmp
&& runtime
) {
2427 if (runtime
->tmp_dir
)
2428 tmp
= strjoina(runtime
->tmp_dir
, "/tmp");
2429 if (runtime
->var_tmp_dir
)
2430 var
= strjoina(runtime
->var_tmp_dir
, "/tmp");
2433 if (params
->flags
& EXEC_APPLY_CHROOT
) {
2434 root_image
= context
->root_image
;
2437 root_dir
= context
->root_directory
;
2440 r
= compile_bind_mounts(context
, params
, &bind_mounts
, &n_bind_mounts
, &empty_directories
);
2444 needs_sandboxing
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && !(command
->flags
& EXEC_COMMAND_FULLY_PRIVILEGED
);
2445 if (needs_sandboxing
)
2446 ns_info
= (NamespaceInfo
) {
2447 .ignore_protect_paths
= false,
2448 .private_dev
= context
->private_devices
,
2449 .protect_control_groups
= context
->protect_control_groups
,
2450 .protect_kernel_tunables
= context
->protect_kernel_tunables
,
2451 .protect_kernel_modules
= context
->protect_kernel_modules
,
2452 .protect_hostname
= context
->protect_hostname
,
2453 .mount_apivfs
= context
->mount_apivfs
,
2454 .private_mounts
= context
->private_mounts
,
2456 else if (!context
->dynamic_user
&& root_dir
)
2458 * If DynamicUser=no and RootDirectory= is set then lets pass a relaxed
2459 * sandbox info, otherwise enforce it, don't ignore protected paths and
2460 * fail if we are enable to apply the sandbox inside the mount namespace.
2462 ns_info
= (NamespaceInfo
) {
2463 .ignore_protect_paths
= true,
2466 ns_info
= (NamespaceInfo
) {};
2468 if (context
->mount_flags
== MS_SHARED
)
2469 log_unit_debug(u
, "shared mount propagation hidden by other fs namespacing unit settings: ignoring");
2471 r
= setup_namespace(root_dir
, root_image
,
2472 &ns_info
, context
->read_write_paths
,
2473 needs_sandboxing
? context
->read_only_paths
: NULL
,
2474 needs_sandboxing
? context
->inaccessible_paths
: NULL
,
2478 context
->temporary_filesystems
,
2479 context
->n_temporary_filesystems
,
2482 needs_sandboxing
? context
->protect_home
: PROTECT_HOME_NO
,
2483 needs_sandboxing
? context
->protect_system
: PROTECT_SYSTEM_NO
,
2484 context
->mount_flags
,
2485 DISSECT_IMAGE_DISCARD_ON_LOOP
,
2488 bind_mount_free_many(bind_mounts
, n_bind_mounts
);
2490 /* If we couldn't set up the namespace this is probably due to a missing capability. setup_namespace() reports
2491 * that with a special, recognizable error ENOANO. In this case, silently proceed, but only if exclusively
2492 * sandboxing options were used, i.e. nothing such as RootDirectory= or BindMount= that would result in a
2493 * completely different execution environment. */
2495 if (n_bind_mounts
== 0 &&
2496 context
->n_temporary_filesystems
== 0 &&
2497 !root_dir
&& !root_image
&&
2498 !context
->dynamic_user
) {
2499 log_unit_debug(u
, "Failed to set up namespace, assuming containerized execution and ignoring.");
2503 log_unit_debug(u
, "Failed to set up namespace, and refusing to continue since the selected namespacing options alter mount environment non-trivially.\n"
2504 "Bind mounts: %zu, temporary filesystems: %zu, root directory: %s, root image: %s, dynamic user: %s",
2505 n_bind_mounts
, context
->n_temporary_filesystems
, yes_no(root_dir
), yes_no(root_image
), yes_no(context
->dynamic_user
));
2513 static int apply_working_directory(
2514 const ExecContext
*context
,
2515 const ExecParameters
*params
,
2517 const bool needs_mount_ns
,
2523 assert(exit_status
);
2525 if (context
->working_directory_home
) {
2528 *exit_status
= EXIT_CHDIR
;
2534 } else if (context
->working_directory
)
2535 wd
= context
->working_directory
;
2539 if (params
->flags
& EXEC_APPLY_CHROOT
) {
2540 if (!needs_mount_ns
&& context
->root_directory
)
2541 if (chroot(context
->root_directory
) < 0) {
2542 *exit_status
= EXIT_CHROOT
;
2548 d
= prefix_roota(context
->root_directory
, wd
);
2550 if (chdir(d
) < 0 && !context
->working_directory_missing_ok
) {
2551 *exit_status
= EXIT_CHDIR
;
2558 static int setup_keyring(
2560 const ExecContext
*context
,
2561 const ExecParameters
*p
,
2562 uid_t uid
, gid_t gid
) {
2564 key_serial_t keyring
;
2573 /* Let's set up a new per-service "session" kernel keyring for each system service. This has the benefit that
2574 * each service runs with its own keyring shared among all processes of the service, but with no hook-up beyond
2575 * that scope, and in particular no link to the per-UID keyring. If we don't do this the keyring will be
2576 * automatically created on-demand and then linked to the per-UID keyring, by the kernel. The kernel's built-in
2577 * on-demand behaviour is very appropriate for login users, but probably not so much for system services, where
2578 * UIDs are not necessarily specific to a service but reused (at least in the case of UID 0). */
2580 if (context
->keyring_mode
== EXEC_KEYRING_INHERIT
)
2583 /* Acquiring a reference to the user keyring is nasty. We briefly change identity in order to get things set up
2584 * properly by the kernel. If we don't do that then we can't create it atomically, and that sucks for parallel
2585 * execution. This mimics what pam_keyinit does, too. Setting up session keyring, to be owned by the right user
2586 * & group is just as nasty as acquiring a reference to the user keyring. */
2588 saved_uid
= getuid();
2589 saved_gid
= getgid();
2591 if (gid_is_valid(gid
) && gid
!= saved_gid
) {
2592 if (setregid(gid
, -1) < 0)
2593 return log_unit_error_errno(u
, errno
, "Failed to change GID for user keyring: %m");
2596 if (uid_is_valid(uid
) && uid
!= saved_uid
) {
2597 if (setreuid(uid
, -1) < 0) {
2598 r
= log_unit_error_errno(u
, errno
, "Failed to change UID for user keyring: %m");
2603 keyring
= keyctl(KEYCTL_JOIN_SESSION_KEYRING
, 0, 0, 0, 0);
2604 if (keyring
== -1) {
2605 if (errno
== ENOSYS
)
2606 log_unit_debug_errno(u
, errno
, "Kernel keyring not supported, ignoring.");
2607 else if (IN_SET(errno
, EACCES
, EPERM
))
2608 log_unit_debug_errno(u
, errno
, "Kernel keyring access prohibited, ignoring.");
2609 else if (errno
== EDQUOT
)
2610 log_unit_debug_errno(u
, errno
, "Out of kernel keyrings to allocate, ignoring.");
2612 r
= log_unit_error_errno(u
, errno
, "Setting up kernel keyring failed: %m");
2617 /* When requested link the user keyring into the session keyring. */
2618 if (context
->keyring_mode
== EXEC_KEYRING_SHARED
) {
2620 if (keyctl(KEYCTL_LINK
,
2621 KEY_SPEC_USER_KEYRING
,
2622 KEY_SPEC_SESSION_KEYRING
, 0, 0) < 0) {
2623 r
= log_unit_error_errno(u
, errno
, "Failed to link user keyring into session keyring: %m");
2628 /* Restore uid/gid back */
2629 if (uid_is_valid(uid
) && uid
!= saved_uid
) {
2630 if (setreuid(saved_uid
, -1) < 0) {
2631 r
= log_unit_error_errno(u
, errno
, "Failed to change UID back for user keyring: %m");
2636 if (gid_is_valid(gid
) && gid
!= saved_gid
) {
2637 if (setregid(saved_gid
, -1) < 0)
2638 return log_unit_error_errno(u
, errno
, "Failed to change GID back for user keyring: %m");
2641 /* Populate they keyring with the invocation ID by default, as original saved_uid. */
2642 if (!sd_id128_is_null(u
->invocation_id
)) {
2645 key
= add_key("user", "invocation_id", &u
->invocation_id
, sizeof(u
->invocation_id
), KEY_SPEC_SESSION_KEYRING
);
2647 log_unit_debug_errno(u
, errno
, "Failed to add invocation ID to keyring, ignoring: %m");
2649 if (keyctl(KEYCTL_SETPERM
, key
,
2650 KEY_POS_VIEW
|KEY_POS_READ
|KEY_POS_SEARCH
|
2651 KEY_USR_VIEW
|KEY_USR_READ
|KEY_USR_SEARCH
, 0, 0) < 0)
2652 r
= log_unit_error_errno(u
, errno
, "Failed to restrict invocation ID permission: %m");
2657 /* Revert back uid & gid for the the last time, and exit */
2658 /* no extra logging, as only the first already reported error matters */
2659 if (getuid() != saved_uid
)
2660 (void) setreuid(saved_uid
, -1);
2662 if (getgid() != saved_gid
)
2663 (void) setregid(saved_gid
, -1);
2668 static void append_socket_pair(int *array
, size_t *n
, const int pair
[static 2]) {
2676 array
[(*n
)++] = pair
[0];
2678 array
[(*n
)++] = pair
[1];
2681 static int close_remaining_fds(
2682 const ExecParameters
*params
,
2683 const ExecRuntime
*runtime
,
2684 const DynamicCreds
*dcreds
,
2688 int *fds
, size_t n_fds
) {
2690 size_t n_dont_close
= 0;
2691 int dont_close
[n_fds
+ 12];
2695 if (params
->stdin_fd
>= 0)
2696 dont_close
[n_dont_close
++] = params
->stdin_fd
;
2697 if (params
->stdout_fd
>= 0)
2698 dont_close
[n_dont_close
++] = params
->stdout_fd
;
2699 if (params
->stderr_fd
>= 0)
2700 dont_close
[n_dont_close
++] = params
->stderr_fd
;
2703 dont_close
[n_dont_close
++] = socket_fd
;
2705 dont_close
[n_dont_close
++] = exec_fd
;
2707 memcpy(dont_close
+ n_dont_close
, fds
, sizeof(int) * n_fds
);
2708 n_dont_close
+= n_fds
;
2712 append_socket_pair(dont_close
, &n_dont_close
, runtime
->netns_storage_socket
);
2716 append_socket_pair(dont_close
, &n_dont_close
, dcreds
->user
->storage_socket
);
2718 append_socket_pair(dont_close
, &n_dont_close
, dcreds
->group
->storage_socket
);
2721 if (user_lookup_fd
>= 0)
2722 dont_close
[n_dont_close
++] = user_lookup_fd
;
2724 return close_all_fds(dont_close
, n_dont_close
);
2727 static int send_user_lookup(
2735 /* Send the resolved UID/GID to PID 1 after we learnt it. We send a single datagram, containing the UID/GID
2736 * data as well as the unit name. Note that we suppress sending this if no user/group to resolve was
2739 if (user_lookup_fd
< 0)
2742 if (!uid_is_valid(uid
) && !gid_is_valid(gid
))
2745 if (writev(user_lookup_fd
,
2747 IOVEC_INIT(&uid
, sizeof(uid
)),
2748 IOVEC_INIT(&gid
, sizeof(gid
)),
2749 IOVEC_INIT_STRING(unit
->id
) }, 3) < 0)
2755 static int acquire_home(const ExecContext
*c
, uid_t uid
, const char** home
, char **buf
) {
2762 /* If WorkingDirectory=~ is set, try to acquire a usable home directory. */
2767 if (!c
->working_directory_home
)
2770 r
= get_home_dir(buf
);
2778 static int compile_suggested_paths(const ExecContext
*c
, const ExecParameters
*p
, char ***ret
) {
2779 _cleanup_strv_free_
char ** list
= NULL
;
2780 ExecDirectoryType t
;
2787 assert(c
->dynamic_user
);
2789 /* Compile a list of paths that it might make sense to read the owning UID from to use as initial candidate for
2790 * dynamic UID allocation, in order to save us from doing costly recursive chown()s of the special
2793 for (t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
2796 if (t
== EXEC_DIRECTORY_CONFIGURATION
)
2802 STRV_FOREACH(i
, c
->directories
[t
].paths
) {
2805 if (t
== EXEC_DIRECTORY_RUNTIME
)
2806 e
= strjoin(p
->prefix
[t
], "/", *i
);
2808 e
= strjoin(p
->prefix
[t
], "/private/", *i
);
2812 r
= strv_consume(&list
, e
);
2818 *ret
= TAKE_PTR(list
);
2823 static char *exec_command_line(char **argv
);
2825 static int exec_parameters_get_cgroup_path(const ExecParameters
*params
, char **ret
) {
2826 bool using_subcgroup
;
2832 if (!params
->cgroup_path
)
2835 /* If we are called for a unit where cgroup delegation is on, and the payload created its own populated
2836 * subcgroup (which we expect it to do, after all it asked for delegation), then we cannot place the control
2837 * processes started after the main unit's process in the unit's main cgroup because it is now an inner one,
2838 * and inner cgroups may not contain processes. Hence, if delegation is on, and this is a control process,
2839 * let's use ".control" as subcgroup instead. Note that we do so only for ExecStartPost=, ExecReload=,
2840 * ExecStop=, ExecStopPost=, i.e. for the commands where the main process is already forked. For ExecStartPre=
2841 * this is not necessary, the cgroup is still empty. We distinguish these cases with the EXEC_CONTROL_CGROUP
2842 * flag, which is only passed for the former statements, not for the latter. */
2844 using_subcgroup
= FLAGS_SET(params
->flags
, EXEC_CONTROL_CGROUP
|EXEC_CGROUP_DELEGATE
|EXEC_IS_CONTROL
);
2845 if (using_subcgroup
)
2846 p
= strjoin(params
->cgroup_path
, "/.control");
2848 p
= strdup(params
->cgroup_path
);
2853 return using_subcgroup
;
2856 static int exec_child(
2858 const ExecCommand
*command
,
2859 const ExecContext
*context
,
2860 const ExecParameters
*params
,
2861 ExecRuntime
*runtime
,
2862 DynamicCreds
*dcreds
,
2866 size_t n_socket_fds
,
2867 size_t n_storage_fds
,
2872 _cleanup_strv_free_
char **our_env
= NULL
, **pass_env
= NULL
, **accum_env
= NULL
, **replaced_argv
= NULL
;
2873 int *fds_with_exec_fd
, n_fds_with_exec_fd
, r
, ngids
= 0, exec_fd
= -1;
2874 _cleanup_free_ gid_t
*supplementary_gids
= NULL
;
2875 const char *username
= NULL
, *groupname
= NULL
;
2876 _cleanup_free_
char *home_buffer
= NULL
;
2877 const char *home
= NULL
, *shell
= NULL
;
2878 char **final_argv
= NULL
;
2879 dev_t journal_stream_dev
= 0;
2880 ino_t journal_stream_ino
= 0;
2881 bool needs_sandboxing
, /* Do we need to set up full sandboxing? (i.e. all namespacing, all MAC stuff, caps, yadda yadda */
2882 needs_setuid
, /* Do we need to do the actual setresuid()/setresgid() calls? */
2883 needs_mount_namespace
, /* Do we need to set up a mount namespace for this kernel? */
2884 needs_ambient_hack
; /* Do we need to apply the ambient capabilities hack? */
2886 _cleanup_free_
char *mac_selinux_context_net
= NULL
;
2887 bool use_selinux
= false;
2890 bool use_smack
= false;
2893 bool use_apparmor
= false;
2895 uid_t uid
= UID_INVALID
;
2896 gid_t gid
= GID_INVALID
;
2898 ExecDirectoryType dt
;
2905 assert(exit_status
);
2907 rename_process_from_path(command
->path
);
2909 /* We reset exactly these signals, since they are the
2910 * only ones we set to SIG_IGN in the main daemon. All
2911 * others we leave untouched because we set them to
2912 * SIG_DFL or a valid handler initially, both of which
2913 * will be demoted to SIG_DFL. */
2914 (void) default_signals(SIGNALS_CRASH_HANDLER
,
2915 SIGNALS_IGNORE
, -1);
2917 if (context
->ignore_sigpipe
)
2918 (void) ignore_signals(SIGPIPE
, -1);
2920 r
= reset_signal_mask();
2922 *exit_status
= EXIT_SIGNAL_MASK
;
2923 return log_unit_error_errno(unit
, r
, "Failed to set process signal mask: %m");
2926 if (params
->idle_pipe
)
2927 do_idle_pipe_dance(params
->idle_pipe
);
2929 /* Close fds we don't need very early to make sure we don't block init reexecution because it cannot bind its
2930 * sockets. Among the fds we close are the logging fds, and we want to keep them closed, so that we don't have
2931 * any fds open we don't really want open during the transition. In order to make logging work, we switch the
2932 * log subsystem into open_when_needed mode, so that it reopens the logs on every single log call. */
2935 log_set_open_when_needed(true);
2937 /* In case anything used libc syslog(), close this here, too */
2940 n_fds
= n_socket_fds
+ n_storage_fds
;
2941 r
= close_remaining_fds(params
, runtime
, dcreds
, user_lookup_fd
, socket_fd
, params
->exec_fd
, fds
, n_fds
);
2943 *exit_status
= EXIT_FDS
;
2944 return log_unit_error_errno(unit
, r
, "Failed to close unwanted file descriptors: %m");
2947 if (!context
->same_pgrp
)
2949 *exit_status
= EXIT_SETSID
;
2950 return log_unit_error_errno(unit
, errno
, "Failed to create new process session: %m");
2953 exec_context_tty_reset(context
, params
);
2955 if (unit_shall_confirm_spawn(unit
)) {
2956 const char *vc
= params
->confirm_spawn
;
2957 _cleanup_free_
char *cmdline
= NULL
;
2959 cmdline
= exec_command_line(command
->argv
);
2961 *exit_status
= EXIT_MEMORY
;
2965 r
= ask_for_confirmation(vc
, unit
, cmdline
);
2966 if (r
!= CONFIRM_EXECUTE
) {
2967 if (r
== CONFIRM_PRETEND_SUCCESS
) {
2968 *exit_status
= EXIT_SUCCESS
;
2971 *exit_status
= EXIT_CONFIRM
;
2972 log_unit_error(unit
, "Execution cancelled by the user");
2977 /* We are about to invoke NSS and PAM modules. Let's tell them what we are doing here, maybe they care. This is
2978 * used by nss-resolve to disable itself when we are about to start systemd-resolved, to avoid deadlocks. Note
2979 * that these env vars do not survive the execve(), which means they really only apply to the PAM and NSS
2980 * invocations themselves. Also note that while we'll only invoke NSS modules involved in user management they
2981 * might internally call into other NSS modules that are involved in hostname resolution, we never know. */
2982 if (setenv("SYSTEMD_ACTIVATION_UNIT", unit
->id
, true) != 0 ||
2983 setenv("SYSTEMD_ACTIVATION_SCOPE", MANAGER_IS_SYSTEM(unit
->manager
) ? "system" : "user", true) != 0) {
2984 *exit_status
= EXIT_MEMORY
;
2985 return log_unit_error_errno(unit
, errno
, "Failed to update environment: %m");
2988 if (context
->dynamic_user
&& dcreds
) {
2989 _cleanup_strv_free_
char **suggested_paths
= NULL
;
2991 /* On top of that, make sure we bypass our own NSS module nss-systemd comprehensively for any NSS
2992 * checks, if DynamicUser=1 is used, as we shouldn't create a feedback loop with ourselves here.*/
2993 if (putenv((char*) "SYSTEMD_NSS_DYNAMIC_BYPASS=1") != 0) {
2994 *exit_status
= EXIT_USER
;
2995 return log_unit_error_errno(unit
, errno
, "Failed to update environment: %m");
2998 r
= compile_suggested_paths(context
, params
, &suggested_paths
);
3000 *exit_status
= EXIT_MEMORY
;
3004 r
= dynamic_creds_realize(dcreds
, suggested_paths
, &uid
, &gid
);
3006 *exit_status
= EXIT_USER
;
3008 log_unit_error(unit
, "Failed to update dynamic user credentials: User or group with specified name already exists.");
3011 return log_unit_error_errno(unit
, r
, "Failed to update dynamic user credentials: %m");
3014 if (!uid_is_valid(uid
)) {
3015 *exit_status
= EXIT_USER
;
3016 log_unit_error(unit
, "UID validation failed for \""UID_FMT
"\"", uid
);
3020 if (!gid_is_valid(gid
)) {
3021 *exit_status
= EXIT_USER
;
3022 log_unit_error(unit
, "GID validation failed for \""GID_FMT
"\"", gid
);
3027 username
= dcreds
->user
->name
;
3030 r
= get_fixed_user(context
, &username
, &uid
, &gid
, &home
, &shell
);
3032 *exit_status
= EXIT_USER
;
3033 return log_unit_error_errno(unit
, r
, "Failed to determine user credentials: %m");
3036 r
= get_fixed_group(context
, &groupname
, &gid
);
3038 *exit_status
= EXIT_GROUP
;
3039 return log_unit_error_errno(unit
, r
, "Failed to determine group credentials: %m");
3043 /* Initialize user supplementary groups and get SupplementaryGroups= ones */
3044 r
= get_supplementary_groups(context
, username
, groupname
, gid
,
3045 &supplementary_gids
, &ngids
);
3047 *exit_status
= EXIT_GROUP
;
3048 return log_unit_error_errno(unit
, r
, "Failed to determine supplementary groups: %m");
3051 r
= send_user_lookup(unit
, user_lookup_fd
, uid
, gid
);
3053 *exit_status
= EXIT_USER
;
3054 return log_unit_error_errno(unit
, r
, "Failed to send user credentials to PID1: %m");
3057 user_lookup_fd
= safe_close(user_lookup_fd
);
3059 r
= acquire_home(context
, uid
, &home
, &home_buffer
);
3061 *exit_status
= EXIT_CHDIR
;
3062 return log_unit_error_errno(unit
, r
, "Failed to determine $HOME for user: %m");
3065 /* If a socket is connected to STDIN/STDOUT/STDERR, we
3066 * must sure to drop O_NONBLOCK */
3068 (void) fd_nonblock(socket_fd
, false);
3070 /* Journald will try to look-up our cgroup in order to populate _SYSTEMD_CGROUP and _SYSTEMD_UNIT fields.
3071 * Hence we need to migrate to the target cgroup from init.scope before connecting to journald */
3072 if (params
->cgroup_path
) {
3073 _cleanup_free_
char *p
= NULL
;
3075 r
= exec_parameters_get_cgroup_path(params
, &p
);
3077 *exit_status
= EXIT_CGROUP
;
3078 return log_unit_error_errno(unit
, r
, "Failed to acquire cgroup path: %m");
3081 r
= cg_attach_everywhere(params
->cgroup_supported
, p
, 0, NULL
, NULL
);
3083 *exit_status
= EXIT_CGROUP
;
3084 return log_unit_error_errno(unit
, r
, "Failed to attach to cgroup %s: %m", p
);
3088 if (context
->network_namespace_path
&& runtime
&& runtime
->netns_storage_socket
[0] >= 0) {
3089 r
= open_netns_path(runtime
->netns_storage_socket
, context
->network_namespace_path
);
3091 *exit_status
= EXIT_NETWORK
;
3092 return log_unit_error_errno(unit
, r
, "Failed to open network namespace path %s: %m", context
->network_namespace_path
);
3096 r
= setup_input(context
, params
, socket_fd
, named_iofds
);
3098 *exit_status
= EXIT_STDIN
;
3099 return log_unit_error_errno(unit
, r
, "Failed to set up standard input: %m");
3102 r
= setup_output(unit
, context
, params
, STDOUT_FILENO
, socket_fd
, named_iofds
, basename(command
->path
), uid
, gid
, &journal_stream_dev
, &journal_stream_ino
);
3104 *exit_status
= EXIT_STDOUT
;
3105 return log_unit_error_errno(unit
, r
, "Failed to set up standard output: %m");
3108 r
= setup_output(unit
, context
, params
, STDERR_FILENO
, socket_fd
, named_iofds
, basename(command
->path
), uid
, gid
, &journal_stream_dev
, &journal_stream_ino
);
3110 *exit_status
= EXIT_STDERR
;
3111 return log_unit_error_errno(unit
, r
, "Failed to set up standard error output: %m");
3114 if (context
->oom_score_adjust_set
) {
3115 /* When we can't make this change due to EPERM, then let's silently skip over it. User namespaces
3116 * prohibit write access to this file, and we shouldn't trip up over that. */
3117 r
= set_oom_score_adjust(context
->oom_score_adjust
);
3118 if (IN_SET(r
, -EPERM
, -EACCES
))
3119 log_unit_debug_errno(unit
, r
, "Failed to adjust OOM setting, assuming containerized execution, ignoring: %m");
3121 *exit_status
= EXIT_OOM_ADJUST
;
3122 return log_unit_error_errno(unit
, r
, "Failed to adjust OOM setting: %m");
3126 if (context
->nice_set
)
3127 if (setpriority(PRIO_PROCESS
, 0, context
->nice
) < 0) {
3128 *exit_status
= EXIT_NICE
;
3129 return log_unit_error_errno(unit
, errno
, "Failed to set up process scheduling priority (nice level): %m");
3132 if (context
->cpu_sched_set
) {
3133 struct sched_param param
= {
3134 .sched_priority
= context
->cpu_sched_priority
,
3137 r
= sched_setscheduler(0,
3138 context
->cpu_sched_policy
|
3139 (context
->cpu_sched_reset_on_fork
?
3140 SCHED_RESET_ON_FORK
: 0),
3143 *exit_status
= EXIT_SETSCHEDULER
;
3144 return log_unit_error_errno(unit
, errno
, "Failed to set up CPU scheduling: %m");
3148 if (context
->cpuset
)
3149 if (sched_setaffinity(0, CPU_ALLOC_SIZE(context
->cpuset_ncpus
), context
->cpuset
) < 0) {
3150 *exit_status
= EXIT_CPUAFFINITY
;
3151 return log_unit_error_errno(unit
, errno
, "Failed to set up CPU affinity: %m");
3154 if (context
->ioprio_set
)
3155 if (ioprio_set(IOPRIO_WHO_PROCESS
, 0, context
->ioprio
) < 0) {
3156 *exit_status
= EXIT_IOPRIO
;
3157 return log_unit_error_errno(unit
, errno
, "Failed to set up IO scheduling priority: %m");
3160 if (context
->timer_slack_nsec
!= NSEC_INFINITY
)
3161 if (prctl(PR_SET_TIMERSLACK
, context
->timer_slack_nsec
) < 0) {
3162 *exit_status
= EXIT_TIMERSLACK
;
3163 return log_unit_error_errno(unit
, errno
, "Failed to set up timer slack: %m");
3166 if (context
->personality
!= PERSONALITY_INVALID
) {
3167 r
= safe_personality(context
->personality
);
3169 *exit_status
= EXIT_PERSONALITY
;
3170 return log_unit_error_errno(unit
, r
, "Failed to set up execution domain (personality): %m");
3174 if (context
->utmp_id
)
3175 utmp_put_init_process(context
->utmp_id
, getpid_cached(), getsid(0),
3177 context
->utmp_mode
== EXEC_UTMP_INIT
? INIT_PROCESS
:
3178 context
->utmp_mode
== EXEC_UTMP_LOGIN
? LOGIN_PROCESS
:
3182 if (uid_is_valid(uid
)) {
3183 r
= chown_terminal(STDIN_FILENO
, uid
);
3185 *exit_status
= EXIT_STDIN
;
3186 return log_unit_error_errno(unit
, r
, "Failed to change ownership of terminal: %m");
3190 /* If delegation is enabled we'll pass ownership of the cgroup to the user of the new process. On cgroup v1
3191 * this is only about systemd's own hierarchy, i.e. not the controller hierarchies, simply because that's not
3192 * safe. On cgroup v2 there's only one hierarchy anyway, and delegation is safe there, hence in that case only
3193 * touch a single hierarchy too. */
3194 if (params
->cgroup_path
&& context
->user
&& (params
->flags
& EXEC_CGROUP_DELEGATE
)) {
3195 r
= cg_set_access(SYSTEMD_CGROUP_CONTROLLER
, params
->cgroup_path
, uid
, gid
);
3197 *exit_status
= EXIT_CGROUP
;
3198 return log_unit_error_errno(unit
, r
, "Failed to adjust control group access: %m");
3202 for (dt
= 0; dt
< _EXEC_DIRECTORY_TYPE_MAX
; dt
++) {
3203 r
= setup_exec_directory(context
, params
, uid
, gid
, dt
, exit_status
);
3205 return log_unit_error_errno(unit
, r
, "Failed to set up special execution directory in %s: %m", params
->prefix
[dt
]);
3208 r
= build_environment(
3220 *exit_status
= EXIT_MEMORY
;
3224 r
= build_pass_environment(context
, &pass_env
);
3226 *exit_status
= EXIT_MEMORY
;
3230 accum_env
= strv_env_merge(5,
3231 params
->environment
,
3234 context
->environment
,
3238 *exit_status
= EXIT_MEMORY
;
3241 accum_env
= strv_env_clean(accum_env
);
3243 (void) umask(context
->umask
);
3245 r
= setup_keyring(unit
, context
, params
, uid
, gid
);
3247 *exit_status
= EXIT_KEYRING
;
3248 return log_unit_error_errno(unit
, r
, "Failed to set up kernel keyring: %m");
3251 /* We need sandboxing if the caller asked us to apply it and the command isn't explicitly excepted from it */
3252 needs_sandboxing
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && !(command
->flags
& EXEC_COMMAND_FULLY_PRIVILEGED
);
3254 /* We need the ambient capability hack, if the caller asked us to apply it and the command is marked for it, and the kernel doesn't actually support ambient caps */
3255 needs_ambient_hack
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && (command
->flags
& EXEC_COMMAND_AMBIENT_MAGIC
) && !ambient_capabilities_supported();
3257 /* We need setresuid() if the caller asked us to apply sandboxing and the command isn't explicitly excepted from either whole sandboxing or just setresuid() itself, and the ambient hack is not desired */
3258 if (needs_ambient_hack
)
3259 needs_setuid
= false;
3261 needs_setuid
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && !(command
->flags
& (EXEC_COMMAND_FULLY_PRIVILEGED
|EXEC_COMMAND_NO_SETUID
));
3263 if (needs_sandboxing
) {
3264 /* MAC enablement checks need to be done before a new mount ns is created, as they rely on /sys being
3265 * present. The actual MAC context application will happen later, as late as possible, to avoid
3266 * impacting our own code paths. */
3269 use_selinux
= mac_selinux_use();
3272 use_smack
= mac_smack_use();
3275 use_apparmor
= mac_apparmor_use();
3279 if (needs_sandboxing
) {
3282 /* Let's set the resource limits before we call into PAM, so that pam_limits wins over what
3283 * is set here. (See below.) */
3285 r
= setrlimit_closest_all((const struct rlimit
* const *) context
->rlimit
, &which_failed
);
3287 *exit_status
= EXIT_LIMITS
;
3288 return log_unit_error_errno(unit
, r
, "Failed to adjust resource limit RLIMIT_%s: %m", rlimit_to_string(which_failed
));
3294 /* Let's call into PAM after we set up our own idea of resource limits to that pam_limits
3295 * wins here. (See above.) */
3297 if (context
->pam_name
&& username
) {
3298 r
= setup_pam(context
->pam_name
, username
, uid
, gid
, context
->tty_path
, &accum_env
, fds
, n_fds
);
3300 *exit_status
= EXIT_PAM
;
3301 return log_unit_error_errno(unit
, r
, "Failed to set up PAM session: %m");
3306 if ((context
->private_network
|| context
->network_namespace_path
) && runtime
&& runtime
->netns_storage_socket
[0] >= 0) {
3308 if (ns_type_supported(NAMESPACE_NET
)) {
3309 r
= setup_netns(runtime
->netns_storage_socket
);
3311 *exit_status
= EXIT_NETWORK
;
3312 return log_unit_error_errno(unit
, r
, "Failed to set up network namespacing: %m");
3314 } else if (context
->network_namespace_path
) {
3315 *exit_status
= EXIT_NETWORK
;
3316 return log_unit_error_errno(unit
, SYNTHETIC_ERRNO(EOPNOTSUPP
), "NetworkNamespacePath= is not supported, refusing.");
3318 log_unit_warning(unit
, "PrivateNetwork=yes is configured, but the kernel does not support network namespaces, ignoring.");
3321 needs_mount_namespace
= exec_needs_mount_namespace(context
, params
, runtime
);
3322 if (needs_mount_namespace
) {
3323 _cleanup_free_
char *error_path
= NULL
;
3325 r
= apply_mount_namespace(unit
, command
, context
, params
, runtime
, &error_path
);
3327 *exit_status
= EXIT_NAMESPACE
;
3328 return log_unit_error_errno(unit
, r
, "Failed to set up mount namespacing%s%s: %m",
3329 error_path
? ": " : "", strempty(error_path
));
3333 if (context
->protect_hostname
) {
3334 if (ns_type_supported(NAMESPACE_UTS
)) {
3335 if (unshare(CLONE_NEWUTS
) < 0) {
3336 *exit_status
= EXIT_NAMESPACE
;
3337 return log_unit_error_errno(unit
, errno
, "Failed to set up UTS namespacing: %m");
3340 log_unit_warning(unit
, "ProtectHostname=yes is configured, but the kernel does not support UTS namespaces, ignoring namespace setup.");
3342 r
= seccomp_protect_hostname();
3344 *exit_status
= EXIT_SECCOMP
;
3345 return log_unit_error_errno(unit
, r
, "Failed to apply hostname restrictions: %m");
3350 /* Drop groups as early as possbile */
3352 r
= enforce_groups(gid
, supplementary_gids
, ngids
);
3354 *exit_status
= EXIT_GROUP
;
3355 return log_unit_error_errno(unit
, r
, "Changing group credentials failed: %m");
3359 if (needs_sandboxing
) {
3361 if (use_selinux
&& params
->selinux_context_net
&& socket_fd
>= 0) {
3362 r
= mac_selinux_get_child_mls_label(socket_fd
, command
->path
, context
->selinux_context
, &mac_selinux_context_net
);
3364 *exit_status
= EXIT_SELINUX_CONTEXT
;
3365 return log_unit_error_errno(unit
, r
, "Failed to determine SELinux context: %m");
3370 if (context
->private_users
) {
3371 r
= setup_private_users(uid
, gid
);
3373 *exit_status
= EXIT_USER
;
3374 return log_unit_error_errno(unit
, r
, "Failed to set up user namespacing: %m");
3379 /* We repeat the fd closing here, to make sure that nothing is leaked from the PAM modules. Note that we are
3380 * more aggressive this time since socket_fd and the netns fds we don't need anymore. We do keep the exec_fd
3381 * however if we have it as we want to keep it open until the final execve(). */
3383 if (params
->exec_fd
>= 0) {
3384 exec_fd
= params
->exec_fd
;
3386 if (exec_fd
< 3 + (int) n_fds
) {
3389 /* Let's move the exec fd far up, so that it's outside of the fd range we want to pass to the
3390 * process we are about to execute. */
3392 moved_fd
= fcntl(exec_fd
, F_DUPFD_CLOEXEC
, 3 + (int) n_fds
);
3394 *exit_status
= EXIT_FDS
;
3395 return log_unit_error_errno(unit
, errno
, "Couldn't move exec fd up: %m");
3398 safe_close(exec_fd
);
3401 /* This fd should be FD_CLOEXEC already, but let's make sure. */
3402 r
= fd_cloexec(exec_fd
, true);
3404 *exit_status
= EXIT_FDS
;
3405 return log_unit_error_errno(unit
, r
, "Failed to make exec fd FD_CLOEXEC: %m");
3409 fds_with_exec_fd
= newa(int, n_fds
+ 1);
3410 memcpy_safe(fds_with_exec_fd
, fds
, n_fds
* sizeof(int));
3411 fds_with_exec_fd
[n_fds
] = exec_fd
;
3412 n_fds_with_exec_fd
= n_fds
+ 1;
3414 fds_with_exec_fd
= fds
;
3415 n_fds_with_exec_fd
= n_fds
;
3418 r
= close_all_fds(fds_with_exec_fd
, n_fds_with_exec_fd
);
3420 r
= shift_fds(fds
, n_fds
);
3422 r
= flags_fds(fds
, n_socket_fds
, n_storage_fds
, context
->non_blocking
);
3424 *exit_status
= EXIT_FDS
;
3425 return log_unit_error_errno(unit
, r
, "Failed to adjust passed file descriptors: %m");
3428 /* At this point, the fds we want to pass to the program are all ready and set up, with O_CLOEXEC turned off
3429 * and at the right fd numbers. The are no other fds open, with one exception: the exec_fd if it is defined,
3430 * and it has O_CLOEXEC set, after all we want it to be closed by the execve(), so that our parent knows we
3433 secure_bits
= context
->secure_bits
;
3435 if (needs_sandboxing
) {
3438 /* Set the RTPRIO resource limit to 0, but only if nothing else was explicitly
3439 * requested. (Note this is placed after the general resource limit initialization, see
3440 * above, in order to take precedence.) */
3441 if (context
->restrict_realtime
&& !context
->rlimit
[RLIMIT_RTPRIO
]) {
3442 if (setrlimit(RLIMIT_RTPRIO
, &RLIMIT_MAKE_CONST(0)) < 0) {
3443 *exit_status
= EXIT_LIMITS
;
3444 return log_unit_error_errno(unit
, errno
, "Failed to adjust RLIMIT_RTPRIO resource limit: %m");
3449 /* LSM Smack needs the capability CAP_MAC_ADMIN to change the current execution security context of the
3450 * process. This is the latest place before dropping capabilities. Other MAC context are set later. */
3452 r
= setup_smack(context
, command
);
3454 *exit_status
= EXIT_SMACK_PROCESS_LABEL
;
3455 return log_unit_error_errno(unit
, r
, "Failed to set SMACK process label: %m");
3460 bset
= context
->capability_bounding_set
;
3461 /* If the ambient caps hack is enabled (which means the kernel can't do them, and the user asked for
3462 * our magic fallback), then let's add some extra caps, so that the service can drop privs of its own,
3463 * instead of us doing that */
3464 if (needs_ambient_hack
)
3465 bset
|= (UINT64_C(1) << CAP_SETPCAP
) |
3466 (UINT64_C(1) << CAP_SETUID
) |
3467 (UINT64_C(1) << CAP_SETGID
);
3469 if (!cap_test_all(bset
)) {
3470 r
= capability_bounding_set_drop(bset
, false);
3472 *exit_status
= EXIT_CAPABILITIES
;
3473 return log_unit_error_errno(unit
, r
, "Failed to drop capabilities: %m");
3477 /* This is done before enforce_user, but ambient set
3478 * does not survive over setresuid() if keep_caps is not set. */
3479 if (!needs_ambient_hack
&&
3480 context
->capability_ambient_set
!= 0) {
3481 r
= capability_ambient_set_apply(context
->capability_ambient_set
, true);
3483 *exit_status
= EXIT_CAPABILITIES
;
3484 return log_unit_error_errno(unit
, r
, "Failed to apply ambient capabilities (before UID change): %m");
3490 if (uid_is_valid(uid
)) {
3491 r
= enforce_user(context
, uid
);
3493 *exit_status
= EXIT_USER
;
3494 return log_unit_error_errno(unit
, r
, "Failed to change UID to " UID_FMT
": %m", uid
);
3497 if (!needs_ambient_hack
&&
3498 context
->capability_ambient_set
!= 0) {
3500 /* Fix the ambient capabilities after user change. */
3501 r
= capability_ambient_set_apply(context
->capability_ambient_set
, false);
3503 *exit_status
= EXIT_CAPABILITIES
;
3504 return log_unit_error_errno(unit
, r
, "Failed to apply ambient capabilities (after UID change): %m");
3507 /* If we were asked to change user and ambient capabilities
3508 * were requested, we had to add keep-caps to the securebits
3509 * so that we would maintain the inherited capability set
3510 * through the setresuid(). Make sure that the bit is added
3511 * also to the context secure_bits so that we don't try to
3512 * drop the bit away next. */
3514 secure_bits
|= 1<<SECURE_KEEP_CAPS
;
3519 /* Apply working directory here, because the working directory might be on NFS and only the user running
3520 * this service might have the correct privilege to change to the working directory */
3521 r
= apply_working_directory(context
, params
, home
, needs_mount_namespace
, exit_status
);
3523 return log_unit_error_errno(unit
, r
, "Changing to the requested working directory failed: %m");
3525 if (needs_sandboxing
) {
3526 /* Apply other MAC contexts late, but before seccomp syscall filtering, as those should really be last to
3527 * influence our own codepaths as little as possible. Moreover, applying MAC contexts usually requires
3528 * syscalls that are subject to seccomp filtering, hence should probably be applied before the syscalls
3529 * are restricted. */
3533 char *exec_context
= mac_selinux_context_net
?: context
->selinux_context
;
3536 r
= setexeccon(exec_context
);
3538 *exit_status
= EXIT_SELINUX_CONTEXT
;
3539 return log_unit_error_errno(unit
, r
, "Failed to change SELinux context to %s: %m", exec_context
);
3546 if (use_apparmor
&& context
->apparmor_profile
) {
3547 r
= aa_change_onexec(context
->apparmor_profile
);
3548 if (r
< 0 && !context
->apparmor_profile_ignore
) {
3549 *exit_status
= EXIT_APPARMOR_PROFILE
;
3550 return log_unit_error_errno(unit
, errno
, "Failed to prepare AppArmor profile change to %s: %m", context
->apparmor_profile
);
3555 /* PR_GET_SECUREBITS is not privileged, while PR_SET_SECUREBITS is. So to suppress potential EPERMs
3556 * we'll try not to call PR_SET_SECUREBITS unless necessary. */
3557 if (prctl(PR_GET_SECUREBITS
) != secure_bits
)
3558 if (prctl(PR_SET_SECUREBITS
, secure_bits
) < 0) {
3559 *exit_status
= EXIT_SECUREBITS
;
3560 return log_unit_error_errno(unit
, errno
, "Failed to set process secure bits: %m");
3563 if (context_has_no_new_privileges(context
))
3564 if (prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0) < 0) {
3565 *exit_status
= EXIT_NO_NEW_PRIVILEGES
;
3566 return log_unit_error_errno(unit
, errno
, "Failed to disable new privileges: %m");
3570 r
= apply_address_families(unit
, context
);
3572 *exit_status
= EXIT_ADDRESS_FAMILIES
;
3573 return log_unit_error_errno(unit
, r
, "Failed to restrict address families: %m");
3576 r
= apply_memory_deny_write_execute(unit
, context
);
3578 *exit_status
= EXIT_SECCOMP
;
3579 return log_unit_error_errno(unit
, r
, "Failed to disable writing to executable memory: %m");
3582 r
= apply_restrict_realtime(unit
, context
);
3584 *exit_status
= EXIT_SECCOMP
;
3585 return log_unit_error_errno(unit
, r
, "Failed to apply realtime restrictions: %m");
3588 r
= apply_restrict_suid_sgid(unit
, context
);
3590 *exit_status
= EXIT_SECCOMP
;
3591 return log_unit_error_errno(unit
, r
, "Failed to apply SUID/SGID restrictions: %m");
3594 r
= apply_restrict_namespaces(unit
, context
);
3596 *exit_status
= EXIT_SECCOMP
;
3597 return log_unit_error_errno(unit
, r
, "Failed to apply namespace restrictions: %m");
3600 r
= apply_protect_sysctl(unit
, context
);
3602 *exit_status
= EXIT_SECCOMP
;
3603 return log_unit_error_errno(unit
, r
, "Failed to apply sysctl restrictions: %m");
3606 r
= apply_protect_kernel_modules(unit
, context
);
3608 *exit_status
= EXIT_SECCOMP
;
3609 return log_unit_error_errno(unit
, r
, "Failed to apply module loading restrictions: %m");
3612 r
= apply_private_devices(unit
, context
);
3614 *exit_status
= EXIT_SECCOMP
;
3615 return log_unit_error_errno(unit
, r
, "Failed to set up private devices: %m");
3618 r
= apply_syscall_archs(unit
, context
);
3620 *exit_status
= EXIT_SECCOMP
;
3621 return log_unit_error_errno(unit
, r
, "Failed to apply syscall architecture restrictions: %m");
3624 r
= apply_lock_personality(unit
, context
);
3626 *exit_status
= EXIT_SECCOMP
;
3627 return log_unit_error_errno(unit
, r
, "Failed to lock personalities: %m");
3630 /* This really should remain the last step before the execve(), to make sure our own code is unaffected
3631 * by the filter as little as possible. */
3632 r
= apply_syscall_filter(unit
, context
, needs_ambient_hack
);
3634 *exit_status
= EXIT_SECCOMP
;
3635 return log_unit_error_errno(unit
, r
, "Failed to apply system call filters: %m");
3640 if (!strv_isempty(context
->unset_environment
)) {
3643 ee
= strv_env_delete(accum_env
, 1, context
->unset_environment
);
3645 *exit_status
= EXIT_MEMORY
;
3649 strv_free_and_replace(accum_env
, ee
);
3652 if (!FLAGS_SET(command
->flags
, EXEC_COMMAND_NO_ENV_EXPAND
)) {
3653 replaced_argv
= replace_env_argv(command
->argv
, accum_env
);
3654 if (!replaced_argv
) {
3655 *exit_status
= EXIT_MEMORY
;
3658 final_argv
= replaced_argv
;
3660 final_argv
= command
->argv
;
3662 if (DEBUG_LOGGING
) {
3663 _cleanup_free_
char *line
;
3665 line
= exec_command_line(final_argv
);
3667 log_struct(LOG_DEBUG
,
3668 "EXECUTABLE=%s", command
->path
,
3669 LOG_UNIT_MESSAGE(unit
, "Executing: %s", line
),
3671 LOG_UNIT_INVOCATION_ID(unit
));
3677 /* We have finished with all our initializations. Let's now let the manager know that. From this point
3678 * on, if the manager sees POLLHUP on the exec_fd, then execve() was successful. */
3680 if (write(exec_fd
, &hot
, sizeof(hot
)) < 0) {
3681 *exit_status
= EXIT_EXEC
;
3682 return log_unit_error_errno(unit
, errno
, "Failed to enable exec_fd: %m");
3686 execve(command
->path
, final_argv
, accum_env
);
3692 /* The execve() failed. This means the exec_fd is still open. Which means we need to tell the manager
3693 * that POLLHUP on it no longer means execve() succeeded. */
3695 if (write(exec_fd
, &hot
, sizeof(hot
)) < 0) {
3696 *exit_status
= EXIT_EXEC
;
3697 return log_unit_error_errno(unit
, errno
, "Failed to disable exec_fd: %m");
3701 if (r
== -ENOENT
&& (command
->flags
& EXEC_COMMAND_IGNORE_FAILURE
)) {
3702 log_struct_errno(LOG_INFO
, r
,
3703 "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR
,
3705 LOG_UNIT_INVOCATION_ID(unit
),
3706 LOG_UNIT_MESSAGE(unit
, "Executable %s missing, skipping: %m",
3708 "EXECUTABLE=%s", command
->path
);
3712 *exit_status
= EXIT_EXEC
;
3713 return log_unit_error_errno(unit
, r
, "Failed to execute command: %m");
3716 static int exec_context_load_environment(const Unit
*unit
, const ExecContext
*c
, char ***l
);
3717 static int exec_context_named_iofds(const ExecContext
*c
, const ExecParameters
*p
, int named_iofds
[3]);
3719 int exec_spawn(Unit
*unit
,
3720 ExecCommand
*command
,
3721 const ExecContext
*context
,
3722 const ExecParameters
*params
,
3723 ExecRuntime
*runtime
,
3724 DynamicCreds
*dcreds
,
3727 int socket_fd
, r
, named_iofds
[3] = { -1, -1, -1 }, *fds
= NULL
;
3728 _cleanup_free_
char *subcgroup_path
= NULL
;
3729 _cleanup_strv_free_
char **files_env
= NULL
;
3730 size_t n_storage_fds
= 0, n_socket_fds
= 0;
3731 _cleanup_free_
char *line
= NULL
;
3739 assert(params
->fds
|| (params
->n_socket_fds
+ params
->n_storage_fds
<= 0));
3741 if (context
->std_input
== EXEC_INPUT_SOCKET
||
3742 context
->std_output
== EXEC_OUTPUT_SOCKET
||
3743 context
->std_error
== EXEC_OUTPUT_SOCKET
) {
3745 if (params
->n_socket_fds
> 1) {
3746 log_unit_error(unit
, "Got more than one socket.");
3750 if (params
->n_socket_fds
== 0) {
3751 log_unit_error(unit
, "Got no socket.");
3755 socket_fd
= params
->fds
[0];
3759 n_socket_fds
= params
->n_socket_fds
;
3760 n_storage_fds
= params
->n_storage_fds
;
3763 r
= exec_context_named_iofds(context
, params
, named_iofds
);
3765 return log_unit_error_errno(unit
, r
, "Failed to load a named file descriptor: %m");
3767 r
= exec_context_load_environment(unit
, context
, &files_env
);
3769 return log_unit_error_errno(unit
, r
, "Failed to load environment files: %m");
3771 line
= exec_command_line(command
->argv
);
3775 log_struct(LOG_DEBUG
,
3776 LOG_UNIT_MESSAGE(unit
, "About to execute: %s", line
),
3777 "EXECUTABLE=%s", command
->path
,
3779 LOG_UNIT_INVOCATION_ID(unit
));
3781 if (params
->cgroup_path
) {
3782 r
= exec_parameters_get_cgroup_path(params
, &subcgroup_path
);
3784 return log_unit_error_errno(unit
, r
, "Failed to acquire subcgroup path: %m");
3785 if (r
> 0) { /* We are using a child cgroup */
3786 r
= cg_create(SYSTEMD_CGROUP_CONTROLLER
, subcgroup_path
);
3788 return log_unit_error_errno(unit
, r
, "Failed to create control group '%s': %m", subcgroup_path
);
3794 return log_unit_error_errno(unit
, errno
, "Failed to fork: %m");
3797 int exit_status
= EXIT_SUCCESS
;
3799 r
= exec_child(unit
,
3811 unit
->manager
->user_lookup_fds
[1],
3815 log_struct_errno(LOG_ERR
, r
,
3816 "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR
,
3818 LOG_UNIT_INVOCATION_ID(unit
),
3819 LOG_UNIT_MESSAGE(unit
, "Failed at step %s spawning %s: %m",
3820 exit_status_to_string(exit_status
, EXIT_STATUS_SYSTEMD
),
3822 "EXECUTABLE=%s", command
->path
);
3827 log_unit_debug(unit
, "Forked %s as "PID_FMT
, command
->path
, pid
);
3829 /* We add the new process to the cgroup both in the child (so that we can be sure that no user code is ever
3830 * executed outside of the cgroup) and in the parent (so that we can be sure that when we kill the cgroup the
3831 * process will be killed too). */
3833 (void) cg_attach(SYSTEMD_CGROUP_CONTROLLER
, subcgroup_path
, pid
);
3835 exec_status_start(&command
->exec_status
, pid
);
3841 void exec_context_init(ExecContext
*c
) {
3842 ExecDirectoryType i
;
3847 c
->ioprio
= IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE
, 0);
3848 c
->cpu_sched_policy
= SCHED_OTHER
;
3849 c
->syslog_priority
= LOG_DAEMON
|LOG_INFO
;
3850 c
->syslog_level_prefix
= true;
3851 c
->ignore_sigpipe
= true;
3852 c
->timer_slack_nsec
= NSEC_INFINITY
;
3853 c
->personality
= PERSONALITY_INVALID
;
3854 for (i
= 0; i
< _EXEC_DIRECTORY_TYPE_MAX
; i
++)
3855 c
->directories
[i
].mode
= 0755;
3856 c
->capability_bounding_set
= CAP_ALL
;
3857 assert_cc(NAMESPACE_FLAGS_INITIAL
!= NAMESPACE_FLAGS_ALL
);
3858 c
->restrict_namespaces
= NAMESPACE_FLAGS_INITIAL
;
3859 c
->log_level_max
= -1;
3862 void exec_context_done(ExecContext
*c
) {
3863 ExecDirectoryType i
;
3868 c
->environment
= strv_free(c
->environment
);
3869 c
->environment_files
= strv_free(c
->environment_files
);
3870 c
->pass_environment
= strv_free(c
->pass_environment
);
3871 c
->unset_environment
= strv_free(c
->unset_environment
);
3873 rlimit_free_all(c
->rlimit
);
3875 for (l
= 0; l
< 3; l
++) {
3876 c
->stdio_fdname
[l
] = mfree(c
->stdio_fdname
[l
]);
3877 c
->stdio_file
[l
] = mfree(c
->stdio_file
[l
]);
3880 c
->working_directory
= mfree(c
->working_directory
);
3881 c
->root_directory
= mfree(c
->root_directory
);
3882 c
->root_image
= mfree(c
->root_image
);
3883 c
->tty_path
= mfree(c
->tty_path
);
3884 c
->syslog_identifier
= mfree(c
->syslog_identifier
);
3885 c
->user
= mfree(c
->user
);
3886 c
->group
= mfree(c
->group
);
3888 c
->supplementary_groups
= strv_free(c
->supplementary_groups
);
3890 c
->pam_name
= mfree(c
->pam_name
);
3892 c
->read_only_paths
= strv_free(c
->read_only_paths
);
3893 c
->read_write_paths
= strv_free(c
->read_write_paths
);
3894 c
->inaccessible_paths
= strv_free(c
->inaccessible_paths
);
3896 bind_mount_free_many(c
->bind_mounts
, c
->n_bind_mounts
);
3897 c
->bind_mounts
= NULL
;
3898 c
->n_bind_mounts
= 0;
3899 temporary_filesystem_free_many(c
->temporary_filesystems
, c
->n_temporary_filesystems
);
3900 c
->temporary_filesystems
= NULL
;
3901 c
->n_temporary_filesystems
= 0;
3903 c
->cpuset
= cpu_set_mfree(c
->cpuset
);
3905 c
->utmp_id
= mfree(c
->utmp_id
);
3906 c
->selinux_context
= mfree(c
->selinux_context
);
3907 c
->apparmor_profile
= mfree(c
->apparmor_profile
);
3908 c
->smack_process_label
= mfree(c
->smack_process_label
);
3910 c
->syscall_filter
= hashmap_free(c
->syscall_filter
);
3911 c
->syscall_archs
= set_free(c
->syscall_archs
);
3912 c
->address_families
= set_free(c
->address_families
);
3914 for (i
= 0; i
< _EXEC_DIRECTORY_TYPE_MAX
; i
++)
3915 c
->directories
[i
].paths
= strv_free(c
->directories
[i
].paths
);
3917 c
->log_level_max
= -1;
3919 exec_context_free_log_extra_fields(c
);
3921 c
->log_rate_limit_interval_usec
= 0;
3922 c
->log_rate_limit_burst
= 0;
3924 c
->stdin_data
= mfree(c
->stdin_data
);
3925 c
->stdin_data_size
= 0;
3927 c
->network_namespace_path
= mfree(c
->network_namespace_path
);
3930 int exec_context_destroy_runtime_directory(const ExecContext
*c
, const char *runtime_prefix
) {
3935 if (!runtime_prefix
)
3938 STRV_FOREACH(i
, c
->directories
[EXEC_DIRECTORY_RUNTIME
].paths
) {
3939 _cleanup_free_
char *p
;
3941 p
= path_join(runtime_prefix
, *i
);
3945 /* We execute this synchronously, since we need to be sure this is gone when we start the
3947 (void) rm_rf(p
, REMOVE_ROOT
);
3953 static void exec_command_done(ExecCommand
*c
) {
3956 c
->path
= mfree(c
->path
);
3957 c
->argv
= strv_free(c
->argv
);
3960 void exec_command_done_array(ExecCommand
*c
, size_t n
) {
3963 for (i
= 0; i
< n
; i
++)
3964 exec_command_done(c
+i
);
3967 ExecCommand
* exec_command_free_list(ExecCommand
*c
) {
3971 LIST_REMOVE(command
, c
, i
);
3972 exec_command_done(i
);
3979 void exec_command_free_array(ExecCommand
**c
, size_t n
) {
3982 for (i
= 0; i
< n
; i
++)
3983 c
[i
] = exec_command_free_list(c
[i
]);
3986 void exec_command_reset_status_array(ExecCommand
*c
, size_t n
) {
3989 for (i
= 0; i
< n
; i
++)
3990 exec_status_reset(&c
[i
].exec_status
);
3993 void exec_command_reset_status_list_array(ExecCommand
**c
, size_t n
) {
3996 for (i
= 0; i
< n
; i
++) {
3999 LIST_FOREACH(command
, z
, c
[i
])
4000 exec_status_reset(&z
->exec_status
);
4004 typedef struct InvalidEnvInfo
{
4009 static void invalid_env(const char *p
, void *userdata
) {
4010 InvalidEnvInfo
*info
= userdata
;
4012 log_unit_error(info
->unit
, "Ignoring invalid environment assignment '%s': %s", p
, info
->path
);
4015 const char* exec_context_fdname(const ExecContext
*c
, int fd_index
) {
4021 if (c
->std_input
!= EXEC_INPUT_NAMED_FD
)
4024 return c
->stdio_fdname
[STDIN_FILENO
] ?: "stdin";
4027 if (c
->std_output
!= EXEC_OUTPUT_NAMED_FD
)
4030 return c
->stdio_fdname
[STDOUT_FILENO
] ?: "stdout";
4033 if (c
->std_error
!= EXEC_OUTPUT_NAMED_FD
)
4036 return c
->stdio_fdname
[STDERR_FILENO
] ?: "stderr";
4043 static int exec_context_named_iofds(const ExecContext
*c
, const ExecParameters
*p
, int named_iofds
[static 3]) {
4045 const char* stdio_fdname
[3];
4051 targets
= (c
->std_input
== EXEC_INPUT_NAMED_FD
) +
4052 (c
->std_output
== EXEC_OUTPUT_NAMED_FD
) +
4053 (c
->std_error
== EXEC_OUTPUT_NAMED_FD
);
4055 for (i
= 0; i
< 3; i
++)
4056 stdio_fdname
[i
] = exec_context_fdname(c
, i
);
4058 n_fds
= p
->n_storage_fds
+ p
->n_socket_fds
;
4060 for (i
= 0; i
< n_fds
&& targets
> 0; i
++)
4061 if (named_iofds
[STDIN_FILENO
] < 0 &&
4062 c
->std_input
== EXEC_INPUT_NAMED_FD
&&
4063 stdio_fdname
[STDIN_FILENO
] &&
4064 streq(p
->fd_names
[i
], stdio_fdname
[STDIN_FILENO
])) {
4066 named_iofds
[STDIN_FILENO
] = p
->fds
[i
];
4069 } else if (named_iofds
[STDOUT_FILENO
] < 0 &&
4070 c
->std_output
== EXEC_OUTPUT_NAMED_FD
&&
4071 stdio_fdname
[STDOUT_FILENO
] &&
4072 streq(p
->fd_names
[i
], stdio_fdname
[STDOUT_FILENO
])) {
4074 named_iofds
[STDOUT_FILENO
] = p
->fds
[i
];
4077 } else if (named_iofds
[STDERR_FILENO
] < 0 &&
4078 c
->std_error
== EXEC_OUTPUT_NAMED_FD
&&
4079 stdio_fdname
[STDERR_FILENO
] &&
4080 streq(p
->fd_names
[i
], stdio_fdname
[STDERR_FILENO
])) {
4082 named_iofds
[STDERR_FILENO
] = p
->fds
[i
];
4086 return targets
== 0 ? 0 : -ENOENT
;
4089 static int exec_context_load_environment(const Unit
*unit
, const ExecContext
*c
, char ***l
) {
4090 char **i
, **r
= NULL
;
4095 STRV_FOREACH(i
, c
->environment_files
) {
4099 bool ignore
= false;
4101 _cleanup_globfree_ glob_t pglob
= {};
4110 if (!path_is_absolute(fn
)) {
4118 /* Filename supports globbing, take all matching files */
4119 k
= safe_glob(fn
, 0, &pglob
);
4128 /* When we don't match anything, -ENOENT should be returned */
4129 assert(pglob
.gl_pathc
> 0);
4131 for (n
= 0; n
< pglob
.gl_pathc
; n
++) {
4132 k
= load_env_file(NULL
, pglob
.gl_pathv
[n
], &p
);
4140 /* Log invalid environment variables with filename */
4142 InvalidEnvInfo info
= {
4144 .path
= pglob
.gl_pathv
[n
]
4147 p
= strv_env_clean_with_callback(p
, invalid_env
, &info
);
4155 m
= strv_env_merge(2, r
, p
);
4171 static bool tty_may_match_dev_console(const char *tty
) {
4172 _cleanup_free_
char *resolved
= NULL
;
4177 tty
= skip_dev_prefix(tty
);
4179 /* trivial identity? */
4180 if (streq(tty
, "console"))
4183 if (resolve_dev_console(&resolved
) < 0)
4184 return true; /* if we could not resolve, assume it may */
4186 /* "tty0" means the active VC, so it may be the same sometimes */
4187 return path_equal(resolved
, tty
) || (streq(resolved
, "tty0") && tty_is_vc(tty
));
4190 static bool exec_context_may_touch_tty(const ExecContext
*ec
) {
4193 return ec
->tty_reset
||
4195 ec
->tty_vt_disallocate
||
4196 is_terminal_input(ec
->std_input
) ||
4197 is_terminal_output(ec
->std_output
) ||
4198 is_terminal_output(ec
->std_error
);
4201 bool exec_context_may_touch_console(const ExecContext
*ec
) {
4203 return exec_context_may_touch_tty(ec
) &&
4204 tty_may_match_dev_console(exec_context_tty_path(ec
));
4207 static void strv_fprintf(FILE *f
, char **l
) {
4213 fprintf(f
, " %s", *g
);
4216 void exec_context_dump(const ExecContext
*c
, FILE* f
, const char *prefix
) {
4217 ExecDirectoryType dt
;
4225 prefix
= strempty(prefix
);
4229 "%sWorkingDirectory: %s\n"
4230 "%sRootDirectory: %s\n"
4231 "%sNonBlocking: %s\n"
4232 "%sPrivateTmp: %s\n"
4233 "%sPrivateDevices: %s\n"
4234 "%sProtectKernelTunables: %s\n"
4235 "%sProtectKernelModules: %s\n"
4236 "%sProtectControlGroups: %s\n"
4237 "%sPrivateNetwork: %s\n"
4238 "%sPrivateUsers: %s\n"
4239 "%sProtectHome: %s\n"
4240 "%sProtectSystem: %s\n"
4241 "%sMountAPIVFS: %s\n"
4242 "%sIgnoreSIGPIPE: %s\n"
4243 "%sMemoryDenyWriteExecute: %s\n"
4244 "%sRestrictRealtime: %s\n"
4245 "%sRestrictSUIDSGID: %s\n"
4246 "%sKeyringMode: %s\n"
4247 "%sProtectHostname: %s\n",
4249 prefix
, c
->working_directory
? c
->working_directory
: "/",
4250 prefix
, c
->root_directory
? c
->root_directory
: "/",
4251 prefix
, yes_no(c
->non_blocking
),
4252 prefix
, yes_no(c
->private_tmp
),
4253 prefix
, yes_no(c
->private_devices
),
4254 prefix
, yes_no(c
->protect_kernel_tunables
),
4255 prefix
, yes_no(c
->protect_kernel_modules
),
4256 prefix
, yes_no(c
->protect_control_groups
),
4257 prefix
, yes_no(c
->private_network
),
4258 prefix
, yes_no(c
->private_users
),
4259 prefix
, protect_home_to_string(c
->protect_home
),
4260 prefix
, protect_system_to_string(c
->protect_system
),
4261 prefix
, yes_no(c
->mount_apivfs
),
4262 prefix
, yes_no(c
->ignore_sigpipe
),
4263 prefix
, yes_no(c
->memory_deny_write_execute
),
4264 prefix
, yes_no(c
->restrict_realtime
),
4265 prefix
, yes_no(c
->restrict_suid_sgid
),
4266 prefix
, exec_keyring_mode_to_string(c
->keyring_mode
),
4267 prefix
, yes_no(c
->protect_hostname
));
4270 fprintf(f
, "%sRootImage: %s\n", prefix
, c
->root_image
);
4272 STRV_FOREACH(e
, c
->environment
)
4273 fprintf(f
, "%sEnvironment: %s\n", prefix
, *e
);
4275 STRV_FOREACH(e
, c
->environment_files
)
4276 fprintf(f
, "%sEnvironmentFile: %s\n", prefix
, *e
);
4278 STRV_FOREACH(e
, c
->pass_environment
)
4279 fprintf(f
, "%sPassEnvironment: %s\n", prefix
, *e
);
4281 STRV_FOREACH(e
, c
->unset_environment
)
4282 fprintf(f
, "%sUnsetEnvironment: %s\n", prefix
, *e
);
4284 fprintf(f
, "%sRuntimeDirectoryPreserve: %s\n", prefix
, exec_preserve_mode_to_string(c
->runtime_directory_preserve_mode
));
4286 for (dt
= 0; dt
< _EXEC_DIRECTORY_TYPE_MAX
; dt
++) {
4287 fprintf(f
, "%s%sMode: %04o\n", prefix
, exec_directory_type_to_string(dt
), c
->directories
[dt
].mode
);
4289 STRV_FOREACH(d
, c
->directories
[dt
].paths
)
4290 fprintf(f
, "%s%s: %s\n", prefix
, exec_directory_type_to_string(dt
), *d
);
4298 if (c
->oom_score_adjust_set
)
4300 "%sOOMScoreAdjust: %i\n",
4301 prefix
, c
->oom_score_adjust
);
4303 for (i
= 0; i
< RLIM_NLIMITS
; i
++)
4305 fprintf(f
, "%sLimit%s: " RLIM_FMT
"\n",
4306 prefix
, rlimit_to_string(i
), c
->rlimit
[i
]->rlim_max
);
4307 fprintf(f
, "%sLimit%sSoft: " RLIM_FMT
"\n",
4308 prefix
, rlimit_to_string(i
), c
->rlimit
[i
]->rlim_cur
);
4311 if (c
->ioprio_set
) {
4312 _cleanup_free_
char *class_str
= NULL
;
4314 r
= ioprio_class_to_string_alloc(IOPRIO_PRIO_CLASS(c
->ioprio
), &class_str
);
4316 fprintf(f
, "%sIOSchedulingClass: %s\n", prefix
, class_str
);
4318 fprintf(f
, "%sIOPriority: %lu\n", prefix
, IOPRIO_PRIO_DATA(c
->ioprio
));
4321 if (c
->cpu_sched_set
) {
4322 _cleanup_free_
char *policy_str
= NULL
;
4324 r
= sched_policy_to_string_alloc(c
->cpu_sched_policy
, &policy_str
);
4326 fprintf(f
, "%sCPUSchedulingPolicy: %s\n", prefix
, policy_str
);
4329 "%sCPUSchedulingPriority: %i\n"
4330 "%sCPUSchedulingResetOnFork: %s\n",
4331 prefix
, c
->cpu_sched_priority
,
4332 prefix
, yes_no(c
->cpu_sched_reset_on_fork
));
4336 fprintf(f
, "%sCPUAffinity:", prefix
);
4337 for (i
= 0; i
< c
->cpuset_ncpus
; i
++)
4338 if (CPU_ISSET_S(i
, CPU_ALLOC_SIZE(c
->cpuset_ncpus
), c
->cpuset
))
4339 fprintf(f
, " %u", i
);
4343 if (c
->timer_slack_nsec
!= NSEC_INFINITY
)
4344 fprintf(f
, "%sTimerSlackNSec: "NSEC_FMT
"\n", prefix
, c
->timer_slack_nsec
);
4347 "%sStandardInput: %s\n"
4348 "%sStandardOutput: %s\n"
4349 "%sStandardError: %s\n",
4350 prefix
, exec_input_to_string(c
->std_input
),
4351 prefix
, exec_output_to_string(c
->std_output
),
4352 prefix
, exec_output_to_string(c
->std_error
));
4354 if (c
->std_input
== EXEC_INPUT_NAMED_FD
)
4355 fprintf(f
, "%sStandardInputFileDescriptorName: %s\n", prefix
, c
->stdio_fdname
[STDIN_FILENO
]);
4356 if (c
->std_output
== EXEC_OUTPUT_NAMED_FD
)
4357 fprintf(f
, "%sStandardOutputFileDescriptorName: %s\n", prefix
, c
->stdio_fdname
[STDOUT_FILENO
]);
4358 if (c
->std_error
== EXEC_OUTPUT_NAMED_FD
)
4359 fprintf(f
, "%sStandardErrorFileDescriptorName: %s\n", prefix
, c
->stdio_fdname
[STDERR_FILENO
]);
4361 if (c
->std_input
== EXEC_INPUT_FILE
)
4362 fprintf(f
, "%sStandardInputFile: %s\n", prefix
, c
->stdio_file
[STDIN_FILENO
]);
4363 if (c
->std_output
== EXEC_OUTPUT_FILE
)
4364 fprintf(f
, "%sStandardOutputFile: %s\n", prefix
, c
->stdio_file
[STDOUT_FILENO
]);
4365 if (c
->std_output
== EXEC_OUTPUT_FILE_APPEND
)
4366 fprintf(f
, "%sStandardOutputFileToAppend: %s\n", prefix
, c
->stdio_file
[STDOUT_FILENO
]);
4367 if (c
->std_error
== EXEC_OUTPUT_FILE
)
4368 fprintf(f
, "%sStandardErrorFile: %s\n", prefix
, c
->stdio_file
[STDERR_FILENO
]);
4369 if (c
->std_error
== EXEC_OUTPUT_FILE_APPEND
)
4370 fprintf(f
, "%sStandardErrorFileToAppend: %s\n", prefix
, c
->stdio_file
[STDERR_FILENO
]);
4376 "%sTTYVHangup: %s\n"
4377 "%sTTYVTDisallocate: %s\n",
4378 prefix
, c
->tty_path
,
4379 prefix
, yes_no(c
->tty_reset
),
4380 prefix
, yes_no(c
->tty_vhangup
),
4381 prefix
, yes_no(c
->tty_vt_disallocate
));
4383 if (IN_SET(c
->std_output
,
4386 EXEC_OUTPUT_JOURNAL
,
4387 EXEC_OUTPUT_SYSLOG_AND_CONSOLE
,
4388 EXEC_OUTPUT_KMSG_AND_CONSOLE
,
4389 EXEC_OUTPUT_JOURNAL_AND_CONSOLE
) ||
4390 IN_SET(c
->std_error
,
4393 EXEC_OUTPUT_JOURNAL
,
4394 EXEC_OUTPUT_SYSLOG_AND_CONSOLE
,
4395 EXEC_OUTPUT_KMSG_AND_CONSOLE
,
4396 EXEC_OUTPUT_JOURNAL_AND_CONSOLE
)) {
4398 _cleanup_free_
char *fac_str
= NULL
, *lvl_str
= NULL
;
4400 r
= log_facility_unshifted_to_string_alloc(c
->syslog_priority
>> 3, &fac_str
);
4402 fprintf(f
, "%sSyslogFacility: %s\n", prefix
, fac_str
);
4404 r
= log_level_to_string_alloc(LOG_PRI(c
->syslog_priority
), &lvl_str
);
4406 fprintf(f
, "%sSyslogLevel: %s\n", prefix
, lvl_str
);
4409 if (c
->log_level_max
>= 0) {
4410 _cleanup_free_
char *t
= NULL
;
4412 (void) log_level_to_string_alloc(c
->log_level_max
, &t
);
4414 fprintf(f
, "%sLogLevelMax: %s\n", prefix
, strna(t
));
4417 if (c
->log_rate_limit_interval_usec
> 0) {
4418 char buf_timespan
[FORMAT_TIMESPAN_MAX
];
4421 "%sLogRateLimitIntervalSec: %s\n",
4422 prefix
, format_timespan(buf_timespan
, sizeof(buf_timespan
), c
->log_rate_limit_interval_usec
, USEC_PER_SEC
));
4425 if (c
->log_rate_limit_burst
> 0)
4426 fprintf(f
, "%sLogRateLimitBurst: %u\n", prefix
, c
->log_rate_limit_burst
);
4428 if (c
->n_log_extra_fields
> 0) {
4431 for (j
= 0; j
< c
->n_log_extra_fields
; j
++) {
4432 fprintf(f
, "%sLogExtraFields: ", prefix
);
4433 fwrite(c
->log_extra_fields
[j
].iov_base
,
4434 1, c
->log_extra_fields
[j
].iov_len
,
4440 if (c
->secure_bits
) {
4441 _cleanup_free_
char *str
= NULL
;
4443 r
= secure_bits_to_string_alloc(c
->secure_bits
, &str
);
4445 fprintf(f
, "%sSecure Bits: %s\n", prefix
, str
);
4448 if (c
->capability_bounding_set
!= CAP_ALL
) {
4449 _cleanup_free_
char *str
= NULL
;
4451 r
= capability_set_to_string_alloc(c
->capability_bounding_set
, &str
);
4453 fprintf(f
, "%sCapabilityBoundingSet: %s\n", prefix
, str
);
4456 if (c
->capability_ambient_set
!= 0) {
4457 _cleanup_free_
char *str
= NULL
;
4459 r
= capability_set_to_string_alloc(c
->capability_ambient_set
, &str
);
4461 fprintf(f
, "%sAmbientCapabilities: %s\n", prefix
, str
);
4465 fprintf(f
, "%sUser: %s\n", prefix
, c
->user
);
4467 fprintf(f
, "%sGroup: %s\n", prefix
, c
->group
);
4469 fprintf(f
, "%sDynamicUser: %s\n", prefix
, yes_no(c
->dynamic_user
));
4471 if (!strv_isempty(c
->supplementary_groups
)) {
4472 fprintf(f
, "%sSupplementaryGroups:", prefix
);
4473 strv_fprintf(f
, c
->supplementary_groups
);
4478 fprintf(f
, "%sPAMName: %s\n", prefix
, c
->pam_name
);
4480 if (!strv_isempty(c
->read_write_paths
)) {
4481 fprintf(f
, "%sReadWritePaths:", prefix
);
4482 strv_fprintf(f
, c
->read_write_paths
);
4486 if (!strv_isempty(c
->read_only_paths
)) {
4487 fprintf(f
, "%sReadOnlyPaths:", prefix
);
4488 strv_fprintf(f
, c
->read_only_paths
);
4492 if (!strv_isempty(c
->inaccessible_paths
)) {
4493 fprintf(f
, "%sInaccessiblePaths:", prefix
);
4494 strv_fprintf(f
, c
->inaccessible_paths
);
4498 if (c
->n_bind_mounts
> 0)
4499 for (i
= 0; i
< c
->n_bind_mounts
; i
++)
4500 fprintf(f
, "%s%s: %s%s:%s:%s\n", prefix
,
4501 c
->bind_mounts
[i
].read_only
? "BindReadOnlyPaths" : "BindPaths",
4502 c
->bind_mounts
[i
].ignore_enoent
? "-": "",
4503 c
->bind_mounts
[i
].source
,
4504 c
->bind_mounts
[i
].destination
,
4505 c
->bind_mounts
[i
].recursive
? "rbind" : "norbind");
4507 if (c
->n_temporary_filesystems
> 0)
4508 for (i
= 0; i
< c
->n_temporary_filesystems
; i
++) {
4509 TemporaryFileSystem
*t
= c
->temporary_filesystems
+ i
;
4511 fprintf(f
, "%sTemporaryFileSystem: %s%s%s\n", prefix
,
4513 isempty(t
->options
) ? "" : ":",
4514 strempty(t
->options
));
4519 "%sUtmpIdentifier: %s\n",
4520 prefix
, c
->utmp_id
);
4522 if (c
->selinux_context
)
4524 "%sSELinuxContext: %s%s\n",
4525 prefix
, c
->selinux_context_ignore
? "-" : "", c
->selinux_context
);
4527 if (c
->apparmor_profile
)
4529 "%sAppArmorProfile: %s%s\n",
4530 prefix
, c
->apparmor_profile_ignore
? "-" : "", c
->apparmor_profile
);
4532 if (c
->smack_process_label
)
4534 "%sSmackProcessLabel: %s%s\n",
4535 prefix
, c
->smack_process_label_ignore
? "-" : "", c
->smack_process_label
);
4537 if (c
->personality
!= PERSONALITY_INVALID
)
4539 "%sPersonality: %s\n",
4540 prefix
, strna(personality_to_string(c
->personality
)));
4543 "%sLockPersonality: %s\n",
4544 prefix
, yes_no(c
->lock_personality
));
4546 if (c
->syscall_filter
) {
4554 "%sSystemCallFilter: ",
4557 if (!c
->syscall_whitelist
)
4561 HASHMAP_FOREACH_KEY(val
, id
, c
->syscall_filter
, j
) {
4562 _cleanup_free_
char *name
= NULL
;
4563 const char *errno_name
= NULL
;
4564 int num
= PTR_TO_INT(val
);
4571 name
= seccomp_syscall_resolve_num_arch(SCMP_ARCH_NATIVE
, PTR_TO_INT(id
) - 1);
4572 fputs(strna(name
), f
);
4575 errno_name
= errno_to_name(num
);
4577 fprintf(f
, ":%s", errno_name
);
4579 fprintf(f
, ":%d", num
);
4587 if (c
->syscall_archs
) {
4594 "%sSystemCallArchitectures:",
4598 SET_FOREACH(id
, c
->syscall_archs
, j
)
4599 fprintf(f
, " %s", strna(seccomp_arch_to_string(PTR_TO_UINT32(id
) - 1)));
4604 if (exec_context_restrict_namespaces_set(c
)) {
4605 _cleanup_free_
char *s
= NULL
;
4607 r
= namespace_flags_to_string(c
->restrict_namespaces
, &s
);
4609 fprintf(f
, "%sRestrictNamespaces: %s\n",
4613 if (c
->network_namespace_path
)
4615 "%sNetworkNamespacePath: %s\n",
4616 prefix
, c
->network_namespace_path
);
4618 if (c
->syscall_errno
> 0) {
4619 const char *errno_name
;
4621 fprintf(f
, "%sSystemCallErrorNumber: ", prefix
);
4623 errno_name
= errno_to_name(c
->syscall_errno
);
4625 fprintf(f
, "%s\n", errno_name
);
4627 fprintf(f
, "%d\n", c
->syscall_errno
);
4631 bool exec_context_maintains_privileges(const ExecContext
*c
) {
4634 /* Returns true if the process forked off would run under
4635 * an unchanged UID or as root. */
4640 if (streq(c
->user
, "root") || streq(c
->user
, "0"))
4646 int exec_context_get_effective_ioprio(const ExecContext
*c
) {
4654 p
= ioprio_get(IOPRIO_WHO_PROCESS
, 0);
4656 return IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE
, 4);
4661 void exec_context_free_log_extra_fields(ExecContext
*c
) {
4666 for (l
= 0; l
< c
->n_log_extra_fields
; l
++)
4667 free(c
->log_extra_fields
[l
].iov_base
);
4668 c
->log_extra_fields
= mfree(c
->log_extra_fields
);
4669 c
->n_log_extra_fields
= 0;
4672 void exec_context_revert_tty(ExecContext
*c
) {
4677 /* First, reset the TTY (possibly kicking everybody else from the TTY) */
4678 exec_context_tty_reset(c
, NULL
);
4680 /* And then undo what chown_terminal() did earlier. Note that we only do this if we have a path
4681 * configured. If the TTY was passed to us as file descriptor we assume the TTY is opened and managed
4682 * by whoever passed it to us and thus knows better when and how to chmod()/chown() it back. */
4684 if (exec_context_may_touch_tty(c
)) {
4687 path
= exec_context_tty_path(c
);
4689 r
= chmod_and_chown(path
, TTY_MODE
, 0, TTY_GID
);
4690 if (r
< 0 && r
!= -ENOENT
)
4691 log_warning_errno(r
, "Failed to reset TTY ownership/access mode of %s, ignoring: %m", path
);
4696 void exec_status_start(ExecStatus
*s
, pid_t pid
) {
4703 dual_timestamp_get(&s
->start_timestamp
);
4706 void exec_status_exit(ExecStatus
*s
, const ExecContext
*context
, pid_t pid
, int code
, int status
) {
4709 if (s
->pid
!= pid
) {
4715 dual_timestamp_get(&s
->exit_timestamp
);
4720 if (context
&& context
->utmp_id
)
4721 (void) utmp_put_dead_process(context
->utmp_id
, pid
, code
, status
);
4724 void exec_status_reset(ExecStatus
*s
) {
4727 *s
= (ExecStatus
) {};
4730 void exec_status_dump(const ExecStatus
*s
, FILE *f
, const char *prefix
) {
4731 char buf
[FORMAT_TIMESTAMP_MAX
];
4739 prefix
= strempty(prefix
);
4742 "%sPID: "PID_FMT
"\n",
4745 if (dual_timestamp_is_set(&s
->start_timestamp
))
4747 "%sStart Timestamp: %s\n",
4748 prefix
, format_timestamp(buf
, sizeof(buf
), s
->start_timestamp
.realtime
));
4750 if (dual_timestamp_is_set(&s
->exit_timestamp
))
4752 "%sExit Timestamp: %s\n"
4754 "%sExit Status: %i\n",
4755 prefix
, format_timestamp(buf
, sizeof(buf
), s
->exit_timestamp
.realtime
),
4756 prefix
, sigchld_code_to_string(s
->code
),
4760 static char *exec_command_line(char **argv
) {
4768 STRV_FOREACH(a
, argv
)
4776 STRV_FOREACH(a
, argv
) {
4783 if (strpbrk(*a
, WHITESPACE
)) {
4794 /* FIXME: this doesn't really handle arguments that have
4795 * spaces and ticks in them */
4800 static void exec_command_dump(ExecCommand
*c
, FILE *f
, const char *prefix
) {
4801 _cleanup_free_
char *cmd
= NULL
;
4802 const char *prefix2
;
4807 prefix
= strempty(prefix
);
4808 prefix2
= strjoina(prefix
, "\t");
4810 cmd
= exec_command_line(c
->argv
);
4812 "%sCommand Line: %s\n",
4813 prefix
, cmd
? cmd
: strerror(ENOMEM
));
4815 exec_status_dump(&c
->exec_status
, f
, prefix2
);
4818 void exec_command_dump_list(ExecCommand
*c
, FILE *f
, const char *prefix
) {
4821 prefix
= strempty(prefix
);
4823 LIST_FOREACH(command
, c
, c
)
4824 exec_command_dump(c
, f
, prefix
);
4827 void exec_command_append_list(ExecCommand
**l
, ExecCommand
*e
) {
4834 /* It's kind of important, that we keep the order here */
4835 LIST_FIND_TAIL(command
, *l
, end
);
4836 LIST_INSERT_AFTER(command
, *l
, end
, e
);
4841 int exec_command_set(ExecCommand
*c
, const char *path
, ...) {
4849 l
= strv_new_ap(path
, ap
);
4861 free_and_replace(c
->path
, p
);
4863 return strv_free_and_replace(c
->argv
, l
);
4866 int exec_command_append(ExecCommand
*c
, const char *path
, ...) {
4867 _cleanup_strv_free_
char **l
= NULL
;
4875 l
= strv_new_ap(path
, ap
);
4881 r
= strv_extend_strv(&c
->argv
, l
, false);
4888 static void *remove_tmpdir_thread(void *p
) {
4889 _cleanup_free_
char *path
= p
;
4891 (void) rm_rf(path
, REMOVE_ROOT
|REMOVE_PHYSICAL
);
4895 static ExecRuntime
* exec_runtime_free(ExecRuntime
*rt
, bool destroy
) {
4902 (void) hashmap_remove(rt
->manager
->exec_runtime_by_id
, rt
->id
);
4904 /* When destroy is true, then rm_rf tmp_dir and var_tmp_dir. */
4905 if (destroy
&& rt
->tmp_dir
) {
4906 log_debug("Spawning thread to nuke %s", rt
->tmp_dir
);
4908 r
= asynchronous_job(remove_tmpdir_thread
, rt
->tmp_dir
);
4910 log_warning_errno(r
, "Failed to nuke %s: %m", rt
->tmp_dir
);
4917 if (destroy
&& rt
->var_tmp_dir
) {
4918 log_debug("Spawning thread to nuke %s", rt
->var_tmp_dir
);
4920 r
= asynchronous_job(remove_tmpdir_thread
, rt
->var_tmp_dir
);
4922 log_warning_errno(r
, "Failed to nuke %s: %m", rt
->var_tmp_dir
);
4923 free(rt
->var_tmp_dir
);
4926 rt
->var_tmp_dir
= NULL
;
4929 rt
->id
= mfree(rt
->id
);
4930 rt
->tmp_dir
= mfree(rt
->tmp_dir
);
4931 rt
->var_tmp_dir
= mfree(rt
->var_tmp_dir
);
4932 safe_close_pair(rt
->netns_storage_socket
);
4936 static void exec_runtime_freep(ExecRuntime
**rt
) {
4937 (void) exec_runtime_free(*rt
, false);
4940 static int exec_runtime_allocate(ExecRuntime
**ret
) {
4945 n
= new(ExecRuntime
, 1);
4949 *n
= (ExecRuntime
) {
4950 .netns_storage_socket
= { -1, -1 },
4957 static int exec_runtime_add(
4960 const char *tmp_dir
,
4961 const char *var_tmp_dir
,
4962 const int netns_storage_socket
[2],
4963 ExecRuntime
**ret
) {
4965 _cleanup_(exec_runtime_freep
) ExecRuntime
*rt
= NULL
;
4971 r
= hashmap_ensure_allocated(&m
->exec_runtime_by_id
, &string_hash_ops
);
4975 r
= exec_runtime_allocate(&rt
);
4979 rt
->id
= strdup(id
);
4984 rt
->tmp_dir
= strdup(tmp_dir
);
4988 /* When tmp_dir is set, then we require var_tmp_dir is also set. */
4989 assert(var_tmp_dir
);
4990 rt
->var_tmp_dir
= strdup(var_tmp_dir
);
4991 if (!rt
->var_tmp_dir
)
4995 if (netns_storage_socket
) {
4996 rt
->netns_storage_socket
[0] = netns_storage_socket
[0];
4997 rt
->netns_storage_socket
[1] = netns_storage_socket
[1];
5000 r
= hashmap_put(m
->exec_runtime_by_id
, rt
->id
, rt
);
5009 /* do not remove created ExecRuntime object when the operation succeeds. */
5014 static int exec_runtime_make(Manager
*m
, const ExecContext
*c
, const char *id
, ExecRuntime
**ret
) {
5015 _cleanup_free_
char *tmp_dir
= NULL
, *var_tmp_dir
= NULL
;
5016 _cleanup_close_pair_
int netns_storage_socket
[2] = { -1, -1 };
5023 /* It is not necessary to create ExecRuntime object. */
5024 if (!c
->private_network
&& !c
->private_tmp
&& !c
->network_namespace_path
)
5027 if (c
->private_tmp
) {
5028 r
= setup_tmp_dirs(id
, &tmp_dir
, &var_tmp_dir
);
5033 if (c
->private_network
|| c
->network_namespace_path
) {
5034 if (socketpair(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
, 0, netns_storage_socket
) < 0)
5038 r
= exec_runtime_add(m
, id
, tmp_dir
, var_tmp_dir
, netns_storage_socket
, ret
);
5043 netns_storage_socket
[0] = netns_storage_socket
[1] = -1;
5047 int exec_runtime_acquire(Manager
*m
, const ExecContext
*c
, const char *id
, bool create
, ExecRuntime
**ret
) {
5055 rt
= hashmap_get(m
->exec_runtime_by_id
, id
);
5057 /* We already have a ExecRuntime object, let's increase the ref count and reuse it */
5063 /* If not found, then create a new object. */
5064 r
= exec_runtime_make(m
, c
, id
, &rt
);
5066 /* When r == 0, it is not necessary to create ExecRuntime object. */
5070 /* increment reference counter. */
5076 ExecRuntime
*exec_runtime_unref(ExecRuntime
*rt
, bool destroy
) {
5080 assert(rt
->n_ref
> 0);
5086 return exec_runtime_free(rt
, destroy
);
5089 int exec_runtime_serialize(const Manager
*m
, FILE *f
, FDSet
*fds
) {
5097 HASHMAP_FOREACH(rt
, m
->exec_runtime_by_id
, i
) {
5098 fprintf(f
, "exec-runtime=%s", rt
->id
);
5101 fprintf(f
, " tmp-dir=%s", rt
->tmp_dir
);
5103 if (rt
->var_tmp_dir
)
5104 fprintf(f
, " var-tmp-dir=%s", rt
->var_tmp_dir
);
5106 if (rt
->netns_storage_socket
[0] >= 0) {
5109 copy
= fdset_put_dup(fds
, rt
->netns_storage_socket
[0]);
5113 fprintf(f
, " netns-socket-0=%i", copy
);
5116 if (rt
->netns_storage_socket
[1] >= 0) {
5119 copy
= fdset_put_dup(fds
, rt
->netns_storage_socket
[1]);
5123 fprintf(f
, " netns-socket-1=%i", copy
);
5132 int exec_runtime_deserialize_compat(Unit
*u
, const char *key
, const char *value
, FDSet
*fds
) {
5133 _cleanup_(exec_runtime_freep
) ExecRuntime
*rt_create
= NULL
;
5137 /* This is for the migration from old (v237 or earlier) deserialization text.
5138 * Due to the bug #7790, this may not work with the units that use JoinsNamespaceOf=.
5139 * Even if the ExecRuntime object originally created by the other unit, we cannot judge
5140 * so or not from the serialized text, then we always creates a new object owned by this. */
5146 /* Manager manages ExecRuntime objects by the unit id.
5147 * So, we omit the serialized text when the unit does not have id (yet?)... */
5148 if (isempty(u
->id
)) {
5149 log_unit_debug(u
, "Invocation ID not found. Dropping runtime parameter.");
5153 r
= hashmap_ensure_allocated(&u
->manager
->exec_runtime_by_id
, &string_hash_ops
);
5155 log_unit_debug_errno(u
, r
, "Failed to allocate storage for runtime parameter: %m");
5159 rt
= hashmap_get(u
->manager
->exec_runtime_by_id
, u
->id
);
5161 r
= exec_runtime_allocate(&rt_create
);
5165 rt_create
->id
= strdup(u
->id
);
5172 if (streq(key
, "tmp-dir")) {
5175 copy
= strdup(value
);
5179 free_and_replace(rt
->tmp_dir
, copy
);
5181 } else if (streq(key
, "var-tmp-dir")) {
5184 copy
= strdup(value
);
5188 free_and_replace(rt
->var_tmp_dir
, copy
);
5190 } else if (streq(key
, "netns-socket-0")) {
5193 if (safe_atoi(value
, &fd
) < 0 || !fdset_contains(fds
, fd
)) {
5194 log_unit_debug(u
, "Failed to parse netns socket value: %s", value
);
5198 safe_close(rt
->netns_storage_socket
[0]);
5199 rt
->netns_storage_socket
[0] = fdset_remove(fds
, fd
);
5201 } else if (streq(key
, "netns-socket-1")) {
5204 if (safe_atoi(value
, &fd
) < 0 || !fdset_contains(fds
, fd
)) {
5205 log_unit_debug(u
, "Failed to parse netns socket value: %s", value
);
5209 safe_close(rt
->netns_storage_socket
[1]);
5210 rt
->netns_storage_socket
[1] = fdset_remove(fds
, fd
);
5214 /* If the object is newly created, then put it to the hashmap which manages ExecRuntime objects. */
5216 r
= hashmap_put(u
->manager
->exec_runtime_by_id
, rt_create
->id
, rt_create
);
5218 log_unit_debug_errno(u
, r
, "Failed to put runtime parameter to manager's storage: %m");
5222 rt_create
->manager
= u
->manager
;
5231 void exec_runtime_deserialize_one(Manager
*m
, const char *value
, FDSet
*fds
) {
5232 char *id
= NULL
, *tmp_dir
= NULL
, *var_tmp_dir
= NULL
;
5233 int r
, fd0
= -1, fd1
= -1;
5234 const char *p
, *v
= value
;
5241 n
= strcspn(v
, " ");
5242 id
= strndupa(v
, n
);
5247 v
= startswith(p
, "tmp-dir=");
5249 n
= strcspn(v
, " ");
5250 tmp_dir
= strndupa(v
, n
);
5256 v
= startswith(p
, "var-tmp-dir=");
5258 n
= strcspn(v
, " ");
5259 var_tmp_dir
= strndupa(v
, n
);
5265 v
= startswith(p
, "netns-socket-0=");
5269 n
= strcspn(v
, " ");
5270 buf
= strndupa(v
, n
);
5271 if (safe_atoi(buf
, &fd0
) < 0 || !fdset_contains(fds
, fd0
)) {
5272 log_debug("Unable to process exec-runtime netns fd specification.");
5275 fd0
= fdset_remove(fds
, fd0
);
5281 v
= startswith(p
, "netns-socket-1=");
5285 n
= strcspn(v
, " ");
5286 buf
= strndupa(v
, n
);
5287 if (safe_atoi(buf
, &fd1
) < 0 || !fdset_contains(fds
, fd1
)) {
5288 log_debug("Unable to process exec-runtime netns fd specification.");
5291 fd1
= fdset_remove(fds
, fd1
);
5296 r
= exec_runtime_add(m
, id
, tmp_dir
, var_tmp_dir
, (int[]) { fd0
, fd1
}, NULL
);
5298 log_debug_errno(r
, "Failed to add exec-runtime: %m");
5301 void exec_runtime_vacuum(Manager
*m
) {
5307 /* Free unreferenced ExecRuntime objects. This is used after manager deserialization process. */
5309 HASHMAP_FOREACH(rt
, m
->exec_runtime_by_id
, i
) {
5313 (void) exec_runtime_free(rt
, false);
5317 void exec_params_clear(ExecParameters
*p
) {
5321 strv_free(p
->environment
);
5324 static const char* const exec_input_table
[_EXEC_INPUT_MAX
] = {
5325 [EXEC_INPUT_NULL
] = "null",
5326 [EXEC_INPUT_TTY
] = "tty",
5327 [EXEC_INPUT_TTY_FORCE
] = "tty-force",
5328 [EXEC_INPUT_TTY_FAIL
] = "tty-fail",
5329 [EXEC_INPUT_SOCKET
] = "socket",
5330 [EXEC_INPUT_NAMED_FD
] = "fd",
5331 [EXEC_INPUT_DATA
] = "data",
5332 [EXEC_INPUT_FILE
] = "file",
5335 DEFINE_STRING_TABLE_LOOKUP(exec_input
, ExecInput
);
5337 static const char* const exec_output_table
[_EXEC_OUTPUT_MAX
] = {
5338 [EXEC_OUTPUT_INHERIT
] = "inherit",
5339 [EXEC_OUTPUT_NULL
] = "null",
5340 [EXEC_OUTPUT_TTY
] = "tty",
5341 [EXEC_OUTPUT_SYSLOG
] = "syslog",
5342 [EXEC_OUTPUT_SYSLOG_AND_CONSOLE
] = "syslog+console",
5343 [EXEC_OUTPUT_KMSG
] = "kmsg",
5344 [EXEC_OUTPUT_KMSG_AND_CONSOLE
] = "kmsg+console",
5345 [EXEC_OUTPUT_JOURNAL
] = "journal",
5346 [EXEC_OUTPUT_JOURNAL_AND_CONSOLE
] = "journal+console",
5347 [EXEC_OUTPUT_SOCKET
] = "socket",
5348 [EXEC_OUTPUT_NAMED_FD
] = "fd",
5349 [EXEC_OUTPUT_FILE
] = "file",
5350 [EXEC_OUTPUT_FILE_APPEND
] = "append",
5353 DEFINE_STRING_TABLE_LOOKUP(exec_output
, ExecOutput
);
5355 static const char* const exec_utmp_mode_table
[_EXEC_UTMP_MODE_MAX
] = {
5356 [EXEC_UTMP_INIT
] = "init",
5357 [EXEC_UTMP_LOGIN
] = "login",
5358 [EXEC_UTMP_USER
] = "user",
5361 DEFINE_STRING_TABLE_LOOKUP(exec_utmp_mode
, ExecUtmpMode
);
5363 static const char* const exec_preserve_mode_table
[_EXEC_PRESERVE_MODE_MAX
] = {
5364 [EXEC_PRESERVE_NO
] = "no",
5365 [EXEC_PRESERVE_YES
] = "yes",
5366 [EXEC_PRESERVE_RESTART
] = "restart",
5369 DEFINE_STRING_TABLE_LOOKUP_WITH_BOOLEAN(exec_preserve_mode
, ExecPreserveMode
, EXEC_PRESERVE_YES
);
5371 static const char* const exec_directory_type_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
5372 [EXEC_DIRECTORY_RUNTIME
] = "RuntimeDirectory",
5373 [EXEC_DIRECTORY_STATE
] = "StateDirectory",
5374 [EXEC_DIRECTORY_CACHE
] = "CacheDirectory",
5375 [EXEC_DIRECTORY_LOGS
] = "LogsDirectory",
5376 [EXEC_DIRECTORY_CONFIGURATION
] = "ConfigurationDirectory",
5379 DEFINE_STRING_TABLE_LOOKUP(exec_directory_type
, ExecDirectoryType
);
5381 static const char* const exec_directory_env_name_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
5382 [EXEC_DIRECTORY_RUNTIME
] = "RUNTIME_DIRECTORY",
5383 [EXEC_DIRECTORY_STATE
] = "STATE_DIRECTORY",
5384 [EXEC_DIRECTORY_CACHE
] = "CACHE_DIRECTORY",
5385 [EXEC_DIRECTORY_LOGS
] = "LOGS_DIRECTORY",
5386 [EXEC_DIRECTORY_CONFIGURATION
] = "CONFIGURATION_DIRECTORY",
5389 DEFINE_PRIVATE_STRING_TABLE_LOOKUP_TO_STRING(exec_directory_env_name
, ExecDirectoryType
);
5391 static const char* const exec_keyring_mode_table
[_EXEC_KEYRING_MODE_MAX
] = {
5392 [EXEC_KEYRING_INHERIT
] = "inherit",
5393 [EXEC_KEYRING_PRIVATE
] = "private",
5394 [EXEC_KEYRING_SHARED
] = "shared",
5397 DEFINE_STRING_TABLE_LOOKUP(exec_keyring_mode
, ExecKeyringMode
);