1 /* SPDX-License-Identifier: LGPL-2.1-or-later */
6 #include <sys/eventfd.h>
10 #include <sys/personality.h>
11 #include <sys/prctl.h>
13 #include <sys/types.h>
19 #include <security/pam_appl.h>
23 #include <selinux/selinux.h>
31 #include <sys/apparmor.h>
34 #include "sd-messages.h"
38 #include "alloc-util.h"
40 #include "apparmor-util.h"
45 #include "capability-util.h"
46 #include "cgroup-setup.h"
47 #include "chown-recursive.h"
48 #include "cpu-set-util.h"
52 #include "errno-list.h"
54 #include "exit-status.h"
57 #include "format-util.h"
59 #include "glob-util.h"
60 #include "hexdecoct.h"
67 #include "manager-dump.h"
68 #include "memory-util.h"
69 #include "missing_fs.h"
71 #include "mount-util.h"
72 #include "mountpoint-util.h"
73 #include "namespace.h"
74 #include "parse-util.h"
75 #include "path-util.h"
76 #include "process-util.h"
77 #include "random-util.h"
78 #include "rlimit-util.h"
81 #include "seccomp-util.h"
83 #include "securebits-util.h"
84 #include "selinux-util.h"
85 #include "signal-util.h"
86 #include "smack-util.h"
87 #include "socket-util.h"
89 #include "stat-util.h"
90 #include "string-table.h"
91 #include "string-util.h"
93 #include "syslog-util.h"
94 #include "terminal-util.h"
95 #include "tmpfile-util.h"
96 #include "umask-util.h"
97 #include "unit-serialize.h"
98 #include "user-util.h"
99 #include "utmp-wtmp.h"
101 #define IDLE_TIMEOUT_USEC (5*USEC_PER_SEC)
102 #define IDLE_TIMEOUT2_USEC (1*USEC_PER_SEC)
104 #define SNDBUF_SIZE (8*1024*1024)
106 static int shift_fds(int fds
[], size_t n_fds
) {
110 /* Modifies the fds array! (sorts it) */
114 for (int start
= 0;;) {
115 int restart_from
= -1;
117 for (int i
= start
; i
< (int) n_fds
; i
++) {
120 /* Already at right index? */
124 nfd
= fcntl(fds
[i
], F_DUPFD
, i
+ 3);
131 /* Hmm, the fd we wanted isn't free? Then
132 * let's remember that and try again from here */
133 if (nfd
!= i
+3 && restart_from
< 0)
137 if (restart_from
< 0)
140 start
= restart_from
;
146 static int flags_fds(const int fds
[], size_t n_socket_fds
, size_t n_storage_fds
, bool nonblock
) {
150 n_fds
= n_socket_fds
+ n_storage_fds
;
156 /* Drops/Sets O_NONBLOCK and FD_CLOEXEC from the file flags.
157 * O_NONBLOCK only applies to socket activation though. */
159 for (size_t i
= 0; i
< n_fds
; i
++) {
161 if (i
< n_socket_fds
) {
162 r
= fd_nonblock(fds
[i
], nonblock
);
167 /* We unconditionally drop FD_CLOEXEC from the fds,
168 * since after all we want to pass these fds to our
171 r
= fd_cloexec(fds
[i
], false);
179 static const char *exec_context_tty_path(const ExecContext
*context
) {
182 if (context
->stdio_as_fds
)
185 if (context
->tty_path
)
186 return context
->tty_path
;
188 return "/dev/console";
191 static void exec_context_tty_reset(const ExecContext
*context
, const ExecParameters
*p
) {
196 path
= exec_context_tty_path(context
);
198 if (context
->tty_vhangup
) {
199 if (p
&& p
->stdin_fd
>= 0)
200 (void) terminal_vhangup_fd(p
->stdin_fd
);
202 (void) terminal_vhangup(path
);
205 if (context
->tty_reset
) {
206 if (p
&& p
->stdin_fd
>= 0)
207 (void) reset_terminal_fd(p
->stdin_fd
, true);
209 (void) reset_terminal(path
);
212 if (context
->tty_vt_disallocate
&& path
)
213 (void) vt_disallocate(path
);
216 static bool is_terminal_input(ExecInput i
) {
219 EXEC_INPUT_TTY_FORCE
,
220 EXEC_INPUT_TTY_FAIL
);
223 static bool is_terminal_output(ExecOutput o
) {
226 EXEC_OUTPUT_KMSG_AND_CONSOLE
,
227 EXEC_OUTPUT_JOURNAL_AND_CONSOLE
);
230 static bool is_kmsg_output(ExecOutput o
) {
233 EXEC_OUTPUT_KMSG_AND_CONSOLE
);
236 static bool exec_context_needs_term(const ExecContext
*c
) {
239 /* Return true if the execution context suggests we should set $TERM to something useful. */
241 if (is_terminal_input(c
->std_input
))
244 if (is_terminal_output(c
->std_output
))
247 if (is_terminal_output(c
->std_error
))
250 return !!c
->tty_path
;
253 static int open_null_as(int flags
, int nfd
) {
258 fd
= open("/dev/null", flags
|O_NOCTTY
);
262 return move_fd(fd
, nfd
, false);
265 static int connect_journal_socket(
267 const char *log_namespace
,
271 union sockaddr_union sa
;
273 uid_t olduid
= UID_INVALID
;
274 gid_t oldgid
= GID_INVALID
;
279 strjoina("/run/systemd/journal.", log_namespace
, "/stdout") :
280 "/run/systemd/journal/stdout";
281 r
= sockaddr_un_set_path(&sa
.un
, j
);
286 if (gid_is_valid(gid
)) {
289 if (setegid(gid
) < 0)
293 if (uid_is_valid(uid
)) {
296 if (seteuid(uid
) < 0) {
302 r
= connect(fd
, &sa
.sa
, sa_len
) < 0 ? -errno
: 0;
304 /* If we fail to restore the uid or gid, things will likely
305 fail later on. This should only happen if an LSM interferes. */
307 if (uid_is_valid(uid
))
308 (void) seteuid(olduid
);
311 if (gid_is_valid(gid
))
312 (void) setegid(oldgid
);
317 static int connect_logger_as(
319 const ExecContext
*context
,
320 const ExecParameters
*params
,
327 _cleanup_close_
int fd
= -1;
332 assert(output
< _EXEC_OUTPUT_MAX
);
336 fd
= socket(AF_UNIX
, SOCK_STREAM
, 0);
340 r
= connect_journal_socket(fd
, context
->log_namespace
, uid
, gid
);
344 if (shutdown(fd
, SHUT_RD
) < 0)
347 (void) fd_inc_sndbuf(fd
, SNDBUF_SIZE
);
357 context
->syslog_identifier
?: ident
,
358 params
->flags
& EXEC_PASS_LOG_UNIT
? unit
->id
: "",
359 context
->syslog_priority
,
360 !!context
->syslog_level_prefix
,
362 is_kmsg_output(output
),
363 is_terminal_output(output
)) < 0)
366 return move_fd(TAKE_FD(fd
), nfd
, false);
369 static int open_terminal_as(const char *path
, int flags
, int nfd
) {
375 fd
= open_terminal(path
, flags
| O_NOCTTY
);
379 return move_fd(fd
, nfd
, false);
382 static int acquire_path(const char *path
, int flags
, mode_t mode
) {
383 union sockaddr_union sa
;
385 _cleanup_close_
int fd
= -1;
390 if (IN_SET(flags
& O_ACCMODE
, O_WRONLY
, O_RDWR
))
393 fd
= open(path
, flags
|O_NOCTTY
, mode
);
397 if (errno
!= ENXIO
) /* ENXIO is returned when we try to open() an AF_UNIX file system socket on Linux */
400 /* So, it appears the specified path could be an AF_UNIX socket. Let's see if we can connect to it. */
402 r
= sockaddr_un_set_path(&sa
.un
, path
);
404 return r
== -EINVAL
? -ENXIO
: r
;
407 fd
= socket(AF_UNIX
, SOCK_STREAM
, 0);
411 if (connect(fd
, &sa
.sa
, sa_len
) < 0)
412 return errno
== EINVAL
? -ENXIO
: -errno
; /* Propagate initial error if we get EINVAL, i.e. we have
413 * indication that this wasn't an AF_UNIX socket after all */
415 if ((flags
& O_ACCMODE
) == O_RDONLY
)
416 r
= shutdown(fd
, SHUT_WR
);
417 else if ((flags
& O_ACCMODE
) == O_WRONLY
)
418 r
= shutdown(fd
, SHUT_RD
);
427 static int fixup_input(
428 const ExecContext
*context
,
430 bool apply_tty_stdin
) {
436 std_input
= context
->std_input
;
438 if (is_terminal_input(std_input
) && !apply_tty_stdin
)
439 return EXEC_INPUT_NULL
;
441 if (std_input
== EXEC_INPUT_SOCKET
&& socket_fd
< 0)
442 return EXEC_INPUT_NULL
;
444 if (std_input
== EXEC_INPUT_DATA
&& context
->stdin_data_size
== 0)
445 return EXEC_INPUT_NULL
;
450 static int fixup_output(ExecOutput output
, int socket_fd
) {
452 if (output
== EXEC_OUTPUT_SOCKET
&& socket_fd
< 0)
453 return EXEC_OUTPUT_INHERIT
;
458 static int setup_input(
459 const ExecContext
*context
,
460 const ExecParameters
*params
,
462 const int named_iofds
[static 3]) {
470 if (params
->stdin_fd
>= 0) {
471 if (dup2(params
->stdin_fd
, STDIN_FILENO
) < 0)
474 /* Try to make this the controlling tty, if it is a tty, and reset it */
475 if (isatty(STDIN_FILENO
)) {
476 (void) ioctl(STDIN_FILENO
, TIOCSCTTY
, context
->std_input
== EXEC_INPUT_TTY_FORCE
);
477 (void) reset_terminal_fd(STDIN_FILENO
, true);
483 i
= fixup_input(context
, socket_fd
, params
->flags
& EXEC_APPLY_TTY_STDIN
);
487 case EXEC_INPUT_NULL
:
488 return open_null_as(O_RDONLY
, STDIN_FILENO
);
491 case EXEC_INPUT_TTY_FORCE
:
492 case EXEC_INPUT_TTY_FAIL
: {
495 fd
= acquire_terminal(exec_context_tty_path(context
),
496 i
== EXEC_INPUT_TTY_FAIL
? ACQUIRE_TERMINAL_TRY
:
497 i
== EXEC_INPUT_TTY_FORCE
? ACQUIRE_TERMINAL_FORCE
:
498 ACQUIRE_TERMINAL_WAIT
,
503 return move_fd(fd
, STDIN_FILENO
, false);
506 case EXEC_INPUT_SOCKET
:
507 assert(socket_fd
>= 0);
509 return dup2(socket_fd
, STDIN_FILENO
) < 0 ? -errno
: STDIN_FILENO
;
511 case EXEC_INPUT_NAMED_FD
:
512 assert(named_iofds
[STDIN_FILENO
] >= 0);
514 (void) fd_nonblock(named_iofds
[STDIN_FILENO
], false);
515 return dup2(named_iofds
[STDIN_FILENO
], STDIN_FILENO
) < 0 ? -errno
: STDIN_FILENO
;
517 case EXEC_INPUT_DATA
: {
520 fd
= acquire_data_fd(context
->stdin_data
, context
->stdin_data_size
, 0);
524 return move_fd(fd
, STDIN_FILENO
, false);
527 case EXEC_INPUT_FILE
: {
531 assert(context
->stdio_file
[STDIN_FILENO
]);
533 rw
= (context
->std_output
== EXEC_OUTPUT_FILE
&& streq_ptr(context
->stdio_file
[STDIN_FILENO
], context
->stdio_file
[STDOUT_FILENO
])) ||
534 (context
->std_error
== EXEC_OUTPUT_FILE
&& streq_ptr(context
->stdio_file
[STDIN_FILENO
], context
->stdio_file
[STDERR_FILENO
]));
536 fd
= acquire_path(context
->stdio_file
[STDIN_FILENO
], rw
? O_RDWR
: O_RDONLY
, 0666 & ~context
->umask
);
540 return move_fd(fd
, STDIN_FILENO
, false);
544 assert_not_reached("Unknown input type");
548 static bool can_inherit_stderr_from_stdout(
549 const ExecContext
*context
,
555 /* Returns true, if given the specified STDERR and STDOUT output we can directly dup() the stdout fd to the
558 if (e
== EXEC_OUTPUT_INHERIT
)
563 if (e
== EXEC_OUTPUT_NAMED_FD
)
564 return streq_ptr(context
->stdio_fdname
[STDOUT_FILENO
], context
->stdio_fdname
[STDERR_FILENO
]);
566 if (IN_SET(e
, EXEC_OUTPUT_FILE
, EXEC_OUTPUT_FILE_APPEND
, EXEC_OUTPUT_FILE_TRUNCATE
))
567 return streq_ptr(context
->stdio_file
[STDOUT_FILENO
], context
->stdio_file
[STDERR_FILENO
]);
572 static int setup_output(
574 const ExecContext
*context
,
575 const ExecParameters
*params
,
578 const int named_iofds
[static 3],
582 dev_t
*journal_stream_dev
,
583 ino_t
*journal_stream_ino
) {
593 assert(journal_stream_dev
);
594 assert(journal_stream_ino
);
596 if (fileno
== STDOUT_FILENO
&& params
->stdout_fd
>= 0) {
598 if (dup2(params
->stdout_fd
, STDOUT_FILENO
) < 0)
601 return STDOUT_FILENO
;
604 if (fileno
== STDERR_FILENO
&& params
->stderr_fd
>= 0) {
605 if (dup2(params
->stderr_fd
, STDERR_FILENO
) < 0)
608 return STDERR_FILENO
;
611 i
= fixup_input(context
, socket_fd
, params
->flags
& EXEC_APPLY_TTY_STDIN
);
612 o
= fixup_output(context
->std_output
, socket_fd
);
614 if (fileno
== STDERR_FILENO
) {
616 e
= fixup_output(context
->std_error
, socket_fd
);
618 /* This expects the input and output are already set up */
620 /* Don't change the stderr file descriptor if we inherit all
621 * the way and are not on a tty */
622 if (e
== EXEC_OUTPUT_INHERIT
&&
623 o
== EXEC_OUTPUT_INHERIT
&&
624 i
== EXEC_INPUT_NULL
&&
625 !is_terminal_input(context
->std_input
) &&
629 /* Duplicate from stdout if possible */
630 if (can_inherit_stderr_from_stdout(context
, o
, e
))
631 return dup2(STDOUT_FILENO
, fileno
) < 0 ? -errno
: fileno
;
635 } else if (o
== EXEC_OUTPUT_INHERIT
) {
636 /* If input got downgraded, inherit the original value */
637 if (i
== EXEC_INPUT_NULL
&& is_terminal_input(context
->std_input
))
638 return open_terminal_as(exec_context_tty_path(context
), O_WRONLY
, fileno
);
640 /* If the input is connected to anything that's not a /dev/null or a data fd, inherit that... */
641 if (!IN_SET(i
, EXEC_INPUT_NULL
, EXEC_INPUT_DATA
))
642 return dup2(STDIN_FILENO
, fileno
) < 0 ? -errno
: fileno
;
644 /* If we are not started from PID 1 we just inherit STDOUT from our parent process. */
648 /* We need to open /dev/null here anew, to get the right access mode. */
649 return open_null_as(O_WRONLY
, fileno
);
654 case EXEC_OUTPUT_NULL
:
655 return open_null_as(O_WRONLY
, fileno
);
657 case EXEC_OUTPUT_TTY
:
658 if (is_terminal_input(i
))
659 return dup2(STDIN_FILENO
, fileno
) < 0 ? -errno
: fileno
;
661 /* We don't reset the terminal if this is just about output */
662 return open_terminal_as(exec_context_tty_path(context
), O_WRONLY
, fileno
);
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",
671 fileno
== STDOUT_FILENO
? "stdout" : "stderr");
672 r
= open_null_as(O_WRONLY
, fileno
);
676 /* If we connected this fd to the journal via a stream, patch the device/inode into the passed
677 * parameters, but only then. This is useful so that we can set $JOURNAL_STREAM that permits
678 * services to detect whether they are connected to the journal or not.
680 * If both stdout and stderr are connected to a stream then let's make sure to store the data
681 * about STDERR as that's usually the best way to do logging. */
683 if (fstat(fileno
, &st
) >= 0 &&
684 (*journal_stream_ino
== 0 || fileno
== STDERR_FILENO
)) {
685 *journal_stream_dev
= st
.st_dev
;
686 *journal_stream_ino
= st
.st_ino
;
691 case EXEC_OUTPUT_SOCKET
:
692 assert(socket_fd
>= 0);
694 return dup2(socket_fd
, fileno
) < 0 ? -errno
: fileno
;
696 case EXEC_OUTPUT_NAMED_FD
:
697 assert(named_iofds
[fileno
] >= 0);
699 (void) fd_nonblock(named_iofds
[fileno
], false);
700 return dup2(named_iofds
[fileno
], fileno
) < 0 ? -errno
: fileno
;
702 case EXEC_OUTPUT_FILE
:
703 case EXEC_OUTPUT_FILE_APPEND
:
704 case EXEC_OUTPUT_FILE_TRUNCATE
: {
708 assert(context
->stdio_file
[fileno
]);
710 rw
= context
->std_input
== EXEC_INPUT_FILE
&&
711 streq_ptr(context
->stdio_file
[fileno
], context
->stdio_file
[STDIN_FILENO
]);
714 return dup2(STDIN_FILENO
, fileno
) < 0 ? -errno
: fileno
;
717 if (o
== EXEC_OUTPUT_FILE_APPEND
)
719 else if (o
== EXEC_OUTPUT_FILE_TRUNCATE
)
722 fd
= acquire_path(context
->stdio_file
[fileno
], flags
, 0666 & ~context
->umask
);
726 return move_fd(fd
, fileno
, 0);
730 assert_not_reached("Unknown error type");
734 static int chown_terminal(int fd
, uid_t uid
) {
739 /* Before we chown/chmod the TTY, let's ensure this is actually a tty */
740 if (isatty(fd
) < 1) {
741 if (IN_SET(errno
, EINVAL
, ENOTTY
))
742 return 0; /* not a tty */
747 /* This might fail. What matters are the results. */
748 r
= fchmod_and_chown(fd
, TTY_MODE
, uid
, GID_INVALID
);
755 static int setup_confirm_stdio(const char *vc
, int *_saved_stdin
, int *_saved_stdout
) {
756 _cleanup_close_
int fd
= -1, saved_stdin
= -1, saved_stdout
= -1;
759 assert(_saved_stdin
);
760 assert(_saved_stdout
);
762 saved_stdin
= fcntl(STDIN_FILENO
, F_DUPFD
, 3);
766 saved_stdout
= fcntl(STDOUT_FILENO
, F_DUPFD
, 3);
767 if (saved_stdout
< 0)
770 fd
= acquire_terminal(vc
, ACQUIRE_TERMINAL_WAIT
, DEFAULT_CONFIRM_USEC
);
774 r
= chown_terminal(fd
, getuid());
778 r
= reset_terminal_fd(fd
, true);
782 r
= rearrange_stdio(fd
, fd
, STDERR_FILENO
);
787 *_saved_stdin
= saved_stdin
;
788 *_saved_stdout
= saved_stdout
;
790 saved_stdin
= saved_stdout
= -1;
795 static void write_confirm_error_fd(int err
, int fd
, const Unit
*u
) {
798 if (err
== -ETIMEDOUT
)
799 dprintf(fd
, "Confirmation question timed out for %s, assuming positive response.\n", u
->id
);
802 dprintf(fd
, "Couldn't ask confirmation for %s: %m, assuming positive response.\n", u
->id
);
806 static void write_confirm_error(int err
, const char *vc
, const Unit
*u
) {
807 _cleanup_close_
int fd
= -1;
811 fd
= open_terminal(vc
, O_WRONLY
|O_NOCTTY
|O_CLOEXEC
);
815 write_confirm_error_fd(err
, fd
, u
);
818 static int restore_confirm_stdio(int *saved_stdin
, int *saved_stdout
) {
822 assert(saved_stdout
);
826 if (*saved_stdin
>= 0)
827 if (dup2(*saved_stdin
, STDIN_FILENO
) < 0)
830 if (*saved_stdout
>= 0)
831 if (dup2(*saved_stdout
, STDOUT_FILENO
) < 0)
834 *saved_stdin
= safe_close(*saved_stdin
);
835 *saved_stdout
= safe_close(*saved_stdout
);
841 CONFIRM_PRETEND_FAILURE
= -1,
842 CONFIRM_PRETEND_SUCCESS
= 0,
846 static int ask_for_confirmation(const char *vc
, Unit
*u
, const char *cmdline
) {
847 int saved_stdout
= -1, saved_stdin
= -1, r
;
848 _cleanup_free_
char *e
= NULL
;
851 /* For any internal errors, assume a positive response. */
852 r
= setup_confirm_stdio(vc
, &saved_stdin
, &saved_stdout
);
854 write_confirm_error(r
, vc
, u
);
855 return CONFIRM_EXECUTE
;
858 /* confirm_spawn might have been disabled while we were sleeping. */
859 if (manager_is_confirm_spawn_disabled(u
->manager
)) {
864 e
= ellipsize(cmdline
, 60, 100);
872 r
= ask_char(&c
, "yfshiDjcn", "Execute %s? [y, f, s – h for help] ", e
);
874 write_confirm_error_fd(r
, STDOUT_FILENO
, u
);
881 printf("Resuming normal execution.\n");
882 manager_disable_confirm_spawn();
886 unit_dump(u
, stdout
, " ");
887 continue; /* ask again */
889 printf("Failing execution.\n");
890 r
= CONFIRM_PRETEND_FAILURE
;
893 printf(" c - continue, proceed without asking anymore\n"
894 " D - dump, show the state of the unit\n"
895 " f - fail, don't execute the command and pretend it failed\n"
897 " i - info, show a short summary of the unit\n"
898 " j - jobs, show jobs that are in progress\n"
899 " s - skip, don't execute the command and pretend it succeeded\n"
900 " y - yes, execute the command\n");
901 continue; /* ask again */
903 printf(" Description: %s\n"
906 u
->id
, u
->description
, cmdline
);
907 continue; /* ask again */
909 manager_dump_jobs(u
->manager
, stdout
, " ");
910 continue; /* ask again */
912 /* 'n' was removed in favor of 'f'. */
913 printf("Didn't understand 'n', did you mean 'f'?\n");
914 continue; /* ask again */
916 printf("Skipping execution.\n");
917 r
= CONFIRM_PRETEND_SUCCESS
;
923 assert_not_reached("Unhandled choice");
929 restore_confirm_stdio(&saved_stdin
, &saved_stdout
);
933 static int get_fixed_user(const ExecContext
*c
, const char **user
,
934 uid_t
*uid
, gid_t
*gid
,
935 const char **home
, const char **shell
) {
944 /* Note that we don't set $HOME or $SHELL if they are not particularly enlightening anyway
945 * (i.e. are "/" or "/bin/nologin"). */
948 r
= get_user_creds(&name
, uid
, gid
, home
, shell
, USER_CREDS_CLEAN
);
956 static int get_fixed_group(const ExecContext
*c
, const char **group
, gid_t
*gid
) {
966 r
= get_group_creds(&name
, gid
, 0);
974 static int get_supplementary_groups(const ExecContext
*c
, const char *user
,
975 const char *group
, gid_t gid
,
976 gid_t
**supplementary_gids
, int *ngids
) {
980 bool keep_groups
= false;
981 gid_t
*groups
= NULL
;
982 _cleanup_free_ gid_t
*l_gids
= NULL
;
987 * If user is given, then lookup GID and supplementary groups list.
988 * We avoid NSS lookups for gid=0. Also we have to initialize groups
989 * here and as early as possible so we keep the list of supplementary
990 * groups of the caller.
992 if (user
&& gid_is_valid(gid
) && gid
!= 0) {
993 /* First step, initialize groups from /etc/groups */
994 if (initgroups(user
, gid
) < 0)
1000 if (strv_isempty(c
->supplementary_groups
))
1004 * If SupplementaryGroups= was passed then NGROUPS_MAX has to
1005 * be positive, otherwise fail.
1008 ngroups_max
= (int) sysconf(_SC_NGROUPS_MAX
);
1009 if (ngroups_max
<= 0)
1010 return errno_or_else(EOPNOTSUPP
);
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 set_securebits(int bits
, int mask
) {
1083 int current
, applied
;
1084 current
= prctl(PR_GET_SECUREBITS
);
1087 /* Clear all securebits defined in mask and set bits */
1088 applied
= (current
& ~mask
) | bits
;
1089 if (current
== applied
)
1091 if (prctl(PR_SET_SECUREBITS
, applied
) < 0)
1096 static int enforce_user(const ExecContext
*context
, uid_t uid
) {
1100 if (!uid_is_valid(uid
))
1103 /* Sets (but doesn't look up) the uid and make sure we keep the
1104 * capabilities while doing so. For setting secure bits the capability CAP_SETPCAP is
1105 * required, so we also need keep-caps in this case.
1108 if (context
->capability_ambient_set
!= 0 || context
->secure_bits
!= 0) {
1110 /* First step: If we need to keep capabilities but
1111 * drop privileges we need to make sure we keep our
1112 * caps, while we drop privileges. */
1114 /* Add KEEP_CAPS to the securebits */
1115 r
= set_securebits(1<<SECURE_KEEP_CAPS
, 0);
1121 /* Second step: actually set the uids */
1122 if (setresuid(uid
, uid
, uid
) < 0)
1125 /* At this point we should have all necessary capabilities but
1126 are otherwise a normal user. However, the caps might got
1127 corrupted due to the setresuid() so we need clean them up
1128 later. This is done outside of this call. */
1135 static int null_conv(
1137 const struct pam_message
**msg
,
1138 struct pam_response
**resp
,
1139 void *appdata_ptr
) {
1141 /* We don't support conversations */
1143 return PAM_CONV_ERR
;
1148 static int setup_pam(
1155 const int fds
[], size_t n_fds
) {
1159 static const struct pam_conv conv
= {
1164 _cleanup_(barrier_destroy
) Barrier barrier
= BARRIER_NULL
;
1165 pam_handle_t
*handle
= NULL
;
1167 int pam_code
= PAM_SUCCESS
, r
;
1168 char **nv
, **e
= NULL
;
1169 bool close_session
= false;
1170 pid_t pam_pid
= 0, parent_pid
;
1177 /* We set up PAM in the parent process, then fork. The child
1178 * will then stay around until killed via PR_GET_PDEATHSIG or
1179 * systemd via the cgroup logic. It will then remove the PAM
1180 * session again. The parent process will exec() the actual
1181 * daemon. We do things this way to ensure that the main PID
1182 * of the daemon is the one we initially fork()ed. */
1184 r
= barrier_create(&barrier
);
1188 if (log_get_max_level() < LOG_DEBUG
)
1189 flags
|= PAM_SILENT
;
1191 pam_code
= pam_start(name
, user
, &conv
, &handle
);
1192 if (pam_code
!= PAM_SUCCESS
) {
1198 _cleanup_free_
char *q
= NULL
;
1200 /* Hmm, so no TTY was explicitly passed, but an fd passed to us directly might be a TTY. Let's figure
1201 * out if that's the case, and read the TTY off it. */
1203 if (getttyname_malloc(STDIN_FILENO
, &q
) >= 0)
1204 tty
= strjoina("/dev/", q
);
1208 pam_code
= pam_set_item(handle
, PAM_TTY
, tty
);
1209 if (pam_code
!= PAM_SUCCESS
)
1213 STRV_FOREACH(nv
, *env
) {
1214 pam_code
= pam_putenv(handle
, *nv
);
1215 if (pam_code
!= PAM_SUCCESS
)
1219 pam_code
= pam_acct_mgmt(handle
, flags
);
1220 if (pam_code
!= PAM_SUCCESS
)
1223 pam_code
= pam_setcred(handle
, PAM_ESTABLISH_CRED
| flags
);
1224 if (pam_code
!= PAM_SUCCESS
)
1225 log_debug("pam_setcred() failed, ignoring: %s", pam_strerror(handle
, pam_code
));
1227 pam_code
= pam_open_session(handle
, flags
);
1228 if (pam_code
!= PAM_SUCCESS
)
1231 close_session
= true;
1233 e
= pam_getenvlist(handle
);
1235 pam_code
= PAM_BUF_ERR
;
1239 /* Block SIGTERM, so that we know that it won't get lost in
1242 assert_se(sigprocmask_many(SIG_BLOCK
, &old_ss
, SIGTERM
, -1) >= 0);
1244 parent_pid
= getpid_cached();
1246 r
= safe_fork("(sd-pam)", 0, &pam_pid
);
1250 int sig
, ret
= EXIT_PAM
;
1252 /* The child's job is to reset the PAM session on
1254 barrier_set_role(&barrier
, BARRIER_CHILD
);
1256 /* Make sure we don't keep open the passed fds in this child. We assume that otherwise only
1257 * those fds are open here that have been opened by PAM. */
1258 (void) close_many(fds
, n_fds
);
1260 /* Drop privileges - we don't need any to pam_close_session
1261 * and this will make PR_SET_PDEATHSIG work in most cases.
1262 * If this fails, ignore the error - but expect sd-pam threads
1263 * to fail to exit normally */
1265 r
= maybe_setgroups(0, NULL
);
1267 log_warning_errno(r
, "Failed to setgroups() in sd-pam: %m");
1268 if (setresgid(gid
, gid
, gid
) < 0)
1269 log_warning_errno(errno
, "Failed to setresgid() in sd-pam: %m");
1270 if (setresuid(uid
, uid
, uid
) < 0)
1271 log_warning_errno(errno
, "Failed to setresuid() in sd-pam: %m");
1273 (void) ignore_signals(SIGPIPE
);
1275 /* Wait until our parent died. This will only work if
1276 * the above setresuid() succeeds, otherwise the kernel
1277 * will not allow unprivileged parents kill their privileged
1278 * children this way. We rely on the control groups kill logic
1279 * to do the rest for us. */
1280 if (prctl(PR_SET_PDEATHSIG
, SIGTERM
) < 0)
1283 /* Tell the parent that our setup is done. This is especially
1284 * important regarding dropping privileges. Otherwise, unit
1285 * setup might race against our setresuid(2) call.
1287 * If the parent aborted, we'll detect this below, hence ignore
1288 * return failure here. */
1289 (void) barrier_place(&barrier
);
1291 /* Check if our parent process might already have died? */
1292 if (getppid() == parent_pid
) {
1295 assert_se(sigemptyset(&ss
) >= 0);
1296 assert_se(sigaddset(&ss
, SIGTERM
) >= 0);
1299 if (sigwait(&ss
, &sig
) < 0) {
1306 assert(sig
== SIGTERM
);
1311 pam_code
= pam_setcred(handle
, PAM_DELETE_CRED
| flags
);
1312 if (pam_code
!= PAM_SUCCESS
)
1315 /* If our parent died we'll end the session */
1316 if (getppid() != parent_pid
) {
1317 pam_code
= pam_close_session(handle
, flags
);
1318 if (pam_code
!= PAM_SUCCESS
)
1325 pam_end(handle
, pam_code
| flags
);
1329 barrier_set_role(&barrier
, BARRIER_PARENT
);
1331 /* If the child was forked off successfully it will do all the
1332 * cleanups, so forget about the handle here. */
1335 /* Unblock SIGTERM again in the parent */
1336 assert_se(sigprocmask(SIG_SETMASK
, &old_ss
, NULL
) >= 0);
1338 /* We close the log explicitly here, since the PAM modules
1339 * might have opened it, but we don't want this fd around. */
1342 /* Synchronously wait for the child to initialize. We don't care for
1343 * errors as we cannot recover. However, warn loudly if it happens. */
1344 if (!barrier_place_and_sync(&barrier
))
1345 log_error("PAM initialization failed");
1347 return strv_free_and_replace(*env
, e
);
1350 if (pam_code
!= PAM_SUCCESS
) {
1351 log_error("PAM failed: %s", pam_strerror(handle
, pam_code
));
1352 r
= -EPERM
; /* PAM errors do not map to errno */
1354 log_error_errno(r
, "PAM failed: %m");
1358 pam_code
= pam_close_session(handle
, flags
);
1360 pam_end(handle
, pam_code
| flags
);
1372 static void rename_process_from_path(const char *path
) {
1373 char process_name
[11];
1377 /* This resulting string must fit in 10 chars (i.e. the length
1378 * of "/sbin/init") to look pretty in /bin/ps */
1382 rename_process("(...)");
1388 /* The end of the process name is usually more
1389 * interesting, since the first bit might just be
1395 process_name
[0] = '(';
1396 memcpy(process_name
+1, p
, l
);
1397 process_name
[1+l
] = ')';
1398 process_name
[1+l
+1] = 0;
1400 rename_process(process_name
);
1403 static bool context_has_address_families(const ExecContext
*c
) {
1406 return c
->address_families_allow_list
||
1407 !set_isempty(c
->address_families
);
1410 static bool context_has_syscall_filters(const ExecContext
*c
) {
1413 return c
->syscall_allow_list
||
1414 !hashmap_isempty(c
->syscall_filter
);
1417 static bool context_has_syscall_logs(const ExecContext
*c
) {
1420 return c
->syscall_log_allow_list
||
1421 !hashmap_isempty(c
->syscall_log
);
1424 static bool context_has_no_new_privileges(const ExecContext
*c
) {
1427 if (c
->no_new_privileges
)
1430 if (have_effective_cap(CAP_SYS_ADMIN
)) /* if we are privileged, we don't need NNP */
1433 /* We need NNP if we have any form of seccomp and are unprivileged */
1434 return c
->lock_personality
||
1435 c
->memory_deny_write_execute
||
1436 c
->private_devices
||
1438 c
->protect_hostname
||
1439 c
->protect_kernel_tunables
||
1440 c
->protect_kernel_modules
||
1441 c
->protect_kernel_logs
||
1442 context_has_address_families(c
) ||
1443 exec_context_restrict_namespaces_set(c
) ||
1444 c
->restrict_realtime
||
1445 c
->restrict_suid_sgid
||
1446 !set_isempty(c
->syscall_archs
) ||
1447 context_has_syscall_filters(c
) ||
1448 context_has_syscall_logs(c
);
1451 static bool exec_context_has_credentials(const ExecContext
*context
) {
1455 return !hashmap_isempty(context
->set_credentials
) ||
1456 context
->load_credentials
;
1461 static bool skip_seccomp_unavailable(const Unit
* u
, const char* msg
) {
1463 if (is_seccomp_available())
1466 log_unit_debug(u
, "SECCOMP features not detected in the kernel, skipping %s", msg
);
1470 static int apply_syscall_filter(const Unit
* u
, const ExecContext
*c
, bool needs_ambient_hack
) {
1471 uint32_t negative_action
, default_action
, action
;
1477 if (!context_has_syscall_filters(c
))
1480 if (skip_seccomp_unavailable(u
, "SystemCallFilter="))
1483 negative_action
= c
->syscall_errno
== SECCOMP_ERROR_NUMBER_KILL
? scmp_act_kill_process() : SCMP_ACT_ERRNO(c
->syscall_errno
);
1485 if (c
->syscall_allow_list
) {
1486 default_action
= negative_action
;
1487 action
= SCMP_ACT_ALLOW
;
1489 default_action
= SCMP_ACT_ALLOW
;
1490 action
= negative_action
;
1493 if (needs_ambient_hack
) {
1494 r
= seccomp_filter_set_add(c
->syscall_filter
, c
->syscall_allow_list
, syscall_filter_sets
+ SYSCALL_FILTER_SET_SETUID
);
1499 return seccomp_load_syscall_filter_set_raw(default_action
, c
->syscall_filter
, action
, false);
1502 static int apply_syscall_log(const Unit
* u
, const ExecContext
*c
) {
1504 uint32_t default_action
, action
;
1510 if (!context_has_syscall_logs(c
))
1514 if (skip_seccomp_unavailable(u
, "SystemCallLog="))
1517 if (c
->syscall_log_allow_list
) {
1518 /* Log nothing but the ones listed */
1519 default_action
= SCMP_ACT_ALLOW
;
1520 action
= SCMP_ACT_LOG
;
1522 /* Log everything but the ones listed */
1523 default_action
= SCMP_ACT_LOG
;
1524 action
= SCMP_ACT_ALLOW
;
1527 return seccomp_load_syscall_filter_set_raw(default_action
, c
->syscall_log
, action
, false);
1529 /* old libseccomp */
1530 log_unit_debug(u
, "SECCOMP feature SCMP_ACT_LOG not available, skipping SystemCallLog=");
1535 static int apply_syscall_archs(const Unit
*u
, const ExecContext
*c
) {
1539 if (set_isempty(c
->syscall_archs
))
1542 if (skip_seccomp_unavailable(u
, "SystemCallArchitectures="))
1545 return seccomp_restrict_archs(c
->syscall_archs
);
1548 static int apply_address_families(const Unit
* u
, const ExecContext
*c
) {
1552 if (!context_has_address_families(c
))
1555 if (skip_seccomp_unavailable(u
, "RestrictAddressFamilies="))
1558 return seccomp_restrict_address_families(c
->address_families
, c
->address_families_allow_list
);
1561 static int apply_memory_deny_write_execute(const Unit
* u
, const ExecContext
*c
) {
1565 if (!c
->memory_deny_write_execute
)
1568 if (skip_seccomp_unavailable(u
, "MemoryDenyWriteExecute="))
1571 return seccomp_memory_deny_write_execute();
1574 static int apply_restrict_realtime(const Unit
* u
, const ExecContext
*c
) {
1578 if (!c
->restrict_realtime
)
1581 if (skip_seccomp_unavailable(u
, "RestrictRealtime="))
1584 return seccomp_restrict_realtime();
1587 static int apply_restrict_suid_sgid(const Unit
* u
, const ExecContext
*c
) {
1591 if (!c
->restrict_suid_sgid
)
1594 if (skip_seccomp_unavailable(u
, "RestrictSUIDSGID="))
1597 return seccomp_restrict_suid_sgid();
1600 static int apply_protect_sysctl(const Unit
*u
, const ExecContext
*c
) {
1604 /* Turn off the legacy sysctl() system call. Many distributions turn this off while building the kernel, but
1605 * let's protect even those systems where this is left on in the kernel. */
1607 if (!c
->protect_kernel_tunables
)
1610 if (skip_seccomp_unavailable(u
, "ProtectKernelTunables="))
1613 return seccomp_protect_sysctl();
1616 static int apply_protect_kernel_modules(const Unit
*u
, const ExecContext
*c
) {
1620 /* Turn off module syscalls on ProtectKernelModules=yes */
1622 if (!c
->protect_kernel_modules
)
1625 if (skip_seccomp_unavailable(u
, "ProtectKernelModules="))
1628 return seccomp_load_syscall_filter_set(SCMP_ACT_ALLOW
, syscall_filter_sets
+ SYSCALL_FILTER_SET_MODULE
, SCMP_ACT_ERRNO(EPERM
), false);
1631 static int apply_protect_kernel_logs(const Unit
*u
, const ExecContext
*c
) {
1635 if (!c
->protect_kernel_logs
)
1638 if (skip_seccomp_unavailable(u
, "ProtectKernelLogs="))
1641 return seccomp_protect_syslog();
1644 static int apply_protect_clock(const Unit
*u
, const ExecContext
*c
) {
1648 if (!c
->protect_clock
)
1651 if (skip_seccomp_unavailable(u
, "ProtectClock="))
1654 return seccomp_load_syscall_filter_set(SCMP_ACT_ALLOW
, syscall_filter_sets
+ SYSCALL_FILTER_SET_CLOCK
, SCMP_ACT_ERRNO(EPERM
), false);
1657 static int apply_private_devices(const Unit
*u
, const ExecContext
*c
) {
1661 /* If PrivateDevices= is set, also turn off iopl and all @raw-io syscalls. */
1663 if (!c
->private_devices
)
1666 if (skip_seccomp_unavailable(u
, "PrivateDevices="))
1669 return seccomp_load_syscall_filter_set(SCMP_ACT_ALLOW
, syscall_filter_sets
+ SYSCALL_FILTER_SET_RAW_IO
, SCMP_ACT_ERRNO(EPERM
), false);
1672 static int apply_restrict_namespaces(const Unit
*u
, const ExecContext
*c
) {
1676 if (!exec_context_restrict_namespaces_set(c
))
1679 if (skip_seccomp_unavailable(u
, "RestrictNamespaces="))
1682 return seccomp_restrict_namespaces(c
->restrict_namespaces
);
1685 static int apply_lock_personality(const Unit
* u
, const ExecContext
*c
) {
1686 unsigned long personality
;
1692 if (!c
->lock_personality
)
1695 if (skip_seccomp_unavailable(u
, "LockPersonality="))
1698 personality
= c
->personality
;
1700 /* If personality is not specified, use either PER_LINUX or PER_LINUX32 depending on what is currently set. */
1701 if (personality
== PERSONALITY_INVALID
) {
1703 r
= opinionated_personality(&personality
);
1708 return seccomp_lock_personality(personality
);
1713 static int apply_protect_hostname(const Unit
*u
, const ExecContext
*c
, int *ret_exit_status
) {
1717 if (!c
->protect_hostname
)
1720 if (ns_type_supported(NAMESPACE_UTS
)) {
1721 if (unshare(CLONE_NEWUTS
) < 0) {
1722 if (!ERRNO_IS_NOT_SUPPORTED(errno
) && !ERRNO_IS_PRIVILEGE(errno
)) {
1723 *ret_exit_status
= EXIT_NAMESPACE
;
1724 return log_unit_error_errno(u
, errno
, "Failed to set up UTS namespacing: %m");
1727 log_unit_warning(u
, "ProtectHostname=yes is configured, but UTS namespace setup is prohibited (container manager?), ignoring namespace setup.");
1730 log_unit_warning(u
, "ProtectHostname=yes is configured, but the kernel does not support UTS namespaces, ignoring namespace setup.");
1735 if (skip_seccomp_unavailable(u
, "ProtectHostname="))
1738 r
= seccomp_protect_hostname();
1740 *ret_exit_status
= EXIT_SECCOMP
;
1741 return log_unit_error_errno(u
, r
, "Failed to apply hostname restrictions: %m");
1748 static void do_idle_pipe_dance(int idle_pipe
[static 4]) {
1751 idle_pipe
[1] = safe_close(idle_pipe
[1]);
1752 idle_pipe
[2] = safe_close(idle_pipe
[2]);
1754 if (idle_pipe
[0] >= 0) {
1757 r
= fd_wait_for_event(idle_pipe
[0], POLLHUP
, IDLE_TIMEOUT_USEC
);
1759 if (idle_pipe
[3] >= 0 && r
== 0 /* timeout */) {
1762 /* Signal systemd that we are bored and want to continue. */
1763 n
= write(idle_pipe
[3], "x", 1);
1765 /* Wait for systemd to react to the signal above. */
1766 (void) fd_wait_for_event(idle_pipe
[0], POLLHUP
, IDLE_TIMEOUT2_USEC
);
1769 idle_pipe
[0] = safe_close(idle_pipe
[0]);
1773 idle_pipe
[3] = safe_close(idle_pipe
[3]);
1776 static const char *exec_directory_env_name_to_string(ExecDirectoryType t
);
1778 static int build_environment(
1780 const ExecContext
*c
,
1781 const ExecParameters
*p
,
1784 const char *username
,
1786 dev_t journal_stream_dev
,
1787 ino_t journal_stream_ino
,
1790 _cleanup_strv_free_
char **our_env
= NULL
;
1799 #define N_ENV_VARS 17
1800 our_env
= new0(char*, N_ENV_VARS
+ _EXEC_DIRECTORY_TYPE_MAX
);
1805 _cleanup_free_
char *joined
= NULL
;
1807 if (asprintf(&x
, "LISTEN_PID="PID_FMT
, getpid_cached()) < 0)
1809 our_env
[n_env
++] = x
;
1811 if (asprintf(&x
, "LISTEN_FDS=%zu", n_fds
) < 0)
1813 our_env
[n_env
++] = x
;
1815 joined
= strv_join(p
->fd_names
, ":");
1819 x
= strjoin("LISTEN_FDNAMES=", joined
);
1822 our_env
[n_env
++] = x
;
1825 if ((p
->flags
& EXEC_SET_WATCHDOG
) && p
->watchdog_usec
> 0) {
1826 if (asprintf(&x
, "WATCHDOG_PID="PID_FMT
, getpid_cached()) < 0)
1828 our_env
[n_env
++] = x
;
1830 if (asprintf(&x
, "WATCHDOG_USEC="USEC_FMT
, p
->watchdog_usec
) < 0)
1832 our_env
[n_env
++] = x
;
1835 /* If this is D-Bus, tell the nss-systemd module, since it relies on being able to use D-Bus look up dynamic
1836 * users via PID 1, possibly dead-locking the dbus daemon. This way it will not use D-Bus to resolve names, but
1837 * check the database directly. */
1838 if (p
->flags
& EXEC_NSS_BYPASS_BUS
) {
1839 x
= strdup("SYSTEMD_NSS_BYPASS_BUS=1");
1842 our_env
[n_env
++] = x
;
1846 x
= strjoin("HOME=", home
);
1850 path_simplify(x
+ 5);
1851 our_env
[n_env
++] = x
;
1855 x
= strjoin("LOGNAME=", username
);
1858 our_env
[n_env
++] = x
;
1860 x
= strjoin("USER=", username
);
1863 our_env
[n_env
++] = x
;
1867 x
= strjoin("SHELL=", shell
);
1871 path_simplify(x
+ 6);
1872 our_env
[n_env
++] = x
;
1875 if (!sd_id128_is_null(u
->invocation_id
)) {
1876 if (asprintf(&x
, "INVOCATION_ID=" SD_ID128_FORMAT_STR
, SD_ID128_FORMAT_VAL(u
->invocation_id
)) < 0)
1879 our_env
[n_env
++] = x
;
1882 if (exec_context_needs_term(c
)) {
1883 const char *tty_path
, *term
= NULL
;
1885 tty_path
= exec_context_tty_path(c
);
1887 /* If we are forked off PID 1 and we are supposed to operate on /dev/console, then let's try
1888 * to inherit the $TERM set for PID 1. This is useful for containers so that the $TERM the
1889 * container manager passes to PID 1 ends up all the way in the console login shown. */
1891 if (path_equal_ptr(tty_path
, "/dev/console") && getppid() == 1)
1892 term
= getenv("TERM");
1895 term
= default_term_for_tty(tty_path
);
1897 x
= strjoin("TERM=", term
);
1900 our_env
[n_env
++] = x
;
1903 if (journal_stream_dev
!= 0 && journal_stream_ino
!= 0) {
1904 if (asprintf(&x
, "JOURNAL_STREAM=" DEV_FMT
":" INO_FMT
, journal_stream_dev
, journal_stream_ino
) < 0)
1907 our_env
[n_env
++] = x
;
1910 if (c
->log_namespace
) {
1911 x
= strjoin("LOG_NAMESPACE=", c
->log_namespace
);
1915 our_env
[n_env
++] = x
;
1918 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
1919 _cleanup_free_
char *pre
= NULL
, *joined
= NULL
;
1925 if (strv_isempty(c
->directories
[t
].paths
))
1928 n
= exec_directory_env_name_to_string(t
);
1932 pre
= strjoin(p
->prefix
[t
], "/");
1936 joined
= strv_join_full(c
->directories
[t
].paths
, ":", pre
, true);
1940 x
= strjoin(n
, "=", joined
);
1944 our_env
[n_env
++] = x
;
1947 if (exec_context_has_credentials(c
) && p
->prefix
[EXEC_DIRECTORY_RUNTIME
]) {
1948 x
= strjoin("CREDENTIALS_DIRECTORY=", p
->prefix
[EXEC_DIRECTORY_RUNTIME
], "/credentials/", u
->id
);
1952 our_env
[n_env
++] = x
;
1955 if (asprintf(&x
, "SYSTEMD_EXEC_PID=" PID_FMT
, getpid_cached()) < 0)
1958 our_env
[n_env
++] = x
;
1960 our_env
[n_env
++] = NULL
;
1961 assert(n_env
<= N_ENV_VARS
+ _EXEC_DIRECTORY_TYPE_MAX
);
1964 *ret
= TAKE_PTR(our_env
);
1969 static int build_pass_environment(const ExecContext
*c
, char ***ret
) {
1970 _cleanup_strv_free_
char **pass_env
= NULL
;
1974 STRV_FOREACH(i
, c
->pass_environment
) {
1975 _cleanup_free_
char *x
= NULL
;
1981 x
= strjoin(*i
, "=", v
);
1985 if (!GREEDY_REALLOC(pass_env
, n_env
+ 2))
1988 pass_env
[n_env
++] = TAKE_PTR(x
);
1989 pass_env
[n_env
] = NULL
;
1992 *ret
= TAKE_PTR(pass_env
);
1997 bool exec_needs_mount_namespace(
1998 const ExecContext
*context
,
1999 const ExecParameters
*params
,
2000 const ExecRuntime
*runtime
) {
2004 if (context
->root_image
)
2007 if (!strv_isempty(context
->read_write_paths
) ||
2008 !strv_isempty(context
->read_only_paths
) ||
2009 !strv_isempty(context
->inaccessible_paths
) ||
2010 !strv_isempty(context
->exec_paths
) ||
2011 !strv_isempty(context
->no_exec_paths
))
2014 if (context
->n_bind_mounts
> 0)
2017 if (context
->n_temporary_filesystems
> 0)
2020 if (context
->n_mount_images
> 0)
2023 if (context
->n_extension_images
> 0)
2026 if (!IN_SET(context
->mount_flags
, 0, MS_SHARED
))
2029 if (context
->private_tmp
&& runtime
&& (runtime
->tmp_dir
|| runtime
->var_tmp_dir
))
2032 if (context
->private_devices
||
2033 context
->private_mounts
||
2034 context
->protect_system
!= PROTECT_SYSTEM_NO
||
2035 context
->protect_home
!= PROTECT_HOME_NO
||
2036 context
->protect_kernel_tunables
||
2037 context
->protect_kernel_modules
||
2038 context
->protect_kernel_logs
||
2039 context
->protect_control_groups
||
2040 context
->protect_proc
!= PROTECT_PROC_DEFAULT
||
2041 context
->proc_subset
!= PROC_SUBSET_ALL
||
2042 context
->private_ipc
||
2043 context
->ipc_namespace_path
)
2046 if (context
->root_directory
) {
2047 if (exec_context_get_effective_mount_apivfs(context
))
2050 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
2051 if (params
&& !params
->prefix
[t
])
2054 if (!strv_isempty(context
->directories
[t
].paths
))
2059 if (context
->dynamic_user
&&
2060 (!strv_isempty(context
->directories
[EXEC_DIRECTORY_STATE
].paths
) ||
2061 !strv_isempty(context
->directories
[EXEC_DIRECTORY_CACHE
].paths
) ||
2062 !strv_isempty(context
->directories
[EXEC_DIRECTORY_LOGS
].paths
)))
2065 if (context
->log_namespace
)
2071 static int setup_private_users(uid_t ouid
, gid_t ogid
, uid_t uid
, gid_t gid
) {
2072 _cleanup_free_
char *uid_map
= NULL
, *gid_map
= NULL
;
2073 _cleanup_close_pair_
int errno_pipe
[2] = { -1, -1 };
2074 _cleanup_close_
int unshare_ready_fd
= -1;
2075 _cleanup_(sigkill_waitp
) pid_t pid
= 0;
2080 /* Set up a user namespace and map the original UID/GID (IDs from before any user or group changes, i.e.
2081 * the IDs from the user or system manager(s)) to itself, the selected UID/GID to itself, and everything else to
2082 * nobody. In order to be able to write this mapping we need CAP_SETUID in the original user namespace, which
2083 * we however lack after opening the user namespace. To work around this we fork() a temporary child process,
2084 * which waits for the parent to create the new user namespace while staying in the original namespace. The
2085 * child then writes the UID mapping, under full privileges. The parent waits for the child to finish and
2086 * continues execution normally.
2087 * For unprivileged users (i.e. without capabilities), the root to root mapping is excluded. As such, it
2088 * does not need CAP_SETUID to write the single line mapping to itself. */
2090 /* Can only set up multiple mappings with CAP_SETUID. */
2091 if (have_effective_cap(CAP_SETUID
) && uid
!= ouid
&& uid_is_valid(uid
))
2092 r
= asprintf(&uid_map
,
2093 UID_FMT
" " UID_FMT
" 1\n" /* Map $OUID → $OUID */
2094 UID_FMT
" " UID_FMT
" 1\n", /* Map $UID → $UID */
2095 ouid
, ouid
, uid
, uid
);
2097 r
= asprintf(&uid_map
,
2098 UID_FMT
" " UID_FMT
" 1\n", /* Map $OUID → $OUID */
2104 /* Can only set up multiple mappings with CAP_SETGID. */
2105 if (have_effective_cap(CAP_SETGID
) && gid
!= ogid
&& gid_is_valid(gid
))
2106 r
= asprintf(&gid_map
,
2107 GID_FMT
" " GID_FMT
" 1\n" /* Map $OGID → $OGID */
2108 GID_FMT
" " GID_FMT
" 1\n", /* Map $GID → $GID */
2109 ogid
, ogid
, gid
, gid
);
2111 r
= asprintf(&gid_map
,
2112 GID_FMT
" " GID_FMT
" 1\n", /* Map $OGID -> $OGID */
2118 /* Create a communication channel so that the parent can tell the child when it finished creating the user
2120 unshare_ready_fd
= eventfd(0, EFD_CLOEXEC
);
2121 if (unshare_ready_fd
< 0)
2124 /* Create a communication channel so that the child can tell the parent a proper error code in case it
2126 if (pipe2(errno_pipe
, O_CLOEXEC
) < 0)
2129 r
= safe_fork("(sd-userns)", FORK_RESET_SIGNALS
|FORK_DEATHSIG
, &pid
);
2133 _cleanup_close_
int fd
= -1;
2137 /* Child process, running in the original user namespace. Let's update the parent's UID/GID map from
2138 * here, after the parent opened its own user namespace. */
2141 errno_pipe
[0] = safe_close(errno_pipe
[0]);
2143 /* Wait until the parent unshared the user namespace */
2144 if (read(unshare_ready_fd
, &c
, sizeof(c
)) < 0) {
2149 /* Disable the setgroups() system call in the child user namespace, for good. */
2150 a
= procfs_file_alloca(ppid
, "setgroups");
2151 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
2153 if (errno
!= ENOENT
) {
2158 /* If the file is missing the kernel is too old, let's continue anyway. */
2160 if (write(fd
, "deny\n", 5) < 0) {
2165 fd
= safe_close(fd
);
2168 /* First write the GID map */
2169 a
= procfs_file_alloca(ppid
, "gid_map");
2170 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
2175 if (write(fd
, gid_map
, strlen(gid_map
)) < 0) {
2179 fd
= safe_close(fd
);
2181 /* The write the UID map */
2182 a
= procfs_file_alloca(ppid
, "uid_map");
2183 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
2188 if (write(fd
, uid_map
, strlen(uid_map
)) < 0) {
2193 _exit(EXIT_SUCCESS
);
2196 (void) write(errno_pipe
[1], &r
, sizeof(r
));
2197 _exit(EXIT_FAILURE
);
2200 errno_pipe
[1] = safe_close(errno_pipe
[1]);
2202 if (unshare(CLONE_NEWUSER
) < 0)
2205 /* Let the child know that the namespace is ready now */
2206 if (write(unshare_ready_fd
, &c
, sizeof(c
)) < 0)
2209 /* Try to read an error code from the child */
2210 n
= read(errno_pipe
[0], &r
, sizeof(r
));
2213 if (n
== sizeof(r
)) { /* an error code was sent to us */
2218 if (n
!= 0) /* on success we should have read 0 bytes */
2221 r
= wait_for_terminate_and_check("(sd-userns)", pid
, 0);
2225 if (r
!= EXIT_SUCCESS
) /* If something strange happened with the child, let's consider this fatal, too */
2231 static bool exec_directory_is_private(const ExecContext
*context
, ExecDirectoryType type
) {
2232 if (!context
->dynamic_user
)
2235 if (type
== EXEC_DIRECTORY_CONFIGURATION
)
2238 if (type
== EXEC_DIRECTORY_RUNTIME
&& context
->runtime_directory_preserve_mode
== EXEC_PRESERVE_NO
)
2244 static int setup_exec_directory(
2245 const ExecContext
*context
,
2246 const ExecParameters
*params
,
2249 ExecDirectoryType type
,
2252 static const int exit_status_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
2253 [EXEC_DIRECTORY_RUNTIME
] = EXIT_RUNTIME_DIRECTORY
,
2254 [EXEC_DIRECTORY_STATE
] = EXIT_STATE_DIRECTORY
,
2255 [EXEC_DIRECTORY_CACHE
] = EXIT_CACHE_DIRECTORY
,
2256 [EXEC_DIRECTORY_LOGS
] = EXIT_LOGS_DIRECTORY
,
2257 [EXEC_DIRECTORY_CONFIGURATION
] = EXIT_CONFIGURATION_DIRECTORY
,
2264 assert(type
>= 0 && type
< _EXEC_DIRECTORY_TYPE_MAX
);
2265 assert(exit_status
);
2267 if (!params
->prefix
[type
])
2270 if (params
->flags
& EXEC_CHOWN_DIRECTORIES
) {
2271 if (!uid_is_valid(uid
))
2273 if (!gid_is_valid(gid
))
2277 STRV_FOREACH(rt
, context
->directories
[type
].paths
) {
2278 _cleanup_free_
char *p
= NULL
, *pp
= NULL
;
2280 p
= path_join(params
->prefix
[type
], *rt
);
2286 r
= mkdir_parents_label(p
, 0755);
2290 if (exec_directory_is_private(context
, type
)) {
2291 /* So, here's one extra complication when dealing with DynamicUser=1 units. In that
2292 * case we want to avoid leaving a directory around fully accessible that is owned by
2293 * a dynamic user whose UID is later on reused. To lock this down we use the same
2294 * trick used by container managers to prohibit host users to get access to files of
2295 * the same UID in containers: we place everything inside a directory that has an
2296 * access mode of 0700 and is owned root:root, so that it acts as security boundary
2297 * for unprivileged host code. We then use fs namespacing to make this directory
2298 * permeable for the service itself.
2300 * Specifically: for a service which wants a special directory "foo/" we first create
2301 * a directory "private/" with access mode 0700 owned by root:root. Then we place
2302 * "foo" inside of that directory (i.e. "private/foo/"), and make "foo" a symlink to
2303 * "private/foo". This way, privileged host users can access "foo/" as usual, but
2304 * unprivileged host users can't look into it. Inside of the namespace of the unit
2305 * "private/" is replaced by a more liberally accessible tmpfs, into which the host's
2306 * "private/foo/" is mounted under the same name, thus disabling the access boundary
2307 * for the service and making sure it only gets access to the dirs it needs but no
2308 * others. Tricky? Yes, absolutely, but it works!
2310 * Note that we don't do this for EXEC_DIRECTORY_CONFIGURATION as that's assumed not
2311 * to be owned by the service itself.
2313 * Also, note that we don't do this for EXEC_DIRECTORY_RUNTIME as that's often used
2314 * for sharing files or sockets with other services. */
2316 pp
= path_join(params
->prefix
[type
], "private");
2322 /* First set up private root if it doesn't exist yet, with access mode 0700 and owned by root:root */
2323 r
= mkdir_safe_label(pp
, 0700, 0, 0, MKDIR_WARN_MODE
);
2327 if (!path_extend(&pp
, *rt
)) {
2332 /* Create all directories between the configured directory and this private root, and mark them 0755 */
2333 r
= mkdir_parents_label(pp
, 0755);
2337 if (is_dir(p
, false) > 0 &&
2338 (laccess(pp
, F_OK
) < 0 && errno
== ENOENT
)) {
2340 /* Hmm, the private directory doesn't exist yet, but the normal one exists? If so, move
2341 * it over. Most likely the service has been upgraded from one that didn't use
2342 * DynamicUser=1, to one that does. */
2344 log_info("Found pre-existing public %s= directory %s, migrating to %s.\n"
2345 "Apparently, service previously had DynamicUser= turned off, and has now turned it on.",
2346 exec_directory_type_to_string(type
), p
, pp
);
2348 if (rename(p
, pp
) < 0) {
2353 /* Otherwise, create the actual directory for the service */
2355 r
= mkdir_label(pp
, context
->directories
[type
].mode
);
2356 if (r
< 0 && r
!= -EEXIST
)
2360 /* And link it up from the original place */
2361 r
= symlink_idempotent(pp
, p
, true);
2366 _cleanup_free_
char *target
= NULL
;
2368 if (type
!= EXEC_DIRECTORY_CONFIGURATION
&&
2369 readlink_and_make_absolute(p
, &target
) >= 0) {
2370 _cleanup_free_
char *q
= NULL
, *q_resolved
= NULL
, *target_resolved
= NULL
;
2372 /* This already exists and is a symlink? Interesting. Maybe it's one created
2373 * by DynamicUser=1 (see above)?
2375 * We do this for all directory types except for ConfigurationDirectory=,
2376 * since they all support the private/ symlink logic at least in some
2377 * configurations, see above. */
2379 r
= chase_symlinks(target
, NULL
, 0, &target_resolved
, NULL
);
2383 q
= path_join(params
->prefix
[type
], "private", *rt
);
2389 /* /var/lib or friends may be symlinks. So, let's chase them also. */
2390 r
= chase_symlinks(q
, NULL
, CHASE_NONEXISTENT
, &q_resolved
, NULL
);
2394 if (path_equal(q_resolved
, target_resolved
)) {
2396 /* Hmm, apparently DynamicUser= was once turned on for this service,
2397 * but is no longer. Let's move the directory back up. */
2399 log_info("Found pre-existing private %s= directory %s, migrating to %s.\n"
2400 "Apparently, service previously had DynamicUser= turned on, and has now turned it off.",
2401 exec_directory_type_to_string(type
), q
, p
);
2403 if (unlink(p
) < 0) {
2408 if (rename(q
, p
) < 0) {
2415 r
= mkdir_label(p
, context
->directories
[type
].mode
);
2420 if (type
== EXEC_DIRECTORY_CONFIGURATION
) {
2423 /* Don't change the owner/access mode of the configuration directory,
2424 * as in the common case it is not written to by a service, and shall
2425 * not be writable. */
2427 if (stat(p
, &st
) < 0) {
2432 /* Still complain if the access mode doesn't match */
2433 if (((st
.st_mode
^ context
->directories
[type
].mode
) & 07777) != 0)
2434 log_warning("%s \'%s\' already exists but the mode is different. "
2435 "(File system: %o %sMode: %o)",
2436 exec_directory_type_to_string(type
), *rt
,
2437 st
.st_mode
& 07777, exec_directory_type_to_string(type
), context
->directories
[type
].mode
& 07777);
2444 /* Lock down the access mode (we use chmod_and_chown() to make this idempotent. We don't
2445 * specify UID/GID here, so that path_chown_recursive() can optimize things depending on the
2446 * current UID/GID ownership.) */
2447 r
= chmod_and_chown(pp
?: p
, context
->directories
[type
].mode
, UID_INVALID
, GID_INVALID
);
2451 /* Then, change the ownership of the whole tree, if necessary. When dynamic users are used we
2452 * drop the suid/sgid bits, since we really don't want SUID/SGID files for dynamic UID/GID
2453 * assignments to exist. */
2454 r
= path_chown_recursive(pp
?: p
, uid
, gid
, context
->dynamic_user
? 01777 : 07777);
2462 *exit_status
= exit_status_table
[type
];
2466 static int write_credential(
2472 bool ownership_ok
) {
2474 _cleanup_(unlink_and_freep
) char *tmp
= NULL
;
2475 _cleanup_close_
int fd
= -1;
2478 r
= tempfn_random_child("", "cred", &tmp
);
2482 fd
= openat(dfd
, tmp
, O_CREAT
|O_RDWR
|O_CLOEXEC
|O_EXCL
|O_NOFOLLOW
|O_NOCTTY
, 0600);
2488 r
= loop_write(fd
, data
, size
, /* do_pool = */ false);
2492 if (fchmod(fd
, 0400) < 0) /* Take away "w" bit */
2495 if (uid_is_valid(uid
) && uid
!= getuid()) {
2496 r
= fd_add_uid_acl_permission(fd
, uid
, ACL_READ
);
2498 if (!ERRNO_IS_NOT_SUPPORTED(r
) && !ERRNO_IS_PRIVILEGE(r
))
2501 if (!ownership_ok
) /* Ideally we use ACLs, since we can neatly express what we want
2502 * to express: that the user gets read access and nothing
2503 * else. But if the backing fs can't support that (e.g. ramfs)
2504 * then we can use file ownership instead. But that's only safe if
2505 * we can then re-mount the whole thing read-only, so that the
2506 * user can no longer chmod() the file to gain write access. */
2509 if (fchown(fd
, uid
, GID_INVALID
) < 0)
2514 if (renameat(dfd
, tmp
, dfd
, id
) < 0)
2521 #define CREDENTIALS_BYTES_MAX (1024LU * 1024LU) /* Refuse to pass more than 1M, after all this is unswappable memory */
2523 static int acquire_credentials(
2524 const ExecContext
*context
,
2525 const ExecParameters
*params
,
2529 bool ownership_ok
) {
2531 uint64_t left
= CREDENTIALS_BYTES_MAX
;
2532 _cleanup_close_
int dfd
= -1;
2533 ExecSetCredential
*sc
;
2540 dfd
= open(p
, O_DIRECTORY
|O_CLOEXEC
);
2544 /* First we use the literally specified credentials. Note that they might be overridden again below,
2545 * and thus act as a "default" if the same credential is specified multiple times */
2546 HASHMAP_FOREACH(sc
, context
->set_credentials
) {
2549 add
= strlen(sc
->id
) + sc
->size
;
2553 r
= write_credential(dfd
, sc
->id
, sc
->data
, sc
->size
, uid
, ownership_ok
);
2560 /* Then, load credential off disk (or acquire via AF_UNIX socket) */
2561 STRV_FOREACH_PAIR(id
, fn
, context
->load_credentials
) {
2562 ReadFullFileFlags flags
= READ_FULL_FILE_SECURE
;
2563 _cleanup_(erase_and_freep
) char *data
= NULL
;
2564 _cleanup_free_
char *j
= NULL
, *bindname
= NULL
;
2565 bool missing_ok
= true;
2569 if (path_is_absolute(*fn
)) {
2570 /* If this is an absolute path, read the data directly from it, and support AF_UNIX sockets */
2572 flags
|= READ_FULL_FILE_CONNECT_SOCKET
;
2574 /* Pass some minimal info about the unit and the credential name we are looking to acquire
2575 * via the source socket address in case we read off an AF_UNIX socket. */
2576 if (asprintf(&bindname
, "@%" PRIx64
"/unit/%s/%s", random_u64(), unit
, *id
) < 0)
2581 } else if (params
->received_credentials
) {
2582 /* If this is a relative path, take it relative to the credentials we received
2583 * ourselves. We don't support the AF_UNIX stuff in this mode, since we are operating
2584 * on a credential store, i.e. this is guaranteed to be regular files. */
2585 j
= path_join(params
->received_credentials
, *fn
);
2594 r
= read_full_file_full(AT_FDCWD
, source
, UINT64_MAX
, SIZE_MAX
, flags
, bindname
, &data
, &size
);
2597 if (r
== -ENOENT
&& (missing_ok
|| faccessat(dfd
, *id
, F_OK
, AT_SYMLINK_NOFOLLOW
) >= 0)) {
2598 /* Make a missing inherited credential non-fatal, let's just continue. After all apps
2599 * will get clear errors if we don't pass such a missing credential on as they
2600 * themselves will get ENOENT when trying to read them, which should not be much
2601 * worse than when we handle the error here and make it fatal.
2603 * Also, if the source file doesn't exist, but we already acquired the key otherwise,
2604 * then don't fail either. */
2605 log_debug_errno(r
, "Couldn't read inherited credential '%s', skipping: %m", *fn
);
2609 return log_debug_errno(r
, "Failed to read credential '%s': %m", *fn
);
2611 add
= strlen(*id
) + size
;
2615 r
= write_credential(dfd
, *id
, data
, size
, uid
, ownership_ok
);
2622 if (fchmod(dfd
, 0500) < 0) /* Now take away the "w" bit */
2625 /* After we created all keys with the right perms, also make sure the credential store as a whole is
2628 if (uid_is_valid(uid
) && uid
!= getuid()) {
2629 r
= fd_add_uid_acl_permission(dfd
, uid
, ACL_READ
| ACL_EXECUTE
);
2631 if (!ERRNO_IS_NOT_SUPPORTED(r
) && !ERRNO_IS_PRIVILEGE(r
))
2637 if (fchown(dfd
, uid
, GID_INVALID
) < 0)
2645 static int setup_credentials_internal(
2646 const ExecContext
*context
,
2647 const ExecParameters
*params
,
2649 const char *final
, /* This is where the credential store shall eventually end up at */
2650 const char *workspace
, /* This is where we can prepare it before moving it to the final place */
2651 bool reuse_workspace
, /* Whether to reuse any existing workspace mount if it already is a mount */
2652 bool must_mount
, /* Whether to require that we mount something, it's not OK to use the plain directory fall back */
2655 int r
, workspace_mounted
; /* negative if we don't know yet whether we have/can mount something; true
2656 * if we mounted something; false if we definitely can't mount anything */
2664 if (reuse_workspace
) {
2665 r
= path_is_mount_point(workspace
, NULL
, 0);
2669 workspace_mounted
= true; /* If this is already a mount, and we are supposed to reuse it, let's keep this in mind */
2671 workspace_mounted
= -1; /* We need to figure out if we can mount something to the workspace */
2673 workspace_mounted
= -1; /* ditto */
2675 r
= path_is_mount_point(final
, NULL
, 0);
2679 /* If the final place already has something mounted, we use that. If the workspace also has
2680 * something mounted we assume it's actually the same mount (but with MS_RDONLY
2682 final_mounted
= true;
2684 if (workspace_mounted
< 0) {
2685 /* If the final place is mounted, but the workspace we isn't, then let's bind mount
2686 * the final version to the workspace, and make it writable, so that we can make
2689 r
= mount_nofollow_verbose(LOG_DEBUG
, final
, workspace
, NULL
, MS_BIND
|MS_REC
, NULL
);
2693 r
= mount_nofollow_verbose(LOG_DEBUG
, NULL
, workspace
, NULL
, MS_BIND
|MS_REMOUNT
|MS_NODEV
|MS_NOEXEC
|MS_NOSUID
, NULL
);
2697 workspace_mounted
= true;
2700 final_mounted
= false;
2702 if (workspace_mounted
< 0) {
2703 /* Nothing is mounted on the workspace yet, let's try to mount something now */
2704 for (int try = 0;; try++) {
2707 /* Try "ramfs" first, since it's not swap backed */
2708 r
= mount_nofollow_verbose(LOG_DEBUG
, "ramfs", workspace
, "ramfs", MS_NODEV
|MS_NOEXEC
|MS_NOSUID
, "mode=0700");
2710 workspace_mounted
= true;
2714 } else if (try == 1) {
2715 _cleanup_free_
char *opts
= NULL
;
2717 if (asprintf(&opts
, "mode=0700,nr_inodes=1024,size=%lu", CREDENTIALS_BYTES_MAX
) < 0)
2720 /* Fall back to "tmpfs" otherwise */
2721 r
= mount_nofollow_verbose(LOG_DEBUG
, "tmpfs", workspace
, "tmpfs", MS_NODEV
|MS_NOEXEC
|MS_NOSUID
, opts
);
2723 workspace_mounted
= true;
2728 /* If that didn't work, try to make a bind mount from the final to the workspace, so that we can make it writable there. */
2729 r
= mount_nofollow_verbose(LOG_DEBUG
, final
, workspace
, NULL
, MS_BIND
|MS_REC
, NULL
);
2731 if (!ERRNO_IS_PRIVILEGE(r
)) /* Propagate anything that isn't a permission problem */
2734 if (must_mount
) /* If we it's not OK to use the plain directory
2735 * fallback, propagate all errors too */
2738 /* If we lack privileges to bind mount stuff, then let's gracefully
2739 * proceed for compat with container envs, and just use the final dir
2742 workspace_mounted
= false;
2746 /* Make the new bind mount writable (i.e. drop MS_RDONLY) */
2747 r
= mount_nofollow_verbose(LOG_DEBUG
, NULL
, workspace
, NULL
, MS_BIND
|MS_REMOUNT
|MS_NODEV
|MS_NOEXEC
|MS_NOSUID
, NULL
);
2751 workspace_mounted
= true;
2757 assert(!must_mount
|| workspace_mounted
> 0);
2758 where
= workspace_mounted
? workspace
: final
;
2760 r
= acquire_credentials(context
, params
, unit
, where
, uid
, workspace_mounted
);
2764 if (workspace_mounted
) {
2765 /* Make workspace read-only now, so that any bind mount we make from it defaults to read-only too */
2766 r
= mount_nofollow_verbose(LOG_DEBUG
, NULL
, workspace
, NULL
, MS_BIND
|MS_REMOUNT
|MS_RDONLY
|MS_NODEV
|MS_NOEXEC
|MS_NOSUID
, NULL
);
2770 /* And mount it to the final place, read-only */
2772 r
= umount_verbose(LOG_DEBUG
, workspace
, MNT_DETACH
|UMOUNT_NOFOLLOW
);
2774 r
= mount_nofollow_verbose(LOG_DEBUG
, workspace
, final
, NULL
, MS_MOVE
, NULL
);
2778 _cleanup_free_
char *parent
= NULL
;
2780 /* If we do not have our own mount put used the plain directory fallback, then we need to
2781 * open access to the top-level credential directory and the per-service directory now */
2783 parent
= dirname_malloc(final
);
2786 if (chmod(parent
, 0755) < 0)
2793 static int setup_credentials(
2794 const ExecContext
*context
,
2795 const ExecParameters
*params
,
2799 _cleanup_free_
char *p
= NULL
, *q
= NULL
;
2806 if (!exec_context_has_credentials(context
))
2809 if (!params
->prefix
[EXEC_DIRECTORY_RUNTIME
])
2812 /* This where we'll place stuff when we are done; this main credentials directory is world-readable,
2813 * and the subdir we mount over with a read-only file system readable by the service's user */
2814 q
= path_join(params
->prefix
[EXEC_DIRECTORY_RUNTIME
], "credentials");
2818 r
= mkdir_label(q
, 0755); /* top-level dir: world readable/searchable */
2819 if (r
< 0 && r
!= -EEXIST
)
2822 p
= path_join(q
, unit
);
2826 r
= mkdir_label(p
, 0700); /* per-unit dir: private to user */
2827 if (r
< 0 && r
!= -EEXIST
)
2830 r
= safe_fork("(sd-mkdcreds)", FORK_DEATHSIG
|FORK_WAIT
|FORK_NEW_MOUNTNS
, NULL
);
2832 _cleanup_free_
char *t
= NULL
, *u
= NULL
;
2834 /* If this is not a privilege or support issue then propagate the error */
2835 if (!ERRNO_IS_NOT_SUPPORTED(r
) && !ERRNO_IS_PRIVILEGE(r
))
2838 /* Temporary workspace, that remains inaccessible all the time. We prepare stuff there before moving
2839 * it into place, so that users can't access half-initialized credential stores. */
2840 t
= path_join(params
->prefix
[EXEC_DIRECTORY_RUNTIME
], "systemd/temporary-credentials");
2844 /* We can't set up a mount namespace. In that case operate on a fixed, inaccessible per-unit
2845 * directory outside of /run/credentials/ first, and then move it over to /run/credentials/
2846 * after it is fully set up */
2847 u
= path_join(t
, unit
);
2851 FOREACH_STRING(i
, t
, u
) {
2852 r
= mkdir_label(i
, 0700);
2853 if (r
< 0 && r
!= -EEXIST
)
2857 r
= setup_credentials_internal(
2861 p
, /* final mount point */
2862 u
, /* temporary workspace to overmount */
2863 true, /* reuse the workspace if it is already a mount */
2864 false, /* it's OK to fall back to a plain directory if we can't mount anything */
2867 (void) rmdir(u
); /* remove the workspace again if we can. */
2872 } else if (r
== 0) {
2874 /* We managed to set up a mount namespace, and are now in a child. That's great. In this case
2875 * we can use the same directory for all cases, after turning off propagation. Question
2876 * though is: where do we turn off propagation exactly, and where do we place the workspace
2877 * directory? We need some place that is guaranteed to be a mount point in the host, and
2878 * which is guaranteed to have a subdir we can mount over. /run/ is not suitable for this,
2879 * since we ultimately want to move the resulting file system there, i.e. we need propagation
2880 * for /run/ eventually. We could use our own /run/systemd/bind mount on itself, but that
2881 * would be visible in the host mount table all the time, which we want to avoid. Hence, what
2882 * we do here instead we use /dev/ and /dev/shm/ for our purposes. We know for sure that
2883 * /dev/ is a mount point and we now for sure that /dev/shm/ exists. Hence we can turn off
2884 * propagation on the former, and then overmount the latter.
2886 * Yes it's nasty playing games with /dev/ and /dev/shm/ like this, since it does not exist
2887 * for this purpose, but there are few other candidates that work equally well for us, and
2888 * given that the we do this in a privately namespaced short-lived single-threaded process
2889 * that no one else sees this should be OK to do. */
2891 r
= mount_nofollow_verbose(LOG_DEBUG
, NULL
, "/dev", NULL
, MS_SLAVE
|MS_REC
, NULL
); /* Turn off propagation from our namespace to host */
2895 r
= setup_credentials_internal(
2899 p
, /* final mount point */
2900 "/dev/shm", /* temporary workspace to overmount */
2901 false, /* do not reuse /dev/shm if it is already a mount, under no circumstances */
2902 true, /* insist that something is mounted, do not allow fallback to plain directory */
2907 _exit(EXIT_SUCCESS
);
2910 _exit(EXIT_FAILURE
);
2917 static int setup_smack(
2918 const ExecContext
*context
,
2919 int executable_fd
) {
2923 assert(executable_fd
>= 0);
2925 if (context
->smack_process_label
) {
2926 r
= mac_smack_apply_pid(0, context
->smack_process_label
);
2930 #ifdef SMACK_DEFAULT_PROCESS_LABEL
2932 _cleanup_free_
char *exec_label
= NULL
;
2934 r
= mac_smack_read_fd(executable_fd
, SMACK_ATTR_EXEC
, &exec_label
);
2935 if (r
< 0 && !IN_SET(r
, -ENODATA
, -EOPNOTSUPP
))
2938 r
= mac_smack_apply_pid(0, exec_label
? : SMACK_DEFAULT_PROCESS_LABEL
);
2948 static int compile_bind_mounts(
2949 const ExecContext
*context
,
2950 const ExecParameters
*params
,
2951 BindMount
**ret_bind_mounts
,
2952 size_t *ret_n_bind_mounts
,
2953 char ***ret_empty_directories
) {
2955 _cleanup_strv_free_
char **empty_directories
= NULL
;
2956 BindMount
*bind_mounts
;
2962 assert(ret_bind_mounts
);
2963 assert(ret_n_bind_mounts
);
2964 assert(ret_empty_directories
);
2966 n
= context
->n_bind_mounts
;
2967 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
2968 if (!params
->prefix
[t
])
2971 n
+= strv_length(context
->directories
[t
].paths
);
2975 *ret_bind_mounts
= NULL
;
2976 *ret_n_bind_mounts
= 0;
2977 *ret_empty_directories
= NULL
;
2981 bind_mounts
= new(BindMount
, n
);
2985 for (size_t i
= 0; i
< context
->n_bind_mounts
; i
++) {
2986 BindMount
*item
= context
->bind_mounts
+ i
;
2989 s
= strdup(item
->source
);
2995 d
= strdup(item
->destination
);
3002 bind_mounts
[h
++] = (BindMount
) {
3005 .read_only
= item
->read_only
,
3006 .recursive
= item
->recursive
,
3007 .ignore_enoent
= item
->ignore_enoent
,
3011 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
3014 if (!params
->prefix
[t
])
3017 if (strv_isempty(context
->directories
[t
].paths
))
3020 if (exec_directory_is_private(context
, t
) &&
3021 !exec_context_with_rootfs(context
)) {
3024 /* So this is for a dynamic user, and we need to make sure the process can access its own
3025 * directory. For that we overmount the usually inaccessible "private" subdirectory with a
3026 * tmpfs that makes it accessible and is empty except for the submounts we do this for. */
3028 private_root
= path_join(params
->prefix
[t
], "private");
3029 if (!private_root
) {
3034 r
= strv_consume(&empty_directories
, private_root
);
3039 STRV_FOREACH(suffix
, context
->directories
[t
].paths
) {
3042 if (exec_directory_is_private(context
, t
))
3043 s
= path_join(params
->prefix
[t
], "private", *suffix
);
3045 s
= path_join(params
->prefix
[t
], *suffix
);
3051 if (exec_directory_is_private(context
, t
) &&
3052 exec_context_with_rootfs(context
))
3053 /* When RootDirectory= or RootImage= are set, then the symbolic link to the private
3054 * directory is not created on the root directory. So, let's bind-mount the directory
3055 * on the 'non-private' place. */
3056 d
= path_join(params
->prefix
[t
], *suffix
);
3065 bind_mounts
[h
++] = (BindMount
) {
3069 .nosuid
= context
->dynamic_user
, /* don't allow suid/sgid when DynamicUser= is on */
3071 .ignore_enoent
= false,
3078 *ret_bind_mounts
= bind_mounts
;
3079 *ret_n_bind_mounts
= n
;
3080 *ret_empty_directories
= TAKE_PTR(empty_directories
);
3085 bind_mount_free_many(bind_mounts
, h
);
3089 static bool insist_on_sandboxing(
3090 const ExecContext
*context
,
3091 const char *root_dir
,
3092 const char *root_image
,
3093 const BindMount
*bind_mounts
,
3094 size_t n_bind_mounts
) {
3097 assert(n_bind_mounts
== 0 || bind_mounts
);
3099 /* Checks whether we need to insist on fs namespacing. i.e. whether we have settings configured that
3100 * would alter the view on the file system beyond making things read-only or invisible, i.e. would
3101 * rearrange stuff in a way we cannot ignore gracefully. */
3103 if (context
->n_temporary_filesystems
> 0)
3106 if (root_dir
|| root_image
)
3109 if (context
->n_mount_images
> 0)
3112 if (context
->dynamic_user
)
3115 /* If there are any bind mounts set that don't map back onto themselves, fs namespacing becomes
3117 for (size_t i
= 0; i
< n_bind_mounts
; i
++)
3118 if (!path_equal(bind_mounts
[i
].source
, bind_mounts
[i
].destination
))
3121 if (context
->log_namespace
)
3127 static int apply_mount_namespace(
3129 ExecCommandFlags command_flags
,
3130 const ExecContext
*context
,
3131 const ExecParameters
*params
,
3132 const ExecRuntime
*runtime
,
3133 char **error_path
) {
3135 _cleanup_strv_free_
char **empty_directories
= NULL
;
3136 const char *tmp_dir
= NULL
, *var_tmp_dir
= NULL
;
3137 const char *root_dir
= NULL
, *root_image
= NULL
;
3138 _cleanup_free_
char *creds_path
= NULL
, *incoming_dir
= NULL
, *propagate_dir
= NULL
;
3139 NamespaceInfo ns_info
;
3140 bool needs_sandboxing
;
3141 BindMount
*bind_mounts
= NULL
;
3142 size_t n_bind_mounts
= 0;
3147 if (params
->flags
& EXEC_APPLY_CHROOT
) {
3148 root_image
= context
->root_image
;
3151 root_dir
= context
->root_directory
;
3154 r
= compile_bind_mounts(context
, params
, &bind_mounts
, &n_bind_mounts
, &empty_directories
);
3158 needs_sandboxing
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && !(command_flags
& EXEC_COMMAND_FULLY_PRIVILEGED
);
3159 if (needs_sandboxing
) {
3160 /* The runtime struct only contains the parent of the private /tmp,
3161 * which is non-accessible to world users. Inside of it there's a /tmp
3162 * that is sticky, and that's the one we want to use here.
3163 * This does not apply when we are using /run/systemd/empty as fallback. */
3165 if (context
->private_tmp
&& runtime
) {
3166 if (streq_ptr(runtime
->tmp_dir
, RUN_SYSTEMD_EMPTY
))
3167 tmp_dir
= runtime
->tmp_dir
;
3168 else if (runtime
->tmp_dir
)
3169 tmp_dir
= strjoina(runtime
->tmp_dir
, "/tmp");
3171 if (streq_ptr(runtime
->var_tmp_dir
, RUN_SYSTEMD_EMPTY
))
3172 var_tmp_dir
= runtime
->var_tmp_dir
;
3173 else if (runtime
->var_tmp_dir
)
3174 var_tmp_dir
= strjoina(runtime
->var_tmp_dir
, "/tmp");
3177 ns_info
= (NamespaceInfo
) {
3178 .ignore_protect_paths
= false,
3179 .private_dev
= context
->private_devices
,
3180 .protect_control_groups
= context
->protect_control_groups
,
3181 .protect_kernel_tunables
= context
->protect_kernel_tunables
,
3182 .protect_kernel_modules
= context
->protect_kernel_modules
,
3183 .protect_kernel_logs
= context
->protect_kernel_logs
,
3184 .protect_hostname
= context
->protect_hostname
,
3185 .mount_apivfs
= exec_context_get_effective_mount_apivfs(context
),
3186 .private_mounts
= context
->private_mounts
,
3187 .protect_home
= context
->protect_home
,
3188 .protect_system
= context
->protect_system
,
3189 .protect_proc
= context
->protect_proc
,
3190 .proc_subset
= context
->proc_subset
,
3191 .private_ipc
= context
->private_ipc
|| context
->ipc_namespace_path
,
3193 } else if (!context
->dynamic_user
&& root_dir
)
3195 * If DynamicUser=no and RootDirectory= is set then lets pass a relaxed
3196 * sandbox info, otherwise enforce it, don't ignore protected paths and
3197 * fail if we are enable to apply the sandbox inside the mount namespace.
3199 ns_info
= (NamespaceInfo
) {
3200 .ignore_protect_paths
= true,
3203 ns_info
= (NamespaceInfo
) {};
3205 if (context
->mount_flags
== MS_SHARED
)
3206 log_unit_debug(u
, "shared mount propagation hidden by other fs namespacing unit settings: ignoring");
3208 if (exec_context_has_credentials(context
) &&
3209 params
->prefix
[EXEC_DIRECTORY_RUNTIME
] &&
3210 FLAGS_SET(params
->flags
, EXEC_WRITE_CREDENTIALS
)) {
3211 creds_path
= path_join(params
->prefix
[EXEC_DIRECTORY_RUNTIME
], "credentials", u
->id
);
3218 if (MANAGER_IS_SYSTEM(u
->manager
)) {
3219 propagate_dir
= path_join("/run/systemd/propagate/", u
->id
);
3220 if (!propagate_dir
) {
3225 incoming_dir
= strdup("/run/systemd/incoming");
3226 if (!incoming_dir
) {
3232 r
= setup_namespace(root_dir
, root_image
, context
->root_image_options
,
3233 &ns_info
, context
->read_write_paths
,
3234 needs_sandboxing
? context
->read_only_paths
: NULL
,
3235 needs_sandboxing
? context
->inaccessible_paths
: NULL
,
3236 needs_sandboxing
? context
->exec_paths
: NULL
,
3237 needs_sandboxing
? context
->no_exec_paths
: NULL
,
3241 context
->temporary_filesystems
,
3242 context
->n_temporary_filesystems
,
3243 context
->mount_images
,
3244 context
->n_mount_images
,
3248 context
->log_namespace
,
3249 context
->mount_flags
,
3250 context
->root_hash
, context
->root_hash_size
, context
->root_hash_path
,
3251 context
->root_hash_sig
, context
->root_hash_sig_size
, context
->root_hash_sig_path
,
3252 context
->root_verity
,
3253 context
->extension_images
,
3254 context
->n_extension_images
,
3257 root_dir
|| root_image
? params
->notify_socket
: NULL
,
3260 /* If we couldn't set up the namespace this is probably due to a missing capability. setup_namespace() reports
3261 * that with a special, recognizable error ENOANO. In this case, silently proceed, but only if exclusively
3262 * sandboxing options were used, i.e. nothing such as RootDirectory= or BindMount= that would result in a
3263 * completely different execution environment. */
3265 if (insist_on_sandboxing(
3267 root_dir
, root_image
,
3270 log_unit_debug(u
, "Failed to set up namespace, and refusing to continue since the selected namespacing options alter mount environment non-trivially.\n"
3271 "Bind mounts: %zu, temporary filesystems: %zu, root directory: %s, root image: %s, dynamic user: %s",
3272 n_bind_mounts
, context
->n_temporary_filesystems
, yes_no(root_dir
), yes_no(root_image
), yes_no(context
->dynamic_user
));
3276 log_unit_debug(u
, "Failed to set up namespace, assuming containerized execution and ignoring.");
3282 bind_mount_free_many(bind_mounts
, n_bind_mounts
);
3286 static int apply_working_directory(
3287 const ExecContext
*context
,
3288 const ExecParameters
*params
,
3295 assert(exit_status
);
3297 if (context
->working_directory_home
) {
3300 *exit_status
= EXIT_CHDIR
;
3307 wd
= empty_to_root(context
->working_directory
);
3309 if (params
->flags
& EXEC_APPLY_CHROOT
)
3312 d
= prefix_roota(context
->root_directory
, wd
);
3314 if (chdir(d
) < 0 && !context
->working_directory_missing_ok
) {
3315 *exit_status
= EXIT_CHDIR
;
3322 static int apply_root_directory(
3323 const ExecContext
*context
,
3324 const ExecParameters
*params
,
3325 const bool needs_mount_ns
,
3329 assert(exit_status
);
3331 if (params
->flags
& EXEC_APPLY_CHROOT
)
3332 if (!needs_mount_ns
&& context
->root_directory
)
3333 if (chroot(context
->root_directory
) < 0) {
3334 *exit_status
= EXIT_CHROOT
;
3341 static int setup_keyring(
3343 const ExecContext
*context
,
3344 const ExecParameters
*p
,
3345 uid_t uid
, gid_t gid
) {
3347 key_serial_t keyring
;
3356 /* Let's set up a new per-service "session" kernel keyring for each system service. This has the benefit that
3357 * each service runs with its own keyring shared among all processes of the service, but with no hook-up beyond
3358 * that scope, and in particular no link to the per-UID keyring. If we don't do this the keyring will be
3359 * automatically created on-demand and then linked to the per-UID keyring, by the kernel. The kernel's built-in
3360 * on-demand behaviour is very appropriate for login users, but probably not so much for system services, where
3361 * UIDs are not necessarily specific to a service but reused (at least in the case of UID 0). */
3363 if (context
->keyring_mode
== EXEC_KEYRING_INHERIT
)
3366 /* Acquiring a reference to the user keyring is nasty. We briefly change identity in order to get things set up
3367 * properly by the kernel. If we don't do that then we can't create it atomically, and that sucks for parallel
3368 * execution. This mimics what pam_keyinit does, too. Setting up session keyring, to be owned by the right user
3369 * & group is just as nasty as acquiring a reference to the user keyring. */
3371 saved_uid
= getuid();
3372 saved_gid
= getgid();
3374 if (gid_is_valid(gid
) && gid
!= saved_gid
) {
3375 if (setregid(gid
, -1) < 0)
3376 return log_unit_error_errno(u
, errno
, "Failed to change GID for user keyring: %m");
3379 if (uid_is_valid(uid
) && uid
!= saved_uid
) {
3380 if (setreuid(uid
, -1) < 0) {
3381 r
= log_unit_error_errno(u
, errno
, "Failed to change UID for user keyring: %m");
3386 keyring
= keyctl(KEYCTL_JOIN_SESSION_KEYRING
, 0, 0, 0, 0);
3387 if (keyring
== -1) {
3388 if (errno
== ENOSYS
)
3389 log_unit_debug_errno(u
, errno
, "Kernel keyring not supported, ignoring.");
3390 else if (ERRNO_IS_PRIVILEGE(errno
))
3391 log_unit_debug_errno(u
, errno
, "Kernel keyring access prohibited, ignoring.");
3392 else if (errno
== EDQUOT
)
3393 log_unit_debug_errno(u
, errno
, "Out of kernel keyrings to allocate, ignoring.");
3395 r
= log_unit_error_errno(u
, errno
, "Setting up kernel keyring failed: %m");
3400 /* When requested link the user keyring into the session keyring. */
3401 if (context
->keyring_mode
== EXEC_KEYRING_SHARED
) {
3403 if (keyctl(KEYCTL_LINK
,
3404 KEY_SPEC_USER_KEYRING
,
3405 KEY_SPEC_SESSION_KEYRING
, 0, 0) < 0) {
3406 r
= log_unit_error_errno(u
, errno
, "Failed to link user keyring into session keyring: %m");
3411 /* Restore uid/gid back */
3412 if (uid_is_valid(uid
) && uid
!= saved_uid
) {
3413 if (setreuid(saved_uid
, -1) < 0) {
3414 r
= log_unit_error_errno(u
, errno
, "Failed to change UID back for user keyring: %m");
3419 if (gid_is_valid(gid
) && gid
!= saved_gid
) {
3420 if (setregid(saved_gid
, -1) < 0)
3421 return log_unit_error_errno(u
, errno
, "Failed to change GID back for user keyring: %m");
3424 /* Populate they keyring with the invocation ID by default, as original saved_uid. */
3425 if (!sd_id128_is_null(u
->invocation_id
)) {
3428 key
= add_key("user", "invocation_id", &u
->invocation_id
, sizeof(u
->invocation_id
), KEY_SPEC_SESSION_KEYRING
);
3430 log_unit_debug_errno(u
, errno
, "Failed to add invocation ID to keyring, ignoring: %m");
3432 if (keyctl(KEYCTL_SETPERM
, key
,
3433 KEY_POS_VIEW
|KEY_POS_READ
|KEY_POS_SEARCH
|
3434 KEY_USR_VIEW
|KEY_USR_READ
|KEY_USR_SEARCH
, 0, 0) < 0)
3435 r
= log_unit_error_errno(u
, errno
, "Failed to restrict invocation ID permission: %m");
3440 /* Revert back uid & gid for the last time, and exit */
3441 /* no extra logging, as only the first already reported error matters */
3442 if (getuid() != saved_uid
)
3443 (void) setreuid(saved_uid
, -1);
3445 if (getgid() != saved_gid
)
3446 (void) setregid(saved_gid
, -1);
3451 static void append_socket_pair(int *array
, size_t *n
, const int pair
[static 2]) {
3457 array
[(*n
)++] = pair
[0];
3459 array
[(*n
)++] = pair
[1];
3462 static int close_remaining_fds(
3463 const ExecParameters
*params
,
3464 const ExecRuntime
*runtime
,
3465 const DynamicCreds
*dcreds
,
3468 const int *fds
, size_t n_fds
) {
3470 size_t n_dont_close
= 0;
3471 int dont_close
[n_fds
+ 12];
3475 if (params
->stdin_fd
>= 0)
3476 dont_close
[n_dont_close
++] = params
->stdin_fd
;
3477 if (params
->stdout_fd
>= 0)
3478 dont_close
[n_dont_close
++] = params
->stdout_fd
;
3479 if (params
->stderr_fd
>= 0)
3480 dont_close
[n_dont_close
++] = params
->stderr_fd
;
3483 dont_close
[n_dont_close
++] = socket_fd
;
3485 memcpy(dont_close
+ n_dont_close
, fds
, sizeof(int) * n_fds
);
3486 n_dont_close
+= n_fds
;
3490 append_socket_pair(dont_close
, &n_dont_close
, runtime
->netns_storage_socket
);
3491 append_socket_pair(dont_close
, &n_dont_close
, runtime
->ipcns_storage_socket
);
3496 append_socket_pair(dont_close
, &n_dont_close
, dcreds
->user
->storage_socket
);
3498 append_socket_pair(dont_close
, &n_dont_close
, dcreds
->group
->storage_socket
);
3501 if (user_lookup_fd
>= 0)
3502 dont_close
[n_dont_close
++] = user_lookup_fd
;
3504 return close_all_fds(dont_close
, n_dont_close
);
3507 static int send_user_lookup(
3515 /* Send the resolved UID/GID to PID 1 after we learnt it. We send a single datagram, containing the UID/GID
3516 * data as well as the unit name. Note that we suppress sending this if no user/group to resolve was
3519 if (user_lookup_fd
< 0)
3522 if (!uid_is_valid(uid
) && !gid_is_valid(gid
))
3525 if (writev(user_lookup_fd
,
3527 IOVEC_INIT(&uid
, sizeof(uid
)),
3528 IOVEC_INIT(&gid
, sizeof(gid
)),
3529 IOVEC_INIT_STRING(unit
->id
) }, 3) < 0)
3535 static int acquire_home(const ExecContext
*c
, uid_t uid
, const char** home
, char **buf
) {
3542 /* If WorkingDirectory=~ is set, try to acquire a usable home directory. */
3547 if (!c
->working_directory_home
)
3550 r
= get_home_dir(buf
);
3558 static int compile_suggested_paths(const ExecContext
*c
, const ExecParameters
*p
, char ***ret
) {
3559 _cleanup_strv_free_
char ** list
= NULL
;
3566 assert(c
->dynamic_user
);
3568 /* Compile a list of paths that it might make sense to read the owning UID from to use as initial candidate for
3569 * dynamic UID allocation, in order to save us from doing costly recursive chown()s of the special
3572 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
3575 if (t
== EXEC_DIRECTORY_CONFIGURATION
)
3581 STRV_FOREACH(i
, c
->directories
[t
].paths
) {
3584 if (exec_directory_is_private(c
, t
))
3585 e
= path_join(p
->prefix
[t
], "private", *i
);
3587 e
= path_join(p
->prefix
[t
], *i
);
3591 r
= strv_consume(&list
, e
);
3597 *ret
= TAKE_PTR(list
);
3602 static char *exec_command_line(char **argv
);
3604 static int exec_parameters_get_cgroup_path(const ExecParameters
*params
, char **ret
) {
3605 bool using_subcgroup
;
3611 if (!params
->cgroup_path
)
3614 /* If we are called for a unit where cgroup delegation is on, and the payload created its own populated
3615 * subcgroup (which we expect it to do, after all it asked for delegation), then we cannot place the control
3616 * processes started after the main unit's process in the unit's main cgroup because it is now an inner one,
3617 * and inner cgroups may not contain processes. Hence, if delegation is on, and this is a control process,
3618 * let's use ".control" as subcgroup instead. Note that we do so only for ExecStartPost=, ExecReload=,
3619 * ExecStop=, ExecStopPost=, i.e. for the commands where the main process is already forked. For ExecStartPre=
3620 * this is not necessary, the cgroup is still empty. We distinguish these cases with the EXEC_CONTROL_CGROUP
3621 * flag, which is only passed for the former statements, not for the latter. */
3623 using_subcgroup
= FLAGS_SET(params
->flags
, EXEC_CONTROL_CGROUP
|EXEC_CGROUP_DELEGATE
|EXEC_IS_CONTROL
);
3624 if (using_subcgroup
)
3625 p
= path_join(params
->cgroup_path
, ".control");
3627 p
= strdup(params
->cgroup_path
);
3632 return using_subcgroup
;
3635 static int exec_context_cpu_affinity_from_numa(const ExecContext
*c
, CPUSet
*ret
) {
3636 _cleanup_(cpu_set_reset
) CPUSet s
= {};
3642 if (!c
->numa_policy
.nodes
.set
) {
3643 log_debug("Can't derive CPU affinity mask from NUMA mask because NUMA mask is not set, ignoring");
3647 r
= numa_to_cpu_set(&c
->numa_policy
, &s
);
3653 return cpu_set_add_all(ret
, &s
);
3656 bool exec_context_get_cpu_affinity_from_numa(const ExecContext
*c
) {
3659 return c
->cpu_affinity_from_numa
;
3662 static int add_shifted_fd(int *fds
, size_t fds_size
, size_t *n_fds
, int fd
, int *ret_fd
) {
3667 assert(*n_fds
< fds_size
);
3675 if (fd
< 3 + (int) *n_fds
) {
3676 /* Let's move the fd up, so that it's outside of the fd range we will use to store
3677 * the fds we pass to the process (or which are closed only during execve). */
3679 r
= fcntl(fd
, F_DUPFD_CLOEXEC
, 3 + (int) *n_fds
);
3683 CLOSE_AND_REPLACE(fd
, r
);
3686 *ret_fd
= fds
[*n_fds
] = fd
;
3691 static int exec_child(
3693 const ExecCommand
*command
,
3694 const ExecContext
*context
,
3695 const ExecParameters
*params
,
3696 ExecRuntime
*runtime
,
3697 DynamicCreds
*dcreds
,
3699 const int named_iofds
[static 3],
3701 size_t n_socket_fds
,
3702 size_t n_storage_fds
,
3707 _cleanup_strv_free_
char **our_env
= NULL
, **pass_env
= NULL
, **accum_env
= NULL
, **replaced_argv
= NULL
;
3708 int r
, ngids
= 0, exec_fd
;
3709 _cleanup_free_ gid_t
*supplementary_gids
= NULL
;
3710 const char *username
= NULL
, *groupname
= NULL
;
3711 _cleanup_free_
char *home_buffer
= NULL
;
3712 const char *home
= NULL
, *shell
= NULL
;
3713 char **final_argv
= NULL
;
3714 dev_t journal_stream_dev
= 0;
3715 ino_t journal_stream_ino
= 0;
3716 bool userns_set_up
= false;
3717 bool needs_sandboxing
, /* Do we need to set up full sandboxing? (i.e. all namespacing, all MAC stuff, caps, yadda yadda */
3718 needs_setuid
, /* Do we need to do the actual setresuid()/setresgid() calls? */
3719 needs_mount_namespace
, /* Do we need to set up a mount namespace for this kernel? */
3720 needs_ambient_hack
; /* Do we need to apply the ambient capabilities hack? */
3722 _cleanup_free_
char *mac_selinux_context_net
= NULL
;
3723 bool use_selinux
= false;
3726 bool use_smack
= false;
3729 bool use_apparmor
= false;
3731 uid_t saved_uid
= getuid();
3732 gid_t saved_gid
= getgid();
3733 uid_t uid
= UID_INVALID
;
3734 gid_t gid
= GID_INVALID
;
3735 size_t n_fds
= n_socket_fds
+ n_storage_fds
, /* fds to pass to the child */
3736 n_keep_fds
; /* total number of fds not to close */
3738 _cleanup_free_ gid_t
*gids_after_pam
= NULL
;
3739 int ngids_after_pam
= 0;
3745 assert(exit_status
);
3747 rename_process_from_path(command
->path
);
3749 /* We reset exactly these signals, since they are the only ones we set to SIG_IGN in the main
3750 * daemon. All others we leave untouched because we set them to SIG_DFL or a valid handler initially,
3751 * both of which will be demoted to SIG_DFL. */
3752 (void) default_signals(SIGNALS_CRASH_HANDLER
,
3755 if (context
->ignore_sigpipe
)
3756 (void) ignore_signals(SIGPIPE
);
3758 r
= reset_signal_mask();
3760 *exit_status
= EXIT_SIGNAL_MASK
;
3761 return log_unit_error_errno(unit
, r
, "Failed to set process signal mask: %m");
3764 if (params
->idle_pipe
)
3765 do_idle_pipe_dance(params
->idle_pipe
);
3767 /* Close fds we don't need very early to make sure we don't block init reexecution because it cannot bind its
3768 * sockets. Among the fds we close are the logging fds, and we want to keep them closed, so that we don't have
3769 * any fds open we don't really want open during the transition. In order to make logging work, we switch the
3770 * log subsystem into open_when_needed mode, so that it reopens the logs on every single log call. */
3773 log_set_open_when_needed(true);
3775 /* In case anything used libc syslog(), close this here, too */
3778 int keep_fds
[n_fds
+ 2];
3779 memcpy_safe(keep_fds
, fds
, n_fds
* sizeof(int));
3782 r
= add_shifted_fd(keep_fds
, ELEMENTSOF(keep_fds
), &n_keep_fds
, params
->exec_fd
, &exec_fd
);
3784 *exit_status
= EXIT_FDS
;
3785 return log_unit_error_errno(unit
, r
, "Failed to shift fd and set FD_CLOEXEC: %m");
3788 r
= close_remaining_fds(params
, runtime
, dcreds
, user_lookup_fd
, socket_fd
, keep_fds
, n_keep_fds
);
3790 *exit_status
= EXIT_FDS
;
3791 return log_unit_error_errno(unit
, r
, "Failed to close unwanted file descriptors: %m");
3794 if (!context
->same_pgrp
&&
3796 *exit_status
= EXIT_SETSID
;
3797 return log_unit_error_errno(unit
, errno
, "Failed to create new process session: %m");
3800 exec_context_tty_reset(context
, params
);
3802 if (unit_shall_confirm_spawn(unit
)) {
3803 const char *vc
= params
->confirm_spawn
;
3804 _cleanup_free_
char *cmdline
= NULL
;
3806 cmdline
= exec_command_line(command
->argv
);
3808 *exit_status
= EXIT_MEMORY
;
3812 r
= ask_for_confirmation(vc
, unit
, cmdline
);
3813 if (r
!= CONFIRM_EXECUTE
) {
3814 if (r
== CONFIRM_PRETEND_SUCCESS
) {
3815 *exit_status
= EXIT_SUCCESS
;
3818 *exit_status
= EXIT_CONFIRM
;
3819 return log_unit_error_errno(unit
, SYNTHETIC_ERRNO(ECANCELED
),
3820 "Execution cancelled by the user");
3824 /* We are about to invoke NSS and PAM modules. Let's tell them what we are doing here, maybe they care. This is
3825 * used by nss-resolve to disable itself when we are about to start systemd-resolved, to avoid deadlocks. Note
3826 * that these env vars do not survive the execve(), which means they really only apply to the PAM and NSS
3827 * invocations themselves. Also note that while we'll only invoke NSS modules involved in user management they
3828 * might internally call into other NSS modules that are involved in hostname resolution, we never know. */
3829 if (setenv("SYSTEMD_ACTIVATION_UNIT", unit
->id
, true) != 0 ||
3830 setenv("SYSTEMD_ACTIVATION_SCOPE", MANAGER_IS_SYSTEM(unit
->manager
) ? "system" : "user", true) != 0) {
3831 *exit_status
= EXIT_MEMORY
;
3832 return log_unit_error_errno(unit
, errno
, "Failed to update environment: %m");
3835 if (context
->dynamic_user
&& dcreds
) {
3836 _cleanup_strv_free_
char **suggested_paths
= NULL
;
3838 /* On top of that, make sure we bypass our own NSS module nss-systemd comprehensively for any NSS
3839 * checks, if DynamicUser=1 is used, as we shouldn't create a feedback loop with ourselves here. */
3840 if (putenv((char*) "SYSTEMD_NSS_DYNAMIC_BYPASS=1") != 0) {
3841 *exit_status
= EXIT_USER
;
3842 return log_unit_error_errno(unit
, errno
, "Failed to update environment: %m");
3845 r
= compile_suggested_paths(context
, params
, &suggested_paths
);
3847 *exit_status
= EXIT_MEMORY
;
3851 r
= dynamic_creds_realize(dcreds
, suggested_paths
, &uid
, &gid
);
3853 *exit_status
= EXIT_USER
;
3855 return log_unit_error_errno(unit
, SYNTHETIC_ERRNO(EOPNOTSUPP
),
3856 "Failed to update dynamic user credentials: User or group with specified name already exists.");
3857 return log_unit_error_errno(unit
, r
, "Failed to update dynamic user credentials: %m");
3860 if (!uid_is_valid(uid
)) {
3861 *exit_status
= EXIT_USER
;
3862 return log_unit_error_errno(unit
, SYNTHETIC_ERRNO(ESRCH
), "UID validation failed for \""UID_FMT
"\"", uid
);
3865 if (!gid_is_valid(gid
)) {
3866 *exit_status
= EXIT_USER
;
3867 return log_unit_error_errno(unit
, SYNTHETIC_ERRNO(ESRCH
), "GID validation failed for \""GID_FMT
"\"", gid
);
3871 username
= dcreds
->user
->name
;
3874 r
= get_fixed_user(context
, &username
, &uid
, &gid
, &home
, &shell
);
3876 *exit_status
= EXIT_USER
;
3877 return log_unit_error_errno(unit
, r
, "Failed to determine user credentials: %m");
3880 r
= get_fixed_group(context
, &groupname
, &gid
);
3882 *exit_status
= EXIT_GROUP
;
3883 return log_unit_error_errno(unit
, r
, "Failed to determine group credentials: %m");
3887 /* Initialize user supplementary groups and get SupplementaryGroups= ones */
3888 r
= get_supplementary_groups(context
, username
, groupname
, gid
,
3889 &supplementary_gids
, &ngids
);
3891 *exit_status
= EXIT_GROUP
;
3892 return log_unit_error_errno(unit
, r
, "Failed to determine supplementary groups: %m");
3895 r
= send_user_lookup(unit
, user_lookup_fd
, uid
, gid
);
3897 *exit_status
= EXIT_USER
;
3898 return log_unit_error_errno(unit
, r
, "Failed to send user credentials to PID1: %m");
3901 user_lookup_fd
= safe_close(user_lookup_fd
);
3903 r
= acquire_home(context
, uid
, &home
, &home_buffer
);
3905 *exit_status
= EXIT_CHDIR
;
3906 return log_unit_error_errno(unit
, r
, "Failed to determine $HOME for user: %m");
3909 /* If a socket is connected to STDIN/STDOUT/STDERR, we
3910 * must sure to drop O_NONBLOCK */
3912 (void) fd_nonblock(socket_fd
, false);
3914 /* Journald will try to look-up our cgroup in order to populate _SYSTEMD_CGROUP and _SYSTEMD_UNIT fields.
3915 * Hence we need to migrate to the target cgroup from init.scope before connecting to journald */
3916 if (params
->cgroup_path
) {
3917 _cleanup_free_
char *p
= NULL
;
3919 r
= exec_parameters_get_cgroup_path(params
, &p
);
3921 *exit_status
= EXIT_CGROUP
;
3922 return log_unit_error_errno(unit
, r
, "Failed to acquire cgroup path: %m");
3925 r
= cg_attach_everywhere(params
->cgroup_supported
, p
, 0, NULL
, NULL
);
3927 *exit_status
= EXIT_CGROUP
;
3928 return log_unit_error_errno(unit
, r
, "Failed to attach to cgroup %s: %m", p
);
3932 if (context
->network_namespace_path
&& runtime
&& runtime
->netns_storage_socket
[0] >= 0) {
3933 r
= open_shareable_ns_path(runtime
->netns_storage_socket
, context
->network_namespace_path
, CLONE_NEWNET
);
3935 *exit_status
= EXIT_NETWORK
;
3936 return log_unit_error_errno(unit
, r
, "Failed to open network namespace path %s: %m", context
->network_namespace_path
);
3940 if (context
->ipc_namespace_path
&& runtime
&& runtime
->ipcns_storage_socket
[0] >= 0) {
3941 r
= open_shareable_ns_path(runtime
->ipcns_storage_socket
, context
->ipc_namespace_path
, CLONE_NEWIPC
);
3943 *exit_status
= EXIT_NAMESPACE
;
3944 return log_unit_error_errno(unit
, r
, "Failed to open IPC namespace path %s: %m", context
->ipc_namespace_path
);
3948 r
= setup_input(context
, params
, socket_fd
, named_iofds
);
3950 *exit_status
= EXIT_STDIN
;
3951 return log_unit_error_errno(unit
, r
, "Failed to set up standard input: %m");
3954 r
= setup_output(unit
, context
, params
, STDOUT_FILENO
, socket_fd
, named_iofds
, basename(command
->path
), uid
, gid
, &journal_stream_dev
, &journal_stream_ino
);
3956 *exit_status
= EXIT_STDOUT
;
3957 return log_unit_error_errno(unit
, r
, "Failed to set up standard output: %m");
3960 r
= setup_output(unit
, context
, params
, STDERR_FILENO
, socket_fd
, named_iofds
, basename(command
->path
), uid
, gid
, &journal_stream_dev
, &journal_stream_ino
);
3962 *exit_status
= EXIT_STDERR
;
3963 return log_unit_error_errno(unit
, r
, "Failed to set up standard error output: %m");
3966 if (context
->oom_score_adjust_set
) {
3967 /* When we can't make this change due to EPERM, then let's silently skip over it. User namespaces
3968 * prohibit write access to this file, and we shouldn't trip up over that. */
3969 r
= set_oom_score_adjust(context
->oom_score_adjust
);
3970 if (ERRNO_IS_PRIVILEGE(r
))
3971 log_unit_debug_errno(unit
, r
, "Failed to adjust OOM setting, assuming containerized execution, ignoring: %m");
3973 *exit_status
= EXIT_OOM_ADJUST
;
3974 return log_unit_error_errno(unit
, r
, "Failed to adjust OOM setting: %m");
3978 if (context
->coredump_filter_set
) {
3979 r
= set_coredump_filter(context
->coredump_filter
);
3980 if (ERRNO_IS_PRIVILEGE(r
))
3981 log_unit_debug_errno(unit
, r
, "Failed to adjust coredump_filter, ignoring: %m");
3983 return log_unit_error_errno(unit
, r
, "Failed to adjust coredump_filter: %m");
3986 if (context
->nice_set
) {
3987 r
= setpriority_closest(context
->nice
);
3989 return log_unit_error_errno(unit
, r
, "Failed to set up process scheduling priority (nice level): %m");
3992 if (context
->cpu_sched_set
) {
3993 struct sched_param param
= {
3994 .sched_priority
= context
->cpu_sched_priority
,
3997 r
= sched_setscheduler(0,
3998 context
->cpu_sched_policy
|
3999 (context
->cpu_sched_reset_on_fork
?
4000 SCHED_RESET_ON_FORK
: 0),
4003 *exit_status
= EXIT_SETSCHEDULER
;
4004 return log_unit_error_errno(unit
, errno
, "Failed to set up CPU scheduling: %m");
4008 if (context
->cpu_affinity_from_numa
|| context
->cpu_set
.set
) {
4009 _cleanup_(cpu_set_reset
) CPUSet converted_cpu_set
= {};
4010 const CPUSet
*cpu_set
;
4012 if (context
->cpu_affinity_from_numa
) {
4013 r
= exec_context_cpu_affinity_from_numa(context
, &converted_cpu_set
);
4015 *exit_status
= EXIT_CPUAFFINITY
;
4016 return log_unit_error_errno(unit
, r
, "Failed to derive CPU affinity mask from NUMA mask: %m");
4019 cpu_set
= &converted_cpu_set
;
4021 cpu_set
= &context
->cpu_set
;
4023 if (sched_setaffinity(0, cpu_set
->allocated
, cpu_set
->set
) < 0) {
4024 *exit_status
= EXIT_CPUAFFINITY
;
4025 return log_unit_error_errno(unit
, errno
, "Failed to set up CPU affinity: %m");
4029 if (mpol_is_valid(numa_policy_get_type(&context
->numa_policy
))) {
4030 r
= apply_numa_policy(&context
->numa_policy
);
4031 if (r
== -EOPNOTSUPP
)
4032 log_unit_debug_errno(unit
, r
, "NUMA support not available, ignoring.");
4034 *exit_status
= EXIT_NUMA_POLICY
;
4035 return log_unit_error_errno(unit
, r
, "Failed to set NUMA memory policy: %m");
4039 if (context
->ioprio_set
)
4040 if (ioprio_set(IOPRIO_WHO_PROCESS
, 0, context
->ioprio
) < 0) {
4041 *exit_status
= EXIT_IOPRIO
;
4042 return log_unit_error_errno(unit
, errno
, "Failed to set up IO scheduling priority: %m");
4045 if (context
->timer_slack_nsec
!= NSEC_INFINITY
)
4046 if (prctl(PR_SET_TIMERSLACK
, context
->timer_slack_nsec
) < 0) {
4047 *exit_status
= EXIT_TIMERSLACK
;
4048 return log_unit_error_errno(unit
, errno
, "Failed to set up timer slack: %m");
4051 if (context
->personality
!= PERSONALITY_INVALID
) {
4052 r
= safe_personality(context
->personality
);
4054 *exit_status
= EXIT_PERSONALITY
;
4055 return log_unit_error_errno(unit
, r
, "Failed to set up execution domain (personality): %m");
4059 if (context
->utmp_id
)
4060 utmp_put_init_process(context
->utmp_id
, getpid_cached(), getsid(0),
4062 context
->utmp_mode
== EXEC_UTMP_INIT
? INIT_PROCESS
:
4063 context
->utmp_mode
== EXEC_UTMP_LOGIN
? LOGIN_PROCESS
:
4067 if (uid_is_valid(uid
)) {
4068 r
= chown_terminal(STDIN_FILENO
, uid
);
4070 *exit_status
= EXIT_STDIN
;
4071 return log_unit_error_errno(unit
, r
, "Failed to change ownership of terminal: %m");
4075 /* If delegation is enabled we'll pass ownership of the cgroup to the user of the new process. On cgroup v1
4076 * this is only about systemd's own hierarchy, i.e. not the controller hierarchies, simply because that's not
4077 * safe. On cgroup v2 there's only one hierarchy anyway, and delegation is safe there, hence in that case only
4078 * touch a single hierarchy too. */
4079 if (params
->cgroup_path
&& context
->user
&& (params
->flags
& EXEC_CGROUP_DELEGATE
)) {
4080 r
= cg_set_access(SYSTEMD_CGROUP_CONTROLLER
, params
->cgroup_path
, uid
, gid
);
4082 *exit_status
= EXIT_CGROUP
;
4083 return log_unit_error_errno(unit
, r
, "Failed to adjust control group access: %m");
4087 for (ExecDirectoryType dt
= 0; dt
< _EXEC_DIRECTORY_TYPE_MAX
; dt
++) {
4088 r
= setup_exec_directory(context
, params
, uid
, gid
, dt
, exit_status
);
4090 return log_unit_error_errno(unit
, r
, "Failed to set up special execution directory in %s: %m", params
->prefix
[dt
]);
4093 if (FLAGS_SET(params
->flags
, EXEC_WRITE_CREDENTIALS
)) {
4094 r
= setup_credentials(context
, params
, unit
->id
, uid
);
4096 *exit_status
= EXIT_CREDENTIALS
;
4097 return log_unit_error_errno(unit
, r
, "Failed to set up credentials: %m");
4101 r
= build_environment(
4113 *exit_status
= EXIT_MEMORY
;
4117 r
= build_pass_environment(context
, &pass_env
);
4119 *exit_status
= EXIT_MEMORY
;
4123 accum_env
= strv_env_merge(5,
4124 params
->environment
,
4127 context
->environment
,
4130 *exit_status
= EXIT_MEMORY
;
4133 accum_env
= strv_env_clean(accum_env
);
4135 (void) umask(context
->umask
);
4137 r
= setup_keyring(unit
, context
, params
, uid
, gid
);
4139 *exit_status
= EXIT_KEYRING
;
4140 return log_unit_error_errno(unit
, r
, "Failed to set up kernel keyring: %m");
4143 /* We need sandboxing if the caller asked us to apply it and the command isn't explicitly excepted from it */
4144 needs_sandboxing
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && !(command
->flags
& EXEC_COMMAND_FULLY_PRIVILEGED
);
4146 /* 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 */
4147 needs_ambient_hack
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && (command
->flags
& EXEC_COMMAND_AMBIENT_MAGIC
) && !ambient_capabilities_supported();
4149 /* 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 */
4150 if (needs_ambient_hack
)
4151 needs_setuid
= false;
4153 needs_setuid
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && !(command
->flags
& (EXEC_COMMAND_FULLY_PRIVILEGED
|EXEC_COMMAND_NO_SETUID
));
4155 if (needs_sandboxing
) {
4156 /* MAC enablement checks need to be done before a new mount ns is created, as they rely on /sys being
4157 * present. The actual MAC context application will happen later, as late as possible, to avoid
4158 * impacting our own code paths. */
4161 use_selinux
= mac_selinux_use();
4164 use_smack
= mac_smack_use();
4167 use_apparmor
= mac_apparmor_use();
4171 if (needs_sandboxing
) {
4174 /* Let's set the resource limits before we call into PAM, so that pam_limits wins over what
4175 * is set here. (See below.) */
4177 r
= setrlimit_closest_all((const struct rlimit
* const *) context
->rlimit
, &which_failed
);
4179 *exit_status
= EXIT_LIMITS
;
4180 return log_unit_error_errno(unit
, r
, "Failed to adjust resource limit RLIMIT_%s: %m", rlimit_to_string(which_failed
));
4184 if (needs_setuid
&& context
->pam_name
&& username
) {
4185 /* Let's call into PAM after we set up our own idea of resource limits to that pam_limits
4186 * wins here. (See above.) */
4188 /* All fds passed in the fds array will be closed in the pam child process. */
4189 r
= setup_pam(context
->pam_name
, username
, uid
, gid
, context
->tty_path
, &accum_env
, fds
, n_fds
);
4191 *exit_status
= EXIT_PAM
;
4192 return log_unit_error_errno(unit
, r
, "Failed to set up PAM session: %m");
4195 ngids_after_pam
= getgroups_alloc(&gids_after_pam
);
4196 if (ngids_after_pam
< 0) {
4197 *exit_status
= EXIT_MEMORY
;
4198 return log_unit_error_errno(unit
, ngids_after_pam
, "Failed to obtain groups after setting up PAM: %m");
4202 if (needs_sandboxing
&& context
->private_users
&& !have_effective_cap(CAP_SYS_ADMIN
)) {
4203 /* If we're unprivileged, set up the user namespace first to enable use of the other namespaces.
4204 * Users with CAP_SYS_ADMIN can set up user namespaces last because they will be able to
4205 * set up the all of the other namespaces (i.e. network, mount, UTS) without a user namespace. */
4207 userns_set_up
= true;
4208 r
= setup_private_users(saved_uid
, saved_gid
, uid
, gid
);
4210 *exit_status
= EXIT_USER
;
4211 return log_unit_error_errno(unit
, r
, "Failed to set up user namespacing for unprivileged user: %m");
4215 if ((context
->private_network
|| context
->network_namespace_path
) && runtime
&& runtime
->netns_storage_socket
[0] >= 0) {
4217 if (ns_type_supported(NAMESPACE_NET
)) {
4218 r
= setup_shareable_ns(runtime
->netns_storage_socket
, CLONE_NEWNET
);
4220 log_unit_warning_errno(unit
, r
,
4221 "PrivateNetwork=yes is configured, but network namespace setup failed, ignoring: %m");
4223 *exit_status
= EXIT_NETWORK
;
4224 return log_unit_error_errno(unit
, r
, "Failed to set up network namespacing: %m");
4226 } else if (context
->network_namespace_path
) {
4227 *exit_status
= EXIT_NETWORK
;
4228 return log_unit_error_errno(unit
, SYNTHETIC_ERRNO(EOPNOTSUPP
),
4229 "NetworkNamespacePath= is not supported, refusing.");
4231 log_unit_warning(unit
, "PrivateNetwork=yes is configured, but the kernel does not support network namespaces, ignoring.");
4234 if ((context
->private_ipc
|| context
->ipc_namespace_path
) && runtime
&& runtime
->ipcns_storage_socket
[0] >= 0) {
4236 if (ns_type_supported(NAMESPACE_IPC
)) {
4237 r
= setup_shareable_ns(runtime
->ipcns_storage_socket
, CLONE_NEWIPC
);
4239 log_unit_warning_errno(unit
, r
,
4240 "PrivateIPC=yes is configured, but IPC namespace setup failed, ignoring: %m");
4242 *exit_status
= EXIT_NAMESPACE
;
4243 return log_unit_error_errno(unit
, r
, "Failed to set up IPC namespacing: %m");
4245 } else if (context
->ipc_namespace_path
) {
4246 *exit_status
= EXIT_NAMESPACE
;
4247 return log_unit_error_errno(unit
, SYNTHETIC_ERRNO(EOPNOTSUPP
),
4248 "IPCNamespacePath= is not supported, refusing.");
4250 log_unit_warning(unit
, "PrivateIPC=yes is configured, but the kernel does not support IPC namespaces, ignoring.");
4253 needs_mount_namespace
= exec_needs_mount_namespace(context
, params
, runtime
);
4254 if (needs_mount_namespace
) {
4255 _cleanup_free_
char *error_path
= NULL
;
4257 r
= apply_mount_namespace(unit
, command
->flags
, context
, params
, runtime
, &error_path
);
4259 *exit_status
= EXIT_NAMESPACE
;
4260 return log_unit_error_errno(unit
, r
, "Failed to set up mount namespacing%s%s: %m",
4261 error_path
? ": " : "", strempty(error_path
));
4265 if (needs_sandboxing
) {
4266 r
= apply_protect_hostname(unit
, context
, exit_status
);
4271 /* Drop groups as early as possible.
4272 * This needs to be done after PrivateDevices=y setup as device nodes should be owned by the host's root.
4273 * For non-root in a userns, devices will be owned by the user/group before the group change, and nobody. */
4275 _cleanup_free_ gid_t
*gids_to_enforce
= NULL
;
4276 int ngids_to_enforce
= 0;
4278 ngids_to_enforce
= merge_gid_lists(supplementary_gids
,
4283 if (ngids_to_enforce
< 0) {
4284 *exit_status
= EXIT_MEMORY
;
4285 return log_unit_error_errno(unit
,
4287 "Failed to merge group lists. Group membership might be incorrect: %m");
4290 r
= enforce_groups(gid
, gids_to_enforce
, ngids_to_enforce
);
4292 *exit_status
= EXIT_GROUP
;
4293 return log_unit_error_errno(unit
, r
, "Changing group credentials failed: %m");
4297 /* If the user namespace was not set up above, try to do it now.
4298 * It's preferred to set up the user namespace later (after all other namespaces) so as not to be
4299 * restricted by rules pertaining to combining user namspaces with other namespaces (e.g. in the
4300 * case of mount namespaces being less privileged when the mount point list is copied from a
4301 * different user namespace). */
4303 if (needs_sandboxing
&& context
->private_users
&& !userns_set_up
) {
4304 r
= setup_private_users(saved_uid
, saved_gid
, uid
, gid
);
4306 *exit_status
= EXIT_USER
;
4307 return log_unit_error_errno(unit
, r
, "Failed to set up user namespacing: %m");
4311 /* Now that the mount namespace has been set up and privileges adjusted, let's look for the thing we
4314 _cleanup_free_
char *executable
= NULL
;
4315 _cleanup_close_
int executable_fd
= -1;
4316 r
= find_executable_full(command
->path
, false, &executable
, &executable_fd
);
4318 if (r
!= -ENOMEM
&& (command
->flags
& EXEC_COMMAND_IGNORE_FAILURE
)) {
4319 log_unit_struct_errno(unit
, LOG_INFO
, r
,
4320 "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR
,
4321 LOG_UNIT_INVOCATION_ID(unit
),
4322 LOG_UNIT_MESSAGE(unit
, "Executable %s missing, skipping: %m",
4324 "EXECUTABLE=%s", command
->path
);
4328 *exit_status
= EXIT_EXEC
;
4330 return log_unit_struct_errno(unit
, LOG_INFO
, r
,
4331 "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR
,
4332 LOG_UNIT_INVOCATION_ID(unit
),
4333 LOG_UNIT_MESSAGE(unit
, "Failed to locate executable %s: %m",
4335 "EXECUTABLE=%s", command
->path
);
4338 r
= add_shifted_fd(keep_fds
, ELEMENTSOF(keep_fds
), &n_keep_fds
, executable_fd
, &executable_fd
);
4340 *exit_status
= EXIT_FDS
;
4341 return log_unit_error_errno(unit
, r
, "Failed to shift fd and set FD_CLOEXEC: %m");
4345 if (needs_sandboxing
&& use_selinux
&& params
->selinux_context_net
&& socket_fd
>= 0) {
4346 r
= mac_selinux_get_child_mls_label(socket_fd
, executable
, context
->selinux_context
, &mac_selinux_context_net
);
4348 *exit_status
= EXIT_SELINUX_CONTEXT
;
4349 return log_unit_error_errno(unit
, r
, "Failed to determine SELinux context: %m");
4354 /* We repeat the fd closing here, to make sure that nothing is leaked from the PAM modules. Note that we are
4355 * more aggressive this time since socket_fd and the netns and ipcns fds we don't need anymore. We do keep the exec_fd
4356 * however if we have it as we want to keep it open until the final execve(). */
4358 r
= close_all_fds(keep_fds
, n_keep_fds
);
4360 r
= shift_fds(fds
, n_fds
);
4362 r
= flags_fds(fds
, n_socket_fds
, n_storage_fds
, context
->non_blocking
);
4364 *exit_status
= EXIT_FDS
;
4365 return log_unit_error_errno(unit
, r
, "Failed to adjust passed file descriptors: %m");
4368 /* At this point, the fds we want to pass to the program are all ready and set up, with O_CLOEXEC turned off
4369 * and at the right fd numbers. The are no other fds open, with one exception: the exec_fd if it is defined,
4370 * and it has O_CLOEXEC set, after all we want it to be closed by the execve(), so that our parent knows we
4373 secure_bits
= context
->secure_bits
;
4375 if (needs_sandboxing
) {
4378 /* Set the RTPRIO resource limit to 0, but only if nothing else was explicitly
4379 * requested. (Note this is placed after the general resource limit initialization, see
4380 * above, in order to take precedence.) */
4381 if (context
->restrict_realtime
&& !context
->rlimit
[RLIMIT_RTPRIO
]) {
4382 if (setrlimit(RLIMIT_RTPRIO
, &RLIMIT_MAKE_CONST(0)) < 0) {
4383 *exit_status
= EXIT_LIMITS
;
4384 return log_unit_error_errno(unit
, errno
, "Failed to adjust RLIMIT_RTPRIO resource limit: %m");
4389 /* LSM Smack needs the capability CAP_MAC_ADMIN to change the current execution security context of the
4390 * process. This is the latest place before dropping capabilities. Other MAC context are set later. */
4392 r
= setup_smack(context
, executable_fd
);
4394 *exit_status
= EXIT_SMACK_PROCESS_LABEL
;
4395 return log_unit_error_errno(unit
, r
, "Failed to set SMACK process label: %m");
4400 bset
= context
->capability_bounding_set
;
4401 /* If the ambient caps hack is enabled (which means the kernel can't do them, and the user asked for
4402 * our magic fallback), then let's add some extra caps, so that the service can drop privs of its own,
4403 * instead of us doing that */
4404 if (needs_ambient_hack
)
4405 bset
|= (UINT64_C(1) << CAP_SETPCAP
) |
4406 (UINT64_C(1) << CAP_SETUID
) |
4407 (UINT64_C(1) << CAP_SETGID
);
4409 if (!cap_test_all(bset
)) {
4410 r
= capability_bounding_set_drop(bset
, false);
4412 *exit_status
= EXIT_CAPABILITIES
;
4413 return log_unit_error_errno(unit
, r
, "Failed to drop capabilities: %m");
4417 /* Ambient capabilities are cleared during setresuid() (in enforce_user()) even with
4419 * To be able to raise the ambient capabilities after setresuid() they have to be
4420 * added to the inherited set and keep caps has to be set (done in enforce_user()).
4421 * After setresuid() the ambient capabilities can be raised as they are present in
4422 * the permitted and inhertiable set. However it is possible that someone wants to
4423 * set ambient capabilities without changing the user, so we also set the ambient
4424 * capabilities here.
4425 * The requested ambient capabilities are raised in the inheritable set if the
4426 * second argument is true. */
4427 if (!needs_ambient_hack
) {
4428 r
= capability_ambient_set_apply(context
->capability_ambient_set
, true);
4430 *exit_status
= EXIT_CAPABILITIES
;
4431 return log_unit_error_errno(unit
, r
, "Failed to apply ambient capabilities (before UID change): %m");
4436 /* chroot to root directory first, before we lose the ability to chroot */
4437 r
= apply_root_directory(context
, params
, needs_mount_namespace
, exit_status
);
4439 return log_unit_error_errno(unit
, r
, "Chrooting to the requested root directory failed: %m");
4442 if (uid_is_valid(uid
)) {
4443 r
= enforce_user(context
, uid
);
4445 *exit_status
= EXIT_USER
;
4446 return log_unit_error_errno(unit
, r
, "Failed to change UID to " UID_FMT
": %m", uid
);
4449 if (!needs_ambient_hack
&&
4450 context
->capability_ambient_set
!= 0) {
4452 /* Raise the ambient capabilities after user change. */
4453 r
= capability_ambient_set_apply(context
->capability_ambient_set
, false);
4455 *exit_status
= EXIT_CAPABILITIES
;
4456 return log_unit_error_errno(unit
, r
, "Failed to apply ambient capabilities (after UID change): %m");
4462 /* Apply working directory here, because the working directory might be on NFS and only the user running
4463 * this service might have the correct privilege to change to the working directory */
4464 r
= apply_working_directory(context
, params
, home
, exit_status
);
4466 return log_unit_error_errno(unit
, r
, "Changing to the requested working directory failed: %m");
4468 if (needs_sandboxing
) {
4469 /* Apply other MAC contexts late, but before seccomp syscall filtering, as those should really be last to
4470 * influence our own codepaths as little as possible. Moreover, applying MAC contexts usually requires
4471 * syscalls that are subject to seccomp filtering, hence should probably be applied before the syscalls
4472 * are restricted. */
4476 char *exec_context
= mac_selinux_context_net
?: context
->selinux_context
;
4479 r
= setexeccon(exec_context
);
4481 *exit_status
= EXIT_SELINUX_CONTEXT
;
4482 return log_unit_error_errno(unit
, r
, "Failed to change SELinux context to %s: %m", exec_context
);
4489 if (use_apparmor
&& context
->apparmor_profile
) {
4490 r
= aa_change_onexec(context
->apparmor_profile
);
4491 if (r
< 0 && !context
->apparmor_profile_ignore
) {
4492 *exit_status
= EXIT_APPARMOR_PROFILE
;
4493 return log_unit_error_errno(unit
, errno
, "Failed to prepare AppArmor profile change to %s: %m", context
->apparmor_profile
);
4498 /* PR_GET_SECUREBITS is not privileged, while PR_SET_SECUREBITS is. So to suppress potential EPERMs
4499 * we'll try not to call PR_SET_SECUREBITS unless necessary. Setting securebits requires
4501 if (prctl(PR_GET_SECUREBITS
) != secure_bits
) {
4502 /* CAP_SETPCAP is required to set securebits. This capability is raised into the
4503 * effective set here.
4504 * The effective set is overwritten during execve with the following values:
4505 * - ambient set (for non-root processes)
4506 * - (inheritable | bounding) set for root processes)
4508 * Hence there is no security impact to raise it in the effective set before execve
4510 r
= capability_gain_cap_setpcap(NULL
);
4512 *exit_status
= EXIT_CAPABILITIES
;
4513 return log_unit_error_errno(unit
, r
, "Failed to gain CAP_SETPCAP for setting secure bits");
4515 if (prctl(PR_SET_SECUREBITS
, secure_bits
) < 0) {
4516 *exit_status
= EXIT_SECUREBITS
;
4517 return log_unit_error_errno(unit
, errno
, "Failed to set process secure bits: %m");
4521 if (context_has_no_new_privileges(context
))
4522 if (prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0) < 0) {
4523 *exit_status
= EXIT_NO_NEW_PRIVILEGES
;
4524 return log_unit_error_errno(unit
, errno
, "Failed to disable new privileges: %m");
4528 r
= apply_address_families(unit
, context
);
4530 *exit_status
= EXIT_ADDRESS_FAMILIES
;
4531 return log_unit_error_errno(unit
, r
, "Failed to restrict address families: %m");
4534 r
= apply_memory_deny_write_execute(unit
, context
);
4536 *exit_status
= EXIT_SECCOMP
;
4537 return log_unit_error_errno(unit
, r
, "Failed to disable writing to executable memory: %m");
4540 r
= apply_restrict_realtime(unit
, context
);
4542 *exit_status
= EXIT_SECCOMP
;
4543 return log_unit_error_errno(unit
, r
, "Failed to apply realtime restrictions: %m");
4546 r
= apply_restrict_suid_sgid(unit
, context
);
4548 *exit_status
= EXIT_SECCOMP
;
4549 return log_unit_error_errno(unit
, r
, "Failed to apply SUID/SGID restrictions: %m");
4552 r
= apply_restrict_namespaces(unit
, context
);
4554 *exit_status
= EXIT_SECCOMP
;
4555 return log_unit_error_errno(unit
, r
, "Failed to apply namespace restrictions: %m");
4558 r
= apply_protect_sysctl(unit
, context
);
4560 *exit_status
= EXIT_SECCOMP
;
4561 return log_unit_error_errno(unit
, r
, "Failed to apply sysctl restrictions: %m");
4564 r
= apply_protect_kernel_modules(unit
, context
);
4566 *exit_status
= EXIT_SECCOMP
;
4567 return log_unit_error_errno(unit
, r
, "Failed to apply module loading restrictions: %m");
4570 r
= apply_protect_kernel_logs(unit
, context
);
4572 *exit_status
= EXIT_SECCOMP
;
4573 return log_unit_error_errno(unit
, r
, "Failed to apply kernel log restrictions: %m");
4576 r
= apply_protect_clock(unit
, context
);
4578 *exit_status
= EXIT_SECCOMP
;
4579 return log_unit_error_errno(unit
, r
, "Failed to apply clock restrictions: %m");
4582 r
= apply_private_devices(unit
, context
);
4584 *exit_status
= EXIT_SECCOMP
;
4585 return log_unit_error_errno(unit
, r
, "Failed to set up private devices: %m");
4588 r
= apply_syscall_archs(unit
, context
);
4590 *exit_status
= EXIT_SECCOMP
;
4591 return log_unit_error_errno(unit
, r
, "Failed to apply syscall architecture restrictions: %m");
4594 r
= apply_lock_personality(unit
, context
);
4596 *exit_status
= EXIT_SECCOMP
;
4597 return log_unit_error_errno(unit
, r
, "Failed to lock personalities: %m");
4600 r
= apply_syscall_log(unit
, context
);
4602 *exit_status
= EXIT_SECCOMP
;
4603 return log_unit_error_errno(unit
, r
, "Failed to apply system call log filters: %m");
4606 /* This really should remain the last step before the execve(), to make sure our own code is unaffected
4607 * by the filter as little as possible. */
4608 r
= apply_syscall_filter(unit
, context
, needs_ambient_hack
);
4610 *exit_status
= EXIT_SECCOMP
;
4611 return log_unit_error_errno(unit
, r
, "Failed to apply system call filters: %m");
4616 if (!strv_isempty(context
->unset_environment
)) {
4619 ee
= strv_env_delete(accum_env
, 1, context
->unset_environment
);
4621 *exit_status
= EXIT_MEMORY
;
4625 strv_free_and_replace(accum_env
, ee
);
4628 if (!FLAGS_SET(command
->flags
, EXEC_COMMAND_NO_ENV_EXPAND
)) {
4629 replaced_argv
= replace_env_argv(command
->argv
, accum_env
);
4630 if (!replaced_argv
) {
4631 *exit_status
= EXIT_MEMORY
;
4634 final_argv
= replaced_argv
;
4636 final_argv
= command
->argv
;
4638 if (DEBUG_LOGGING
) {
4639 _cleanup_free_
char *line
= NULL
;
4641 line
= exec_command_line(final_argv
);
4643 log_unit_struct(unit
, LOG_DEBUG
,
4644 "EXECUTABLE=%s", executable
,
4645 LOG_UNIT_MESSAGE(unit
, "Executing: %s", line
),
4646 LOG_UNIT_INVOCATION_ID(unit
));
4652 /* We have finished with all our initializations. Let's now let the manager know that. From this point
4653 * on, if the manager sees POLLHUP on the exec_fd, then execve() was successful. */
4655 if (write(exec_fd
, &hot
, sizeof(hot
)) < 0) {
4656 *exit_status
= EXIT_EXEC
;
4657 return log_unit_error_errno(unit
, errno
, "Failed to enable exec_fd: %m");
4661 r
= fexecve_or_execve(executable_fd
, executable
, final_argv
, accum_env
);
4666 /* The execve() failed. This means the exec_fd is still open. Which means we need to tell the manager
4667 * that POLLHUP on it no longer means execve() succeeded. */
4669 if (write(exec_fd
, &hot
, sizeof(hot
)) < 0) {
4670 *exit_status
= EXIT_EXEC
;
4671 return log_unit_error_errno(unit
, errno
, "Failed to disable exec_fd: %m");
4675 *exit_status
= EXIT_EXEC
;
4676 return log_unit_error_errno(unit
, r
, "Failed to execute %s: %m", executable
);
4679 static int exec_context_load_environment(const Unit
*unit
, const ExecContext
*c
, char ***l
);
4680 static int exec_context_named_iofds(const ExecContext
*c
, const ExecParameters
*p
, int named_iofds
[static 3]);
4682 int exec_spawn(Unit
*unit
,
4683 ExecCommand
*command
,
4684 const ExecContext
*context
,
4685 const ExecParameters
*params
,
4686 ExecRuntime
*runtime
,
4687 DynamicCreds
*dcreds
,
4690 int socket_fd
, r
, named_iofds
[3] = { -1, -1, -1 }, *fds
= NULL
;
4691 _cleanup_free_
char *subcgroup_path
= NULL
;
4692 _cleanup_strv_free_
char **files_env
= NULL
;
4693 size_t n_storage_fds
= 0, n_socket_fds
= 0;
4694 _cleanup_free_
char *line
= NULL
;
4702 assert(params
->fds
|| (params
->n_socket_fds
+ params
->n_storage_fds
<= 0));
4704 if (context
->std_input
== EXEC_INPUT_SOCKET
||
4705 context
->std_output
== EXEC_OUTPUT_SOCKET
||
4706 context
->std_error
== EXEC_OUTPUT_SOCKET
) {
4708 if (params
->n_socket_fds
> 1)
4709 return log_unit_error_errno(unit
, SYNTHETIC_ERRNO(EINVAL
), "Got more than one socket.");
4711 if (params
->n_socket_fds
== 0)
4712 return log_unit_error_errno(unit
, SYNTHETIC_ERRNO(EINVAL
), "Got no socket.");
4714 socket_fd
= params
->fds
[0];
4718 n_socket_fds
= params
->n_socket_fds
;
4719 n_storage_fds
= params
->n_storage_fds
;
4722 r
= exec_context_named_iofds(context
, params
, named_iofds
);
4724 return log_unit_error_errno(unit
, r
, "Failed to load a named file descriptor: %m");
4726 r
= exec_context_load_environment(unit
, context
, &files_env
);
4728 return log_unit_error_errno(unit
, r
, "Failed to load environment files: %m");
4730 line
= exec_command_line(command
->argv
);
4734 /* Fork with up-to-date SELinux label database, so the child inherits the up-to-date db
4735 and, until the next SELinux policy changes, we save further reloads in future children. */
4736 mac_selinux_maybe_reload();
4738 log_unit_struct(unit
, LOG_DEBUG
,
4739 LOG_UNIT_MESSAGE(unit
, "About to execute %s", line
),
4740 "EXECUTABLE=%s", command
->path
, /* We won't know the real executable path until we create
4741 the mount namespace in the child, but we want to log
4742 from the parent, so we need to use the (possibly
4743 inaccurate) path here. */
4744 LOG_UNIT_INVOCATION_ID(unit
));
4746 if (params
->cgroup_path
) {
4747 r
= exec_parameters_get_cgroup_path(params
, &subcgroup_path
);
4749 return log_unit_error_errno(unit
, r
, "Failed to acquire subcgroup path: %m");
4750 if (r
> 0) { /* We are using a child cgroup */
4751 r
= cg_create(SYSTEMD_CGROUP_CONTROLLER
, subcgroup_path
);
4753 return log_unit_error_errno(unit
, r
, "Failed to create control group '%s': %m", subcgroup_path
);
4755 /* Normally we would not propagate the oomd xattrs to children but since we created this
4756 * sub-cgroup internally we should do it. */
4757 cgroup_oomd_xattr_apply(unit
, subcgroup_path
);
4763 return log_unit_error_errno(unit
, errno
, "Failed to fork: %m");
4766 int exit_status
= EXIT_SUCCESS
;
4768 r
= exec_child(unit
,
4780 unit
->manager
->user_lookup_fds
[1],
4784 const char *status
=
4785 exit_status_to_string(exit_status
,
4786 EXIT_STATUS_LIBC
| EXIT_STATUS_SYSTEMD
);
4788 log_unit_struct_errno(unit
, LOG_ERR
, r
,
4789 "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR
,
4790 LOG_UNIT_INVOCATION_ID(unit
),
4791 LOG_UNIT_MESSAGE(unit
, "Failed at step %s spawning %s: %m",
4792 status
, command
->path
),
4793 "EXECUTABLE=%s", command
->path
);
4799 log_unit_debug(unit
, "Forked %s as "PID_FMT
, command
->path
, pid
);
4801 /* We add the new process to the cgroup both in the child (so that we can be sure that no user code is ever
4802 * executed outside of the cgroup) and in the parent (so that we can be sure that when we kill the cgroup the
4803 * process will be killed too). */
4805 (void) cg_attach(SYSTEMD_CGROUP_CONTROLLER
, subcgroup_path
, pid
);
4807 exec_status_start(&command
->exec_status
, pid
);
4813 void exec_context_init(ExecContext
*c
) {
4817 c
->ioprio
= IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE
, 0);
4818 c
->cpu_sched_policy
= SCHED_OTHER
;
4819 c
->syslog_priority
= LOG_DAEMON
|LOG_INFO
;
4820 c
->syslog_level_prefix
= true;
4821 c
->ignore_sigpipe
= true;
4822 c
->timer_slack_nsec
= NSEC_INFINITY
;
4823 c
->personality
= PERSONALITY_INVALID
;
4824 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++)
4825 c
->directories
[t
].mode
= 0755;
4826 c
->timeout_clean_usec
= USEC_INFINITY
;
4827 c
->capability_bounding_set
= CAP_ALL
;
4828 assert_cc(NAMESPACE_FLAGS_INITIAL
!= NAMESPACE_FLAGS_ALL
);
4829 c
->restrict_namespaces
= NAMESPACE_FLAGS_INITIAL
;
4830 c
->log_level_max
= -1;
4832 c
->syscall_errno
= SECCOMP_ERROR_NUMBER_KILL
;
4834 numa_policy_reset(&c
->numa_policy
);
4837 void exec_context_done(ExecContext
*c
) {
4840 c
->environment
= strv_free(c
->environment
);
4841 c
->environment_files
= strv_free(c
->environment_files
);
4842 c
->pass_environment
= strv_free(c
->pass_environment
);
4843 c
->unset_environment
= strv_free(c
->unset_environment
);
4845 rlimit_free_all(c
->rlimit
);
4847 for (size_t l
= 0; l
< 3; l
++) {
4848 c
->stdio_fdname
[l
] = mfree(c
->stdio_fdname
[l
]);
4849 c
->stdio_file
[l
] = mfree(c
->stdio_file
[l
]);
4852 c
->working_directory
= mfree(c
->working_directory
);
4853 c
->root_directory
= mfree(c
->root_directory
);
4854 c
->root_image
= mfree(c
->root_image
);
4855 c
->root_image_options
= mount_options_free_all(c
->root_image_options
);
4856 c
->root_hash
= mfree(c
->root_hash
);
4857 c
->root_hash_size
= 0;
4858 c
->root_hash_path
= mfree(c
->root_hash_path
);
4859 c
->root_hash_sig
= mfree(c
->root_hash_sig
);
4860 c
->root_hash_sig_size
= 0;
4861 c
->root_hash_sig_path
= mfree(c
->root_hash_sig_path
);
4862 c
->root_verity
= mfree(c
->root_verity
);
4863 c
->extension_images
= mount_image_free_many(c
->extension_images
, &c
->n_extension_images
);
4864 c
->tty_path
= mfree(c
->tty_path
);
4865 c
->syslog_identifier
= mfree(c
->syslog_identifier
);
4866 c
->user
= mfree(c
->user
);
4867 c
->group
= mfree(c
->group
);
4869 c
->supplementary_groups
= strv_free(c
->supplementary_groups
);
4871 c
->pam_name
= mfree(c
->pam_name
);
4873 c
->read_only_paths
= strv_free(c
->read_only_paths
);
4874 c
->read_write_paths
= strv_free(c
->read_write_paths
);
4875 c
->inaccessible_paths
= strv_free(c
->inaccessible_paths
);
4876 c
->exec_paths
= strv_free(c
->exec_paths
);
4877 c
->no_exec_paths
= strv_free(c
->no_exec_paths
);
4879 bind_mount_free_many(c
->bind_mounts
, c
->n_bind_mounts
);
4880 c
->bind_mounts
= NULL
;
4881 c
->n_bind_mounts
= 0;
4882 temporary_filesystem_free_many(c
->temporary_filesystems
, c
->n_temporary_filesystems
);
4883 c
->temporary_filesystems
= NULL
;
4884 c
->n_temporary_filesystems
= 0;
4885 c
->mount_images
= mount_image_free_many(c
->mount_images
, &c
->n_mount_images
);
4887 cpu_set_reset(&c
->cpu_set
);
4888 numa_policy_reset(&c
->numa_policy
);
4890 c
->utmp_id
= mfree(c
->utmp_id
);
4891 c
->selinux_context
= mfree(c
->selinux_context
);
4892 c
->apparmor_profile
= mfree(c
->apparmor_profile
);
4893 c
->smack_process_label
= mfree(c
->smack_process_label
);
4895 c
->syscall_filter
= hashmap_free(c
->syscall_filter
);
4896 c
->syscall_archs
= set_free(c
->syscall_archs
);
4897 c
->address_families
= set_free(c
->address_families
);
4899 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++)
4900 c
->directories
[t
].paths
= strv_free(c
->directories
[t
].paths
);
4902 c
->log_level_max
= -1;
4904 exec_context_free_log_extra_fields(c
);
4906 c
->log_ratelimit_interval_usec
= 0;
4907 c
->log_ratelimit_burst
= 0;
4909 c
->stdin_data
= mfree(c
->stdin_data
);
4910 c
->stdin_data_size
= 0;
4912 c
->network_namespace_path
= mfree(c
->network_namespace_path
);
4913 c
->ipc_namespace_path
= mfree(c
->ipc_namespace_path
);
4915 c
->log_namespace
= mfree(c
->log_namespace
);
4917 c
->load_credentials
= strv_free(c
->load_credentials
);
4918 c
->set_credentials
= hashmap_free(c
->set_credentials
);
4921 int exec_context_destroy_runtime_directory(const ExecContext
*c
, const char *runtime_prefix
) {
4926 if (!runtime_prefix
)
4929 STRV_FOREACH(i
, c
->directories
[EXEC_DIRECTORY_RUNTIME
].paths
) {
4930 _cleanup_free_
char *p
= NULL
;
4932 if (exec_directory_is_private(c
, EXEC_DIRECTORY_RUNTIME
))
4933 p
= path_join(runtime_prefix
, "private", *i
);
4935 p
= path_join(runtime_prefix
, *i
);
4939 /* We execute this synchronously, since we need to be sure this is gone when we start the
4941 (void) rm_rf(p
, REMOVE_ROOT
);
4947 int exec_context_destroy_credentials(const ExecContext
*c
, const char *runtime_prefix
, const char *unit
) {
4948 _cleanup_free_
char *p
= NULL
;
4952 if (!runtime_prefix
|| !unit
)
4955 p
= path_join(runtime_prefix
, "credentials", unit
);
4959 /* This is either a tmpfs/ramfs of its own, or a plain directory. Either way, let's first try to
4960 * unmount it, and afterwards remove the mount point */
4961 (void) umount2(p
, MNT_DETACH
|UMOUNT_NOFOLLOW
);
4962 (void) rm_rf(p
, REMOVE_ROOT
|REMOVE_CHMOD
);
4967 static void exec_command_done(ExecCommand
*c
) {
4970 c
->path
= mfree(c
->path
);
4971 c
->argv
= strv_free(c
->argv
);
4974 void exec_command_done_array(ExecCommand
*c
, size_t n
) {
4975 for (size_t i
= 0; i
< n
; i
++)
4976 exec_command_done(c
+i
);
4979 ExecCommand
* exec_command_free_list(ExecCommand
*c
) {
4983 LIST_REMOVE(command
, c
, i
);
4984 exec_command_done(i
);
4991 void exec_command_free_array(ExecCommand
**c
, size_t n
) {
4992 for (size_t i
= 0; i
< n
; i
++)
4993 c
[i
] = exec_command_free_list(c
[i
]);
4996 void exec_command_reset_status_array(ExecCommand
*c
, size_t n
) {
4997 for (size_t i
= 0; i
< n
; i
++)
4998 exec_status_reset(&c
[i
].exec_status
);
5001 void exec_command_reset_status_list_array(ExecCommand
**c
, size_t n
) {
5002 for (size_t i
= 0; i
< n
; i
++) {
5005 LIST_FOREACH(command
, z
, c
[i
])
5006 exec_status_reset(&z
->exec_status
);
5010 typedef struct InvalidEnvInfo
{
5015 static void invalid_env(const char *p
, void *userdata
) {
5016 InvalidEnvInfo
*info
= userdata
;
5018 log_unit_error(info
->unit
, "Ignoring invalid environment assignment '%s': %s", p
, info
->path
);
5021 const char* exec_context_fdname(const ExecContext
*c
, int fd_index
) {
5027 if (c
->std_input
!= EXEC_INPUT_NAMED_FD
)
5030 return c
->stdio_fdname
[STDIN_FILENO
] ?: "stdin";
5033 if (c
->std_output
!= EXEC_OUTPUT_NAMED_FD
)
5036 return c
->stdio_fdname
[STDOUT_FILENO
] ?: "stdout";
5039 if (c
->std_error
!= EXEC_OUTPUT_NAMED_FD
)
5042 return c
->stdio_fdname
[STDERR_FILENO
] ?: "stderr";
5049 static int exec_context_named_iofds(
5050 const ExecContext
*c
,
5051 const ExecParameters
*p
,
5052 int named_iofds
[static 3]) {
5055 const char* stdio_fdname
[3];
5060 assert(named_iofds
);
5062 targets
= (c
->std_input
== EXEC_INPUT_NAMED_FD
) +
5063 (c
->std_output
== EXEC_OUTPUT_NAMED_FD
) +
5064 (c
->std_error
== EXEC_OUTPUT_NAMED_FD
);
5066 for (size_t i
= 0; i
< 3; i
++)
5067 stdio_fdname
[i
] = exec_context_fdname(c
, i
);
5069 n_fds
= p
->n_storage_fds
+ p
->n_socket_fds
;
5071 for (size_t i
= 0; i
< n_fds
&& targets
> 0; i
++)
5072 if (named_iofds
[STDIN_FILENO
] < 0 &&
5073 c
->std_input
== EXEC_INPUT_NAMED_FD
&&
5074 stdio_fdname
[STDIN_FILENO
] &&
5075 streq(p
->fd_names
[i
], stdio_fdname
[STDIN_FILENO
])) {
5077 named_iofds
[STDIN_FILENO
] = p
->fds
[i
];
5080 } else if (named_iofds
[STDOUT_FILENO
] < 0 &&
5081 c
->std_output
== EXEC_OUTPUT_NAMED_FD
&&
5082 stdio_fdname
[STDOUT_FILENO
] &&
5083 streq(p
->fd_names
[i
], stdio_fdname
[STDOUT_FILENO
])) {
5085 named_iofds
[STDOUT_FILENO
] = p
->fds
[i
];
5088 } else if (named_iofds
[STDERR_FILENO
] < 0 &&
5089 c
->std_error
== EXEC_OUTPUT_NAMED_FD
&&
5090 stdio_fdname
[STDERR_FILENO
] &&
5091 streq(p
->fd_names
[i
], stdio_fdname
[STDERR_FILENO
])) {
5093 named_iofds
[STDERR_FILENO
] = p
->fds
[i
];
5097 return targets
== 0 ? 0 : -ENOENT
;
5100 static int exec_context_load_environment(const Unit
*unit
, const ExecContext
*c
, char ***l
) {
5101 char **i
, **r
= NULL
;
5106 STRV_FOREACH(i
, c
->environment_files
) {
5109 bool ignore
= false;
5111 _cleanup_globfree_ glob_t pglob
= {};
5120 if (!path_is_absolute(fn
)) {
5128 /* Filename supports globbing, take all matching files */
5129 k
= safe_glob(fn
, 0, &pglob
);
5138 /* When we don't match anything, -ENOENT should be returned */
5139 assert(pglob
.gl_pathc
> 0);
5141 for (unsigned n
= 0; n
< pglob
.gl_pathc
; n
++) {
5142 k
= load_env_file(NULL
, pglob
.gl_pathv
[n
], &p
);
5150 /* Log invalid environment variables with filename */
5152 InvalidEnvInfo info
= {
5154 .path
= pglob
.gl_pathv
[n
]
5157 p
= strv_env_clean_with_callback(p
, invalid_env
, &info
);
5165 m
= strv_env_merge(2, r
, p
);
5181 static bool tty_may_match_dev_console(const char *tty
) {
5182 _cleanup_free_
char *resolved
= NULL
;
5187 tty
= skip_dev_prefix(tty
);
5189 /* trivial identity? */
5190 if (streq(tty
, "console"))
5193 if (resolve_dev_console(&resolved
) < 0)
5194 return true; /* if we could not resolve, assume it may */
5196 /* "tty0" means the active VC, so it may be the same sometimes */
5197 return path_equal(resolved
, tty
) || (streq(resolved
, "tty0") && tty_is_vc(tty
));
5200 static bool exec_context_may_touch_tty(const ExecContext
*ec
) {
5203 return ec
->tty_reset
||
5205 ec
->tty_vt_disallocate
||
5206 is_terminal_input(ec
->std_input
) ||
5207 is_terminal_output(ec
->std_output
) ||
5208 is_terminal_output(ec
->std_error
);
5211 bool exec_context_may_touch_console(const ExecContext
*ec
) {
5213 return exec_context_may_touch_tty(ec
) &&
5214 tty_may_match_dev_console(exec_context_tty_path(ec
));
5217 static void strv_fprintf(FILE *f
, char **l
) {
5223 fprintf(f
, " %s", *g
);
5226 static void strv_dump(FILE* f
, const char *prefix
, const char *name
, char **strv
) {
5231 if (!strv_isempty(strv
)) {
5232 fprintf(f
, "%s%s:", prefix
, name
);
5233 strv_fprintf(f
, strv
);
5238 void exec_context_dump(const ExecContext
*c
, FILE* f
, const char *prefix
) {
5239 char **e
, **d
, buf_clean
[FORMAT_TIMESPAN_MAX
];
5245 prefix
= strempty(prefix
);
5249 "%sWorkingDirectory: %s\n"
5250 "%sRootDirectory: %s\n"
5251 "%sNonBlocking: %s\n"
5252 "%sPrivateTmp: %s\n"
5253 "%sPrivateDevices: %s\n"
5254 "%sProtectKernelTunables: %s\n"
5255 "%sProtectKernelModules: %s\n"
5256 "%sProtectKernelLogs: %s\n"
5257 "%sProtectClock: %s\n"
5258 "%sProtectControlGroups: %s\n"
5259 "%sPrivateNetwork: %s\n"
5260 "%sPrivateUsers: %s\n"
5261 "%sProtectHome: %s\n"
5262 "%sProtectSystem: %s\n"
5263 "%sMountAPIVFS: %s\n"
5264 "%sIgnoreSIGPIPE: %s\n"
5265 "%sMemoryDenyWriteExecute: %s\n"
5266 "%sRestrictRealtime: %s\n"
5267 "%sRestrictSUIDSGID: %s\n"
5268 "%sKeyringMode: %s\n"
5269 "%sProtectHostname: %s\n"
5270 "%sProtectProc: %s\n"
5271 "%sProcSubset: %s\n",
5273 prefix
, empty_to_root(c
->working_directory
),
5274 prefix
, empty_to_root(c
->root_directory
),
5275 prefix
, yes_no(c
->non_blocking
),
5276 prefix
, yes_no(c
->private_tmp
),
5277 prefix
, yes_no(c
->private_devices
),
5278 prefix
, yes_no(c
->protect_kernel_tunables
),
5279 prefix
, yes_no(c
->protect_kernel_modules
),
5280 prefix
, yes_no(c
->protect_kernel_logs
),
5281 prefix
, yes_no(c
->protect_clock
),
5282 prefix
, yes_no(c
->protect_control_groups
),
5283 prefix
, yes_no(c
->private_network
),
5284 prefix
, yes_no(c
->private_users
),
5285 prefix
, protect_home_to_string(c
->protect_home
),
5286 prefix
, protect_system_to_string(c
->protect_system
),
5287 prefix
, yes_no(exec_context_get_effective_mount_apivfs(c
)),
5288 prefix
, yes_no(c
->ignore_sigpipe
),
5289 prefix
, yes_no(c
->memory_deny_write_execute
),
5290 prefix
, yes_no(c
->restrict_realtime
),
5291 prefix
, yes_no(c
->restrict_suid_sgid
),
5292 prefix
, exec_keyring_mode_to_string(c
->keyring_mode
),
5293 prefix
, yes_no(c
->protect_hostname
),
5294 prefix
, protect_proc_to_string(c
->protect_proc
),
5295 prefix
, proc_subset_to_string(c
->proc_subset
));
5298 fprintf(f
, "%sRootImage: %s\n", prefix
, c
->root_image
);
5300 if (c
->root_image_options
) {
5303 fprintf(f
, "%sRootImageOptions:", prefix
);
5304 LIST_FOREACH(mount_options
, o
, c
->root_image_options
)
5305 if (!isempty(o
->options
))
5306 fprintf(f
, " %s:%s",
5307 partition_designator_to_string(o
->partition_designator
),
5313 _cleanup_free_
char *encoded
= NULL
;
5314 encoded
= hexmem(c
->root_hash
, c
->root_hash_size
);
5316 fprintf(f
, "%sRootHash: %s\n", prefix
, encoded
);
5319 if (c
->root_hash_path
)
5320 fprintf(f
, "%sRootHash: %s\n", prefix
, c
->root_hash_path
);
5322 if (c
->root_hash_sig
) {
5323 _cleanup_free_
char *encoded
= NULL
;
5325 len
= base64mem(c
->root_hash_sig
, c
->root_hash_sig_size
, &encoded
);
5327 fprintf(f
, "%sRootHashSignature: base64:%s\n", prefix
, encoded
);
5330 if (c
->root_hash_sig_path
)
5331 fprintf(f
, "%sRootHashSignature: %s\n", prefix
, c
->root_hash_sig_path
);
5334 fprintf(f
, "%sRootVerity: %s\n", prefix
, c
->root_verity
);
5336 STRV_FOREACH(e
, c
->environment
)
5337 fprintf(f
, "%sEnvironment: %s\n", prefix
, *e
);
5339 STRV_FOREACH(e
, c
->environment_files
)
5340 fprintf(f
, "%sEnvironmentFile: %s\n", prefix
, *e
);
5342 STRV_FOREACH(e
, c
->pass_environment
)
5343 fprintf(f
, "%sPassEnvironment: %s\n", prefix
, *e
);
5345 STRV_FOREACH(e
, c
->unset_environment
)
5346 fprintf(f
, "%sUnsetEnvironment: %s\n", prefix
, *e
);
5348 fprintf(f
, "%sRuntimeDirectoryPreserve: %s\n", prefix
, exec_preserve_mode_to_string(c
->runtime_directory_preserve_mode
));
5350 for (ExecDirectoryType dt
= 0; dt
< _EXEC_DIRECTORY_TYPE_MAX
; dt
++) {
5351 fprintf(f
, "%s%sMode: %04o\n", prefix
, exec_directory_type_to_string(dt
), c
->directories
[dt
].mode
);
5353 STRV_FOREACH(d
, c
->directories
[dt
].paths
)
5354 fprintf(f
, "%s%s: %s\n", prefix
, exec_directory_type_to_string(dt
), *d
);
5358 "%sTimeoutCleanSec: %s\n",
5359 prefix
, format_timespan(buf_clean
, sizeof(buf_clean
), c
->timeout_clean_usec
, USEC_PER_SEC
));
5366 if (c
->oom_score_adjust_set
)
5368 "%sOOMScoreAdjust: %i\n",
5369 prefix
, c
->oom_score_adjust
);
5371 if (c
->coredump_filter_set
)
5373 "%sCoredumpFilter: 0x%"PRIx64
"\n",
5374 prefix
, c
->coredump_filter
);
5376 for (unsigned i
= 0; i
< RLIM_NLIMITS
; i
++)
5378 fprintf(f
, "%sLimit%s: " RLIM_FMT
"\n",
5379 prefix
, rlimit_to_string(i
), c
->rlimit
[i
]->rlim_max
);
5380 fprintf(f
, "%sLimit%sSoft: " RLIM_FMT
"\n",
5381 prefix
, rlimit_to_string(i
), c
->rlimit
[i
]->rlim_cur
);
5384 if (c
->ioprio_set
) {
5385 _cleanup_free_
char *class_str
= NULL
;
5387 r
= ioprio_class_to_string_alloc(IOPRIO_PRIO_CLASS(c
->ioprio
), &class_str
);
5389 fprintf(f
, "%sIOSchedulingClass: %s\n", prefix
, class_str
);
5391 fprintf(f
, "%sIOPriority: %lu\n", prefix
, IOPRIO_PRIO_DATA(c
->ioprio
));
5394 if (c
->cpu_sched_set
) {
5395 _cleanup_free_
char *policy_str
= NULL
;
5397 r
= sched_policy_to_string_alloc(c
->cpu_sched_policy
, &policy_str
);
5399 fprintf(f
, "%sCPUSchedulingPolicy: %s\n", prefix
, policy_str
);
5402 "%sCPUSchedulingPriority: %i\n"
5403 "%sCPUSchedulingResetOnFork: %s\n",
5404 prefix
, c
->cpu_sched_priority
,
5405 prefix
, yes_no(c
->cpu_sched_reset_on_fork
));
5408 if (c
->cpu_set
.set
) {
5409 _cleanup_free_
char *affinity
= NULL
;
5411 affinity
= cpu_set_to_range_string(&c
->cpu_set
);
5412 fprintf(f
, "%sCPUAffinity: %s\n", prefix
, affinity
);
5415 if (mpol_is_valid(numa_policy_get_type(&c
->numa_policy
))) {
5416 _cleanup_free_
char *nodes
= NULL
;
5418 nodes
= cpu_set_to_range_string(&c
->numa_policy
.nodes
);
5419 fprintf(f
, "%sNUMAPolicy: %s\n", prefix
, mpol_to_string(numa_policy_get_type(&c
->numa_policy
)));
5420 fprintf(f
, "%sNUMAMask: %s\n", prefix
, strnull(nodes
));
5423 if (c
->timer_slack_nsec
!= NSEC_INFINITY
)
5424 fprintf(f
, "%sTimerSlackNSec: "NSEC_FMT
"\n", prefix
, c
->timer_slack_nsec
);
5427 "%sStandardInput: %s\n"
5428 "%sStandardOutput: %s\n"
5429 "%sStandardError: %s\n",
5430 prefix
, exec_input_to_string(c
->std_input
),
5431 prefix
, exec_output_to_string(c
->std_output
),
5432 prefix
, exec_output_to_string(c
->std_error
));
5434 if (c
->std_input
== EXEC_INPUT_NAMED_FD
)
5435 fprintf(f
, "%sStandardInputFileDescriptorName: %s\n", prefix
, c
->stdio_fdname
[STDIN_FILENO
]);
5436 if (c
->std_output
== EXEC_OUTPUT_NAMED_FD
)
5437 fprintf(f
, "%sStandardOutputFileDescriptorName: %s\n", prefix
, c
->stdio_fdname
[STDOUT_FILENO
]);
5438 if (c
->std_error
== EXEC_OUTPUT_NAMED_FD
)
5439 fprintf(f
, "%sStandardErrorFileDescriptorName: %s\n", prefix
, c
->stdio_fdname
[STDERR_FILENO
]);
5441 if (c
->std_input
== EXEC_INPUT_FILE
)
5442 fprintf(f
, "%sStandardInputFile: %s\n", prefix
, c
->stdio_file
[STDIN_FILENO
]);
5443 if (c
->std_output
== EXEC_OUTPUT_FILE
)
5444 fprintf(f
, "%sStandardOutputFile: %s\n", prefix
, c
->stdio_file
[STDOUT_FILENO
]);
5445 if (c
->std_output
== EXEC_OUTPUT_FILE_APPEND
)
5446 fprintf(f
, "%sStandardOutputFileToAppend: %s\n", prefix
, c
->stdio_file
[STDOUT_FILENO
]);
5447 if (c
->std_output
== EXEC_OUTPUT_FILE_TRUNCATE
)
5448 fprintf(f
, "%sStandardOutputFileToTruncate: %s\n", prefix
, c
->stdio_file
[STDOUT_FILENO
]);
5449 if (c
->std_error
== EXEC_OUTPUT_FILE
)
5450 fprintf(f
, "%sStandardErrorFile: %s\n", prefix
, c
->stdio_file
[STDERR_FILENO
]);
5451 if (c
->std_error
== EXEC_OUTPUT_FILE_APPEND
)
5452 fprintf(f
, "%sStandardErrorFileToAppend: %s\n", prefix
, c
->stdio_file
[STDERR_FILENO
]);
5453 if (c
->std_error
== EXEC_OUTPUT_FILE_TRUNCATE
)
5454 fprintf(f
, "%sStandardErrorFileToTruncate: %s\n", prefix
, c
->stdio_file
[STDERR_FILENO
]);
5460 "%sTTYVHangup: %s\n"
5461 "%sTTYVTDisallocate: %s\n",
5462 prefix
, c
->tty_path
,
5463 prefix
, yes_no(c
->tty_reset
),
5464 prefix
, yes_no(c
->tty_vhangup
),
5465 prefix
, yes_no(c
->tty_vt_disallocate
));
5467 if (IN_SET(c
->std_output
,
5469 EXEC_OUTPUT_JOURNAL
,
5470 EXEC_OUTPUT_KMSG_AND_CONSOLE
,
5471 EXEC_OUTPUT_JOURNAL_AND_CONSOLE
) ||
5472 IN_SET(c
->std_error
,
5474 EXEC_OUTPUT_JOURNAL
,
5475 EXEC_OUTPUT_KMSG_AND_CONSOLE
,
5476 EXEC_OUTPUT_JOURNAL_AND_CONSOLE
)) {
5478 _cleanup_free_
char *fac_str
= NULL
, *lvl_str
= NULL
;
5480 r
= log_facility_unshifted_to_string_alloc(c
->syslog_priority
>> 3, &fac_str
);
5482 fprintf(f
, "%sSyslogFacility: %s\n", prefix
, fac_str
);
5484 r
= log_level_to_string_alloc(LOG_PRI(c
->syslog_priority
), &lvl_str
);
5486 fprintf(f
, "%sSyslogLevel: %s\n", prefix
, lvl_str
);
5489 if (c
->log_level_max
>= 0) {
5490 _cleanup_free_
char *t
= NULL
;
5492 (void) log_level_to_string_alloc(c
->log_level_max
, &t
);
5494 fprintf(f
, "%sLogLevelMax: %s\n", prefix
, strna(t
));
5497 if (c
->log_ratelimit_interval_usec
> 0) {
5498 char buf_timespan
[FORMAT_TIMESPAN_MAX
];
5501 "%sLogRateLimitIntervalSec: %s\n",
5502 prefix
, format_timespan(buf_timespan
, sizeof(buf_timespan
), c
->log_ratelimit_interval_usec
, USEC_PER_SEC
));
5505 if (c
->log_ratelimit_burst
> 0)
5506 fprintf(f
, "%sLogRateLimitBurst: %u\n", prefix
, c
->log_ratelimit_burst
);
5508 for (size_t j
= 0; j
< c
->n_log_extra_fields
; j
++) {
5509 fprintf(f
, "%sLogExtraFields: ", prefix
);
5510 fwrite(c
->log_extra_fields
[j
].iov_base
,
5511 1, c
->log_extra_fields
[j
].iov_len
,
5516 if (c
->log_namespace
)
5517 fprintf(f
, "%sLogNamespace: %s\n", prefix
, c
->log_namespace
);
5519 if (c
->secure_bits
) {
5520 _cleanup_free_
char *str
= NULL
;
5522 r
= secure_bits_to_string_alloc(c
->secure_bits
, &str
);
5524 fprintf(f
, "%sSecure Bits: %s\n", prefix
, str
);
5527 if (c
->capability_bounding_set
!= CAP_ALL
) {
5528 _cleanup_free_
char *str
= NULL
;
5530 r
= capability_set_to_string_alloc(c
->capability_bounding_set
, &str
);
5532 fprintf(f
, "%sCapabilityBoundingSet: %s\n", prefix
, str
);
5535 if (c
->capability_ambient_set
!= 0) {
5536 _cleanup_free_
char *str
= NULL
;
5538 r
= capability_set_to_string_alloc(c
->capability_ambient_set
, &str
);
5540 fprintf(f
, "%sAmbientCapabilities: %s\n", prefix
, str
);
5544 fprintf(f
, "%sUser: %s\n", prefix
, c
->user
);
5546 fprintf(f
, "%sGroup: %s\n", prefix
, c
->group
);
5548 fprintf(f
, "%sDynamicUser: %s\n", prefix
, yes_no(c
->dynamic_user
));
5550 strv_dump(f
, prefix
, "SupplementaryGroups", c
->supplementary_groups
);
5553 fprintf(f
, "%sPAMName: %s\n", prefix
, c
->pam_name
);
5555 strv_dump(f
, prefix
, "ReadWritePaths", c
->read_write_paths
);
5556 strv_dump(f
, prefix
, "ReadOnlyPaths", c
->read_only_paths
);
5557 strv_dump(f
, prefix
, "InaccessiblePaths", c
->inaccessible_paths
);
5558 strv_dump(f
, prefix
, "ExecPaths", c
->exec_paths
);
5559 strv_dump(f
, prefix
, "NoExecPaths", c
->no_exec_paths
);
5561 for (size_t i
= 0; i
< c
->n_bind_mounts
; i
++)
5562 fprintf(f
, "%s%s: %s%s:%s:%s\n", prefix
,
5563 c
->bind_mounts
[i
].read_only
? "BindReadOnlyPaths" : "BindPaths",
5564 c
->bind_mounts
[i
].ignore_enoent
? "-": "",
5565 c
->bind_mounts
[i
].source
,
5566 c
->bind_mounts
[i
].destination
,
5567 c
->bind_mounts
[i
].recursive
? "rbind" : "norbind");
5569 for (size_t i
= 0; i
< c
->n_temporary_filesystems
; i
++) {
5570 const TemporaryFileSystem
*t
= c
->temporary_filesystems
+ i
;
5572 fprintf(f
, "%sTemporaryFileSystem: %s%s%s\n", prefix
,
5574 isempty(t
->options
) ? "" : ":",
5575 strempty(t
->options
));
5580 "%sUtmpIdentifier: %s\n",
5581 prefix
, c
->utmp_id
);
5583 if (c
->selinux_context
)
5585 "%sSELinuxContext: %s%s\n",
5586 prefix
, c
->selinux_context_ignore
? "-" : "", c
->selinux_context
);
5588 if (c
->apparmor_profile
)
5590 "%sAppArmorProfile: %s%s\n",
5591 prefix
, c
->apparmor_profile_ignore
? "-" : "", c
->apparmor_profile
);
5593 if (c
->smack_process_label
)
5595 "%sSmackProcessLabel: %s%s\n",
5596 prefix
, c
->smack_process_label_ignore
? "-" : "", c
->smack_process_label
);
5598 if (c
->personality
!= PERSONALITY_INVALID
)
5600 "%sPersonality: %s\n",
5601 prefix
, strna(personality_to_string(c
->personality
)));
5604 "%sLockPersonality: %s\n",
5605 prefix
, yes_no(c
->lock_personality
));
5607 if (c
->syscall_filter
) {
5614 "%sSystemCallFilter: ",
5617 if (!c
->syscall_allow_list
)
5621 HASHMAP_FOREACH_KEY(val
, id
, c
->syscall_filter
) {
5622 _cleanup_free_
char *name
= NULL
;
5623 const char *errno_name
= NULL
;
5624 int num
= PTR_TO_INT(val
);
5631 name
= seccomp_syscall_resolve_num_arch(SCMP_ARCH_NATIVE
, PTR_TO_INT(id
) - 1);
5632 fputs(strna(name
), f
);
5635 errno_name
= seccomp_errno_or_action_to_string(num
);
5637 fprintf(f
, ":%s", errno_name
);
5639 fprintf(f
, ":%d", num
);
5647 if (c
->syscall_archs
) {
5653 "%sSystemCallArchitectures:",
5657 SET_FOREACH(id
, c
->syscall_archs
)
5658 fprintf(f
, " %s", strna(seccomp_arch_to_string(PTR_TO_UINT32(id
) - 1)));
5663 if (exec_context_restrict_namespaces_set(c
)) {
5664 _cleanup_free_
char *s
= NULL
;
5666 r
= namespace_flags_to_string(c
->restrict_namespaces
, &s
);
5668 fprintf(f
, "%sRestrictNamespaces: %s\n",
5672 if (c
->network_namespace_path
)
5674 "%sNetworkNamespacePath: %s\n",
5675 prefix
, c
->network_namespace_path
);
5677 if (c
->syscall_errno
> 0) {
5679 const char *errno_name
;
5682 fprintf(f
, "%sSystemCallErrorNumber: ", prefix
);
5685 errno_name
= seccomp_errno_or_action_to_string(c
->syscall_errno
);
5687 fputs(errno_name
, f
);
5689 fprintf(f
, "%d", c
->syscall_errno
);
5694 for (size_t i
= 0; i
< c
->n_mount_images
; i
++) {
5697 fprintf(f
, "%sMountImages: %s%s:%s", prefix
,
5698 c
->mount_images
[i
].ignore_enoent
? "-": "",
5699 c
->mount_images
[i
].source
,
5700 c
->mount_images
[i
].destination
);
5701 LIST_FOREACH(mount_options
, o
, c
->mount_images
[i
].mount_options
)
5702 fprintf(f
, ":%s:%s",
5703 partition_designator_to_string(o
->partition_designator
),
5704 strempty(o
->options
));
5708 for (size_t i
= 0; i
< c
->n_extension_images
; i
++) {
5711 fprintf(f
, "%sExtensionImages: %s%s", prefix
,
5712 c
->extension_images
[i
].ignore_enoent
? "-": "",
5713 c
->extension_images
[i
].source
);
5714 LIST_FOREACH(mount_options
, o
, c
->extension_images
[i
].mount_options
)
5715 fprintf(f
, ":%s:%s",
5716 partition_designator_to_string(o
->partition_designator
),
5717 strempty(o
->options
));
5722 bool exec_context_maintains_privileges(const ExecContext
*c
) {
5725 /* Returns true if the process forked off would run under
5726 * an unchanged UID or as root. */
5731 if (streq(c
->user
, "root") || streq(c
->user
, "0"))
5737 int exec_context_get_effective_ioprio(const ExecContext
*c
) {
5745 p
= ioprio_get(IOPRIO_WHO_PROCESS
, 0);
5747 return IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE
, 4);
5752 bool exec_context_get_effective_mount_apivfs(const ExecContext
*c
) {
5755 /* Explicit setting wins */
5756 if (c
->mount_apivfs_set
)
5757 return c
->mount_apivfs
;
5759 /* Default to "yes" if root directory or image are specified */
5760 if (exec_context_with_rootfs(c
))
5766 void exec_context_free_log_extra_fields(ExecContext
*c
) {
5769 for (size_t l
= 0; l
< c
->n_log_extra_fields
; l
++)
5770 free(c
->log_extra_fields
[l
].iov_base
);
5771 c
->log_extra_fields
= mfree(c
->log_extra_fields
);
5772 c
->n_log_extra_fields
= 0;
5775 void exec_context_revert_tty(ExecContext
*c
) {
5776 _cleanup_close_
int fd
= -1;
5783 /* First, reset the TTY (possibly kicking everybody else from the TTY) */
5784 exec_context_tty_reset(c
, NULL
);
5786 /* And then undo what chown_terminal() did earlier. Note that we only do this if we have a path
5787 * configured. If the TTY was passed to us as file descriptor we assume the TTY is opened and managed
5788 * by whoever passed it to us and thus knows better when and how to chmod()/chown() it back. */
5789 if (!exec_context_may_touch_tty(c
))
5792 path
= exec_context_tty_path(c
);
5796 fd
= open(path
, O_PATH
|O_CLOEXEC
);
5798 return (void) log_full_errno(errno
== ENOENT
? LOG_DEBUG
: LOG_WARNING
, errno
,
5799 "Failed to open TTY inode of '%s' to adjust ownership/access mode, ignoring: %m",
5802 if (fstat(fd
, &st
) < 0)
5803 return (void) log_warning_errno(errno
, "Failed to stat TTY '%s', ignoring: %m", path
);
5805 /* Let's add a superficial check that we only do this for stuff that looks like a TTY. We only check
5806 * if things are a character device, since a proper check either means we'd have to open the TTY and
5807 * use isatty(), but we'd rather not do that since opening TTYs comes with all kinds of side-effects
5808 * and is slow. Or we'd have to hardcode dev_t major information, which we'd rather avoid. Why bother
5809 * with this at all? → https://github.com/systemd/systemd/issues/19213 */
5810 if (!S_ISCHR(st
.st_mode
))
5811 return log_warning("Configured TTY '%s' is not actually a character device, ignoring.", path
);
5813 r
= fchmod_and_chown(fd
, TTY_MODE
, 0, TTY_GID
);
5815 log_warning_errno(r
, "Failed to reset TTY ownership/access mode of %s, ignoring: %m", path
);
5818 int exec_context_get_clean_directories(
5824 _cleanup_strv_free_
char **l
= NULL
;
5831 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
5834 if (!FLAGS_SET(mask
, 1U << t
))
5840 STRV_FOREACH(i
, c
->directories
[t
].paths
) {
5843 j
= path_join(prefix
[t
], *i
);
5847 r
= strv_consume(&l
, j
);
5851 /* Also remove private directories unconditionally. */
5852 if (t
!= EXEC_DIRECTORY_CONFIGURATION
) {
5853 j
= path_join(prefix
[t
], "private", *i
);
5857 r
= strv_consume(&l
, j
);
5868 int exec_context_get_clean_mask(ExecContext
*c
, ExecCleanMask
*ret
) {
5869 ExecCleanMask mask
= 0;
5874 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++)
5875 if (!strv_isempty(c
->directories
[t
].paths
))
5882 void exec_status_start(ExecStatus
*s
, pid_t pid
) {
5889 dual_timestamp_get(&s
->start_timestamp
);
5892 void exec_status_exit(ExecStatus
*s
, const ExecContext
*context
, pid_t pid
, int code
, int status
) {
5900 dual_timestamp_get(&s
->exit_timestamp
);
5905 if (context
&& context
->utmp_id
)
5906 (void) utmp_put_dead_process(context
->utmp_id
, pid
, code
, status
);
5909 void exec_status_reset(ExecStatus
*s
) {
5912 *s
= (ExecStatus
) {};
5915 void exec_status_dump(const ExecStatus
*s
, FILE *f
, const char *prefix
) {
5916 char buf
[FORMAT_TIMESTAMP_MAX
];
5924 prefix
= strempty(prefix
);
5927 "%sPID: "PID_FMT
"\n",
5930 if (dual_timestamp_is_set(&s
->start_timestamp
))
5932 "%sStart Timestamp: %s\n",
5933 prefix
, format_timestamp(buf
, sizeof(buf
), s
->start_timestamp
.realtime
));
5935 if (dual_timestamp_is_set(&s
->exit_timestamp
))
5937 "%sExit Timestamp: %s\n"
5939 "%sExit Status: %i\n",
5940 prefix
, format_timestamp(buf
, sizeof(buf
), s
->exit_timestamp
.realtime
),
5941 prefix
, sigchld_code_to_string(s
->code
),
5945 static char *exec_command_line(char **argv
) {
5953 STRV_FOREACH(a
, argv
)
5961 STRV_FOREACH(a
, argv
) {
5968 if (strpbrk(*a
, WHITESPACE
)) {
5979 /* FIXME: this doesn't really handle arguments that have
5980 * spaces and ticks in them */
5985 static void exec_command_dump(ExecCommand
*c
, FILE *f
, const char *prefix
) {
5986 _cleanup_free_
char *cmd
= NULL
;
5987 const char *prefix2
;
5992 prefix
= strempty(prefix
);
5993 prefix2
= strjoina(prefix
, "\t");
5995 cmd
= exec_command_line(c
->argv
);
5997 "%sCommand Line: %s\n",
5998 prefix
, cmd
? cmd
: strerror_safe(ENOMEM
));
6000 exec_status_dump(&c
->exec_status
, f
, prefix2
);
6003 void exec_command_dump_list(ExecCommand
*c
, FILE *f
, const char *prefix
) {
6006 prefix
= strempty(prefix
);
6008 LIST_FOREACH(command
, c
, c
)
6009 exec_command_dump(c
, f
, prefix
);
6012 void exec_command_append_list(ExecCommand
**l
, ExecCommand
*e
) {
6019 /* It's kind of important, that we keep the order here */
6020 LIST_FIND_TAIL(command
, *l
, end
);
6021 LIST_INSERT_AFTER(command
, *l
, end
, e
);
6026 int exec_command_set(ExecCommand
*c
, const char *path
, ...) {
6034 l
= strv_new_ap(path
, ap
);
6046 free_and_replace(c
->path
, p
);
6048 return strv_free_and_replace(c
->argv
, l
);
6051 int exec_command_append(ExecCommand
*c
, const char *path
, ...) {
6052 _cleanup_strv_free_
char **l
= NULL
;
6060 l
= strv_new_ap(path
, ap
);
6066 r
= strv_extend_strv(&c
->argv
, l
, false);
6073 static void *remove_tmpdir_thread(void *p
) {
6074 _cleanup_free_
char *path
= p
;
6076 (void) rm_rf(path
, REMOVE_ROOT
|REMOVE_PHYSICAL
);
6080 static ExecRuntime
* exec_runtime_free(ExecRuntime
*rt
, bool destroy
) {
6087 (void) hashmap_remove(rt
->manager
->exec_runtime_by_id
, rt
->id
);
6089 /* When destroy is true, then rm_rf tmp_dir and var_tmp_dir. */
6091 if (destroy
&& rt
->tmp_dir
&& !streq(rt
->tmp_dir
, RUN_SYSTEMD_EMPTY
)) {
6092 log_debug("Spawning thread to nuke %s", rt
->tmp_dir
);
6094 r
= asynchronous_job(remove_tmpdir_thread
, rt
->tmp_dir
);
6096 log_warning_errno(r
, "Failed to nuke %s: %m", rt
->tmp_dir
);
6101 if (destroy
&& rt
->var_tmp_dir
&& !streq(rt
->var_tmp_dir
, RUN_SYSTEMD_EMPTY
)) {
6102 log_debug("Spawning thread to nuke %s", rt
->var_tmp_dir
);
6104 r
= asynchronous_job(remove_tmpdir_thread
, rt
->var_tmp_dir
);
6106 log_warning_errno(r
, "Failed to nuke %s: %m", rt
->var_tmp_dir
);
6108 rt
->var_tmp_dir
= NULL
;
6111 rt
->id
= mfree(rt
->id
);
6112 rt
->tmp_dir
= mfree(rt
->tmp_dir
);
6113 rt
->var_tmp_dir
= mfree(rt
->var_tmp_dir
);
6114 safe_close_pair(rt
->netns_storage_socket
);
6115 safe_close_pair(rt
->ipcns_storage_socket
);
6119 static void exec_runtime_freep(ExecRuntime
**rt
) {
6120 (void) exec_runtime_free(*rt
, false);
6123 static int exec_runtime_allocate(ExecRuntime
**ret
, const char *id
) {
6124 _cleanup_free_
char *id_copy
= NULL
;
6129 id_copy
= strdup(id
);
6133 n
= new(ExecRuntime
, 1);
6137 *n
= (ExecRuntime
) {
6138 .id
= TAKE_PTR(id_copy
),
6139 .netns_storage_socket
= { -1, -1 },
6140 .ipcns_storage_socket
= { -1, -1 },
6147 static int exec_runtime_add(
6152 int netns_storage_socket
[2],
6153 int ipcns_storage_socket
[2],
6154 ExecRuntime
**ret
) {
6156 _cleanup_(exec_runtime_freep
) ExecRuntime
*rt
= NULL
;
6162 /* tmp_dir, var_tmp_dir, {net,ipc}ns_storage_socket fds are donated on success */
6164 r
= exec_runtime_allocate(&rt
, id
);
6168 r
= hashmap_ensure_put(&m
->exec_runtime_by_id
, &string_hash_ops
, rt
->id
, rt
);
6172 assert(!!rt
->tmp_dir
== !!rt
->var_tmp_dir
); /* We require both to be set together */
6173 rt
->tmp_dir
= TAKE_PTR(*tmp_dir
);
6174 rt
->var_tmp_dir
= TAKE_PTR(*var_tmp_dir
);
6176 if (netns_storage_socket
) {
6177 rt
->netns_storage_socket
[0] = TAKE_FD(netns_storage_socket
[0]);
6178 rt
->netns_storage_socket
[1] = TAKE_FD(netns_storage_socket
[1]);
6181 if (ipcns_storage_socket
) {
6182 rt
->ipcns_storage_socket
[0] = TAKE_FD(ipcns_storage_socket
[0]);
6183 rt
->ipcns_storage_socket
[1] = TAKE_FD(ipcns_storage_socket
[1]);
6190 /* do not remove created ExecRuntime object when the operation succeeds. */
6195 static int exec_runtime_make(
6197 const ExecContext
*c
,
6199 ExecRuntime
**ret
) {
6201 _cleanup_(namespace_cleanup_tmpdirp
) char *tmp_dir
= NULL
, *var_tmp_dir
= NULL
;
6202 _cleanup_close_pair_
int netns_storage_socket
[2] = { -1, -1 }, ipcns_storage_socket
[2] = { -1, -1 };
6209 /* It is not necessary to create ExecRuntime object. */
6210 if (!c
->private_network
&& !c
->private_ipc
&& !c
->private_tmp
&& !c
->network_namespace_path
) {
6215 if (c
->private_tmp
&&
6216 !(prefixed_path_strv_contains(c
->inaccessible_paths
, "/tmp") &&
6217 (prefixed_path_strv_contains(c
->inaccessible_paths
, "/var/tmp") ||
6218 prefixed_path_strv_contains(c
->inaccessible_paths
, "/var")))) {
6219 r
= setup_tmp_dirs(id
, &tmp_dir
, &var_tmp_dir
);
6224 if (c
->private_network
|| c
->network_namespace_path
) {
6225 if (socketpair(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
, 0, netns_storage_socket
) < 0)
6229 if (c
->private_ipc
|| c
->ipc_namespace_path
) {
6230 if (socketpair(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
, 0, ipcns_storage_socket
) < 0)
6234 r
= exec_runtime_add(m
, id
, &tmp_dir
, &var_tmp_dir
, netns_storage_socket
, ipcns_storage_socket
, ret
);
6241 int exec_runtime_acquire(Manager
*m
, const ExecContext
*c
, const char *id
, bool create
, ExecRuntime
**ret
) {
6249 rt
= hashmap_get(m
->exec_runtime_by_id
, id
);
6251 /* We already have a ExecRuntime object, let's increase the ref count and reuse it */
6259 /* If not found, then create a new object. */
6260 r
= exec_runtime_make(m
, c
, id
, &rt
);
6264 /* When r == 0, it is not necessary to create ExecRuntime object. */
6270 /* increment reference counter. */
6276 ExecRuntime
*exec_runtime_unref(ExecRuntime
*rt
, bool destroy
) {
6280 assert(rt
->n_ref
> 0);
6286 return exec_runtime_free(rt
, destroy
);
6289 int exec_runtime_serialize(const Manager
*m
, FILE *f
, FDSet
*fds
) {
6296 HASHMAP_FOREACH(rt
, m
->exec_runtime_by_id
) {
6297 fprintf(f
, "exec-runtime=%s", rt
->id
);
6300 fprintf(f
, " tmp-dir=%s", rt
->tmp_dir
);
6302 if (rt
->var_tmp_dir
)
6303 fprintf(f
, " var-tmp-dir=%s", rt
->var_tmp_dir
);
6305 if (rt
->netns_storage_socket
[0] >= 0) {
6308 copy
= fdset_put_dup(fds
, rt
->netns_storage_socket
[0]);
6312 fprintf(f
, " netns-socket-0=%i", copy
);
6315 if (rt
->netns_storage_socket
[1] >= 0) {
6318 copy
= fdset_put_dup(fds
, rt
->netns_storage_socket
[1]);
6322 fprintf(f
, " netns-socket-1=%i", copy
);
6325 if (rt
->ipcns_storage_socket
[0] >= 0) {
6328 copy
= fdset_put_dup(fds
, rt
->ipcns_storage_socket
[0]);
6332 fprintf(f
, " ipcns-socket-0=%i", copy
);
6335 if (rt
->ipcns_storage_socket
[1] >= 0) {
6338 copy
= fdset_put_dup(fds
, rt
->ipcns_storage_socket
[1]);
6342 fprintf(f
, " ipcns-socket-1=%i", copy
);
6351 int exec_runtime_deserialize_compat(Unit
*u
, const char *key
, const char *value
, FDSet
*fds
) {
6352 _cleanup_(exec_runtime_freep
) ExecRuntime
*rt_create
= NULL
;
6356 /* This is for the migration from old (v237 or earlier) deserialization text.
6357 * Due to the bug #7790, this may not work with the units that use JoinsNamespaceOf=.
6358 * Even if the ExecRuntime object originally created by the other unit, we cannot judge
6359 * so or not from the serialized text, then we always creates a new object owned by this. */
6365 /* Manager manages ExecRuntime objects by the unit id.
6366 * So, we omit the serialized text when the unit does not have id (yet?)... */
6367 if (isempty(u
->id
)) {
6368 log_unit_debug(u
, "Invocation ID not found. Dropping runtime parameter.");
6372 if (hashmap_ensure_allocated(&u
->manager
->exec_runtime_by_id
, &string_hash_ops
) < 0)
6375 rt
= hashmap_get(u
->manager
->exec_runtime_by_id
, u
->id
);
6377 if (exec_runtime_allocate(&rt_create
, u
->id
) < 0)
6383 if (streq(key
, "tmp-dir")) {
6384 if (free_and_strdup_warn(&rt
->tmp_dir
, value
) < 0)
6387 } else if (streq(key
, "var-tmp-dir")) {
6388 if (free_and_strdup_warn(&rt
->var_tmp_dir
, value
) < 0)
6391 } else if (streq(key
, "netns-socket-0")) {
6394 if (safe_atoi(value
, &fd
) < 0 || !fdset_contains(fds
, fd
)) {
6395 log_unit_debug(u
, "Failed to parse netns socket value: %s", value
);
6399 safe_close(rt
->netns_storage_socket
[0]);
6400 rt
->netns_storage_socket
[0] = fdset_remove(fds
, fd
);
6402 } else if (streq(key
, "netns-socket-1")) {
6405 if (safe_atoi(value
, &fd
) < 0 || !fdset_contains(fds
, fd
)) {
6406 log_unit_debug(u
, "Failed to parse netns socket value: %s", value
);
6410 safe_close(rt
->netns_storage_socket
[1]);
6411 rt
->netns_storage_socket
[1] = fdset_remove(fds
, fd
);
6416 /* If the object is newly created, then put it to the hashmap which manages ExecRuntime objects. */
6418 r
= hashmap_put(u
->manager
->exec_runtime_by_id
, rt_create
->id
, rt_create
);
6420 log_unit_debug_errno(u
, r
, "Failed to put runtime parameter to manager's storage: %m");
6424 rt_create
->manager
= u
->manager
;
6427 TAKE_PTR(rt_create
);
6433 int exec_runtime_deserialize_one(Manager
*m
, const char *value
, FDSet
*fds
) {
6434 _cleanup_free_
char *tmp_dir
= NULL
, *var_tmp_dir
= NULL
;
6436 int r
, netns_fdpair
[] = {-1, -1}, ipcns_fdpair
[] = {-1, -1};
6437 const char *p
, *v
= value
;
6444 n
= strcspn(v
, " ");
6445 id
= strndupa(v
, n
);
6450 v
= startswith(p
, "tmp-dir=");
6452 n
= strcspn(v
, " ");
6453 tmp_dir
= strndup(v
, n
);
6461 v
= startswith(p
, "var-tmp-dir=");
6463 n
= strcspn(v
, " ");
6464 var_tmp_dir
= strndup(v
, n
);
6472 v
= startswith(p
, "netns-socket-0=");
6476 n
= strcspn(v
, " ");
6477 buf
= strndupa(v
, n
);
6479 r
= safe_atoi(buf
, &netns_fdpair
[0]);
6481 return log_debug_errno(r
, "Unable to parse exec-runtime specification netns-socket-0=%s: %m", buf
);
6482 if (!fdset_contains(fds
, netns_fdpair
[0]))
6483 return log_debug_errno(SYNTHETIC_ERRNO(EBADF
),
6484 "exec-runtime specification netns-socket-0= refers to unknown fd %d: %m", netns_fdpair
[0]);
6485 netns_fdpair
[0] = fdset_remove(fds
, netns_fdpair
[0]);
6491 v
= startswith(p
, "netns-socket-1=");
6495 n
= strcspn(v
, " ");
6496 buf
= strndupa(v
, n
);
6498 r
= safe_atoi(buf
, &netns_fdpair
[1]);
6500 return log_debug_errno(r
, "Unable to parse exec-runtime specification netns-socket-1=%s: %m", buf
);
6501 if (!fdset_contains(fds
, netns_fdpair
[1]))
6502 return log_debug_errno(SYNTHETIC_ERRNO(EBADF
),
6503 "exec-runtime specification netns-socket-1= refers to unknown fd %d: %m", netns_fdpair
[1]);
6504 netns_fdpair
[1] = fdset_remove(fds
, netns_fdpair
[1]);
6510 v
= startswith(p
, "ipcns-socket-0=");
6514 n
= strcspn(v
, " ");
6515 buf
= strndupa(v
, n
);
6517 r
= safe_atoi(buf
, &ipcns_fdpair
[0]);
6519 return log_debug_errno(r
, "Unable to parse exec-runtime specification ipcns-socket-0=%s: %m", buf
);
6520 if (!fdset_contains(fds
, ipcns_fdpair
[0]))
6521 return log_debug_errno(SYNTHETIC_ERRNO(EBADF
),
6522 "exec-runtime specification ipcns-socket-0= refers to unknown fd %d: %m", ipcns_fdpair
[0]);
6523 ipcns_fdpair
[0] = fdset_remove(fds
, ipcns_fdpair
[0]);
6529 v
= startswith(p
, "ipcns-socket-1=");
6533 n
= strcspn(v
, " ");
6534 buf
= strndupa(v
, n
);
6536 r
= safe_atoi(buf
, &ipcns_fdpair
[1]);
6538 return log_debug_errno(r
, "Unable to parse exec-runtime specification ipcns-socket-1=%s: %m", buf
);
6539 if (!fdset_contains(fds
, ipcns_fdpair
[1]))
6540 return log_debug_errno(SYNTHETIC_ERRNO(EBADF
),
6541 "exec-runtime specification ipcns-socket-1= refers to unknown fd %d: %m", ipcns_fdpair
[1]);
6542 ipcns_fdpair
[1] = fdset_remove(fds
, ipcns_fdpair
[1]);
6546 r
= exec_runtime_add(m
, id
, &tmp_dir
, &var_tmp_dir
, netns_fdpair
, ipcns_fdpair
, NULL
);
6548 return log_debug_errno(r
, "Failed to add exec-runtime: %m");
6552 void exec_runtime_vacuum(Manager
*m
) {
6557 /* Free unreferenced ExecRuntime objects. This is used after manager deserialization process. */
6559 HASHMAP_FOREACH(rt
, m
->exec_runtime_by_id
) {
6563 (void) exec_runtime_free(rt
, false);
6567 void exec_params_clear(ExecParameters
*p
) {
6571 p
->environment
= strv_free(p
->environment
);
6572 p
->fd_names
= strv_free(p
->fd_names
);
6573 p
->fds
= mfree(p
->fds
);
6574 p
->exec_fd
= safe_close(p
->exec_fd
);
6577 ExecSetCredential
*exec_set_credential_free(ExecSetCredential
*sc
) {
6586 DEFINE_HASH_OPS_WITH_VALUE_DESTRUCTOR(exec_set_credential_hash_ops
, char, string_hash_func
, string_compare_func
, ExecSetCredential
, exec_set_credential_free
);
6588 static const char* const exec_input_table
[_EXEC_INPUT_MAX
] = {
6589 [EXEC_INPUT_NULL
] = "null",
6590 [EXEC_INPUT_TTY
] = "tty",
6591 [EXEC_INPUT_TTY_FORCE
] = "tty-force",
6592 [EXEC_INPUT_TTY_FAIL
] = "tty-fail",
6593 [EXEC_INPUT_SOCKET
] = "socket",
6594 [EXEC_INPUT_NAMED_FD
] = "fd",
6595 [EXEC_INPUT_DATA
] = "data",
6596 [EXEC_INPUT_FILE
] = "file",
6599 DEFINE_STRING_TABLE_LOOKUP(exec_input
, ExecInput
);
6601 static const char* const exec_output_table
[_EXEC_OUTPUT_MAX
] = {
6602 [EXEC_OUTPUT_INHERIT
] = "inherit",
6603 [EXEC_OUTPUT_NULL
] = "null",
6604 [EXEC_OUTPUT_TTY
] = "tty",
6605 [EXEC_OUTPUT_KMSG
] = "kmsg",
6606 [EXEC_OUTPUT_KMSG_AND_CONSOLE
] = "kmsg+console",
6607 [EXEC_OUTPUT_JOURNAL
] = "journal",
6608 [EXEC_OUTPUT_JOURNAL_AND_CONSOLE
] = "journal+console",
6609 [EXEC_OUTPUT_SOCKET
] = "socket",
6610 [EXEC_OUTPUT_NAMED_FD
] = "fd",
6611 [EXEC_OUTPUT_FILE
] = "file",
6612 [EXEC_OUTPUT_FILE_APPEND
] = "append",
6613 [EXEC_OUTPUT_FILE_TRUNCATE
] = "truncate",
6616 DEFINE_STRING_TABLE_LOOKUP(exec_output
, ExecOutput
);
6618 static const char* const exec_utmp_mode_table
[_EXEC_UTMP_MODE_MAX
] = {
6619 [EXEC_UTMP_INIT
] = "init",
6620 [EXEC_UTMP_LOGIN
] = "login",
6621 [EXEC_UTMP_USER
] = "user",
6624 DEFINE_STRING_TABLE_LOOKUP(exec_utmp_mode
, ExecUtmpMode
);
6626 static const char* const exec_preserve_mode_table
[_EXEC_PRESERVE_MODE_MAX
] = {
6627 [EXEC_PRESERVE_NO
] = "no",
6628 [EXEC_PRESERVE_YES
] = "yes",
6629 [EXEC_PRESERVE_RESTART
] = "restart",
6632 DEFINE_STRING_TABLE_LOOKUP_WITH_BOOLEAN(exec_preserve_mode
, ExecPreserveMode
, EXEC_PRESERVE_YES
);
6634 /* This table maps ExecDirectoryType to the setting it is configured with in the unit */
6635 static const char* const exec_directory_type_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
6636 [EXEC_DIRECTORY_RUNTIME
] = "RuntimeDirectory",
6637 [EXEC_DIRECTORY_STATE
] = "StateDirectory",
6638 [EXEC_DIRECTORY_CACHE
] = "CacheDirectory",
6639 [EXEC_DIRECTORY_LOGS
] = "LogsDirectory",
6640 [EXEC_DIRECTORY_CONFIGURATION
] = "ConfigurationDirectory",
6643 DEFINE_STRING_TABLE_LOOKUP(exec_directory_type
, ExecDirectoryType
);
6645 /* And this table maps ExecDirectoryType too, but to a generic term identifying the type of resource. This
6646 * one is supposed to be generic enough to be used for unit types that don't use ExecContext and per-unit
6647 * directories, specifically .timer units with their timestamp touch file. */
6648 static const char* const exec_resource_type_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
6649 [EXEC_DIRECTORY_RUNTIME
] = "runtime",
6650 [EXEC_DIRECTORY_STATE
] = "state",
6651 [EXEC_DIRECTORY_CACHE
] = "cache",
6652 [EXEC_DIRECTORY_LOGS
] = "logs",
6653 [EXEC_DIRECTORY_CONFIGURATION
] = "configuration",
6656 DEFINE_STRING_TABLE_LOOKUP(exec_resource_type
, ExecDirectoryType
);
6658 /* And this table also maps ExecDirectoryType, to the environment variable we pass the selected directory to
6659 * the service payload in. */
6660 static const char* const exec_directory_env_name_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
6661 [EXEC_DIRECTORY_RUNTIME
] = "RUNTIME_DIRECTORY",
6662 [EXEC_DIRECTORY_STATE
] = "STATE_DIRECTORY",
6663 [EXEC_DIRECTORY_CACHE
] = "CACHE_DIRECTORY",
6664 [EXEC_DIRECTORY_LOGS
] = "LOGS_DIRECTORY",
6665 [EXEC_DIRECTORY_CONFIGURATION
] = "CONFIGURATION_DIRECTORY",
6668 DEFINE_PRIVATE_STRING_TABLE_LOOKUP_TO_STRING(exec_directory_env_name
, ExecDirectoryType
);
6670 static const char* const exec_keyring_mode_table
[_EXEC_KEYRING_MODE_MAX
] = {
6671 [EXEC_KEYRING_INHERIT
] = "inherit",
6672 [EXEC_KEYRING_PRIVATE
] = "private",
6673 [EXEC_KEYRING_SHARED
] = "shared",
6676 DEFINE_STRING_TABLE_LOOKUP(exec_keyring_mode
, ExecKeyringMode
);