1 /* SPDX-License-Identifier: LGPL-2.1-or-later */
6 #include <sys/eventfd.h>
10 #include <sys/personality.h>
11 #include <sys/prctl.h>
13 #include <sys/types.h>
19 #include <security/pam_appl.h>
23 #include <selinux/selinux.h>
31 #include <sys/apparmor.h>
34 #include "sd-messages.h"
38 #include "alloc-util.h"
40 #include "apparmor-util.h"
45 #include "capability-util.h"
46 #include "cgroup-setup.h"
47 #include "chown-recursive.h"
48 #include "cpu-set-util.h"
49 #include "data-fd-util.h"
53 #include "errno-list.h"
55 #include "exit-status.h"
58 #include "format-util.h"
60 #include "glob-util.h"
61 #include "hexdecoct.h"
68 #include "manager-dump.h"
69 #include "memory-util.h"
70 #include "missing_fs.h"
72 #include "mount-util.h"
73 #include "mountpoint-util.h"
74 #include "namespace.h"
75 #include "parse-util.h"
76 #include "path-util.h"
77 #include "process-util.h"
78 #include "random-util.h"
79 #include "rlimit-util.h"
82 #include "seccomp-util.h"
84 #include "securebits-util.h"
85 #include "selinux-util.h"
86 #include "signal-util.h"
87 #include "smack-util.h"
88 #include "socket-util.h"
90 #include "stat-util.h"
91 #include "string-table.h"
92 #include "string-util.h"
94 #include "syslog-util.h"
95 #include "terminal-util.h"
96 #include "tmpfile-util.h"
97 #include "umask-util.h"
98 #include "unit-serialize.h"
99 #include "user-util.h"
100 #include "utmp-wtmp.h"
102 #define IDLE_TIMEOUT_USEC (5*USEC_PER_SEC)
103 #define IDLE_TIMEOUT2_USEC (1*USEC_PER_SEC)
105 #define SNDBUF_SIZE (8*1024*1024)
107 static int shift_fds(int fds
[], size_t n_fds
) {
111 /* Modifies the fds array! (sorts it) */
115 for (int start
= 0;;) {
116 int restart_from
= -1;
118 for (int i
= start
; i
< (int) n_fds
; i
++) {
121 /* Already at right index? */
125 nfd
= fcntl(fds
[i
], F_DUPFD
, i
+ 3);
132 /* Hmm, the fd we wanted isn't free? Then
133 * let's remember that and try again from here */
134 if (nfd
!= i
+3 && restart_from
< 0)
138 if (restart_from
< 0)
141 start
= restart_from
;
147 static int flags_fds(const int fds
[], size_t n_socket_fds
, size_t n_storage_fds
, bool nonblock
) {
151 n_fds
= n_socket_fds
+ n_storage_fds
;
157 /* Drops/Sets O_NONBLOCK and FD_CLOEXEC from the file flags.
158 * O_NONBLOCK only applies to socket activation though. */
160 for (size_t i
= 0; i
< n_fds
; i
++) {
162 if (i
< n_socket_fds
) {
163 r
= fd_nonblock(fds
[i
], nonblock
);
168 /* We unconditionally drop FD_CLOEXEC from the fds,
169 * since after all we want to pass these fds to our
172 r
= fd_cloexec(fds
[i
], false);
180 static const char *exec_context_tty_path(const ExecContext
*context
) {
183 if (context
->stdio_as_fds
)
186 if (context
->tty_path
)
187 return context
->tty_path
;
189 return "/dev/console";
192 static void exec_context_tty_reset(const ExecContext
*context
, const ExecParameters
*p
) {
197 path
= exec_context_tty_path(context
);
199 if (context
->tty_vhangup
) {
200 if (p
&& p
->stdin_fd
>= 0)
201 (void) terminal_vhangup_fd(p
->stdin_fd
);
203 (void) terminal_vhangup(path
);
206 if (context
->tty_reset
) {
207 if (p
&& p
->stdin_fd
>= 0)
208 (void) reset_terminal_fd(p
->stdin_fd
, true);
210 (void) reset_terminal(path
);
213 if (context
->tty_vt_disallocate
&& path
)
214 (void) vt_disallocate(path
);
217 static bool is_terminal_input(ExecInput i
) {
220 EXEC_INPUT_TTY_FORCE
,
221 EXEC_INPUT_TTY_FAIL
);
224 static bool is_terminal_output(ExecOutput o
) {
227 EXEC_OUTPUT_KMSG_AND_CONSOLE
,
228 EXEC_OUTPUT_JOURNAL_AND_CONSOLE
);
231 static bool is_kmsg_output(ExecOutput o
) {
234 EXEC_OUTPUT_KMSG_AND_CONSOLE
);
237 static bool exec_context_needs_term(const ExecContext
*c
) {
240 /* Return true if the execution context suggests we should set $TERM to something useful. */
242 if (is_terminal_input(c
->std_input
))
245 if (is_terminal_output(c
->std_output
))
248 if (is_terminal_output(c
->std_error
))
251 return !!c
->tty_path
;
254 static int open_null_as(int flags
, int nfd
) {
259 fd
= open("/dev/null", flags
|O_NOCTTY
);
263 return move_fd(fd
, nfd
, false);
266 static int connect_journal_socket(
268 const char *log_namespace
,
272 union sockaddr_union sa
;
274 uid_t olduid
= UID_INVALID
;
275 gid_t oldgid
= GID_INVALID
;
280 strjoina("/run/systemd/journal.", log_namespace
, "/stdout") :
281 "/run/systemd/journal/stdout";
282 r
= sockaddr_un_set_path(&sa
.un
, j
);
287 if (gid_is_valid(gid
)) {
290 if (setegid(gid
) < 0)
294 if (uid_is_valid(uid
)) {
297 if (seteuid(uid
) < 0) {
303 r
= connect(fd
, &sa
.sa
, sa_len
) < 0 ? -errno
: 0;
305 /* If we fail to restore the uid or gid, things will likely
306 fail later on. This should only happen if an LSM interferes. */
308 if (uid_is_valid(uid
))
309 (void) seteuid(olduid
);
312 if (gid_is_valid(gid
))
313 (void) setegid(oldgid
);
318 static int connect_logger_as(
320 const ExecContext
*context
,
321 const ExecParameters
*params
,
328 _cleanup_close_
int fd
= -1;
333 assert(output
< _EXEC_OUTPUT_MAX
);
337 fd
= socket(AF_UNIX
, SOCK_STREAM
, 0);
341 r
= connect_journal_socket(fd
, context
->log_namespace
, uid
, gid
);
345 if (shutdown(fd
, SHUT_RD
) < 0)
348 (void) fd_inc_sndbuf(fd
, SNDBUF_SIZE
);
358 context
->syslog_identifier
?: ident
,
359 params
->flags
& EXEC_PASS_LOG_UNIT
? unit
->id
: "",
360 context
->syslog_priority
,
361 !!context
->syslog_level_prefix
,
363 is_kmsg_output(output
),
364 is_terminal_output(output
)) < 0)
367 return move_fd(TAKE_FD(fd
), nfd
, false);
370 static int open_terminal_as(const char *path
, int flags
, int nfd
) {
376 fd
= open_terminal(path
, flags
| O_NOCTTY
);
380 return move_fd(fd
, nfd
, false);
383 static int acquire_path(const char *path
, int flags
, mode_t mode
) {
384 union sockaddr_union sa
;
386 _cleanup_close_
int fd
= -1;
391 if (IN_SET(flags
& O_ACCMODE
, O_WRONLY
, O_RDWR
))
394 fd
= open(path
, flags
|O_NOCTTY
, mode
);
398 if (errno
!= ENXIO
) /* ENXIO is returned when we try to open() an AF_UNIX file system socket on Linux */
401 /* So, it appears the specified path could be an AF_UNIX socket. Let's see if we can connect to it. */
403 r
= sockaddr_un_set_path(&sa
.un
, path
);
405 return r
== -EINVAL
? -ENXIO
: r
;
408 fd
= socket(AF_UNIX
, SOCK_STREAM
, 0);
412 if (connect(fd
, &sa
.sa
, sa_len
) < 0)
413 return errno
== EINVAL
? -ENXIO
: -errno
; /* Propagate initial error if we get EINVAL, i.e. we have
414 * indication that this wasn't an AF_UNIX socket after all */
416 if ((flags
& O_ACCMODE
) == O_RDONLY
)
417 r
= shutdown(fd
, SHUT_WR
);
418 else if ((flags
& O_ACCMODE
) == O_WRONLY
)
419 r
= shutdown(fd
, SHUT_RD
);
428 static int fixup_input(
429 const ExecContext
*context
,
431 bool apply_tty_stdin
) {
437 std_input
= context
->std_input
;
439 if (is_terminal_input(std_input
) && !apply_tty_stdin
)
440 return EXEC_INPUT_NULL
;
442 if (std_input
== EXEC_INPUT_SOCKET
&& socket_fd
< 0)
443 return EXEC_INPUT_NULL
;
445 if (std_input
== EXEC_INPUT_DATA
&& context
->stdin_data_size
== 0)
446 return EXEC_INPUT_NULL
;
451 static int fixup_output(ExecOutput output
, int socket_fd
) {
453 if (output
== EXEC_OUTPUT_SOCKET
&& socket_fd
< 0)
454 return EXEC_OUTPUT_INHERIT
;
459 static int setup_input(
460 const ExecContext
*context
,
461 const ExecParameters
*params
,
463 const int named_iofds
[static 3]) {
471 if (params
->stdin_fd
>= 0) {
472 if (dup2(params
->stdin_fd
, STDIN_FILENO
) < 0)
475 /* Try to make this the controlling tty, if it is a tty, and reset it */
476 if (isatty(STDIN_FILENO
)) {
477 (void) ioctl(STDIN_FILENO
, TIOCSCTTY
, context
->std_input
== EXEC_INPUT_TTY_FORCE
);
478 (void) reset_terminal_fd(STDIN_FILENO
, true);
484 i
= fixup_input(context
, socket_fd
, params
->flags
& EXEC_APPLY_TTY_STDIN
);
488 case EXEC_INPUT_NULL
:
489 return open_null_as(O_RDONLY
, STDIN_FILENO
);
492 case EXEC_INPUT_TTY_FORCE
:
493 case EXEC_INPUT_TTY_FAIL
: {
496 fd
= acquire_terminal(exec_context_tty_path(context
),
497 i
== EXEC_INPUT_TTY_FAIL
? ACQUIRE_TERMINAL_TRY
:
498 i
== EXEC_INPUT_TTY_FORCE
? ACQUIRE_TERMINAL_FORCE
:
499 ACQUIRE_TERMINAL_WAIT
,
504 return move_fd(fd
, STDIN_FILENO
, false);
507 case EXEC_INPUT_SOCKET
:
508 assert(socket_fd
>= 0);
510 return dup2(socket_fd
, STDIN_FILENO
) < 0 ? -errno
: STDIN_FILENO
;
512 case EXEC_INPUT_NAMED_FD
:
513 assert(named_iofds
[STDIN_FILENO
] >= 0);
515 (void) fd_nonblock(named_iofds
[STDIN_FILENO
], false);
516 return dup2(named_iofds
[STDIN_FILENO
], STDIN_FILENO
) < 0 ? -errno
: STDIN_FILENO
;
518 case EXEC_INPUT_DATA
: {
521 fd
= acquire_data_fd(context
->stdin_data
, context
->stdin_data_size
, 0);
525 return move_fd(fd
, STDIN_FILENO
, false);
528 case EXEC_INPUT_FILE
: {
532 assert(context
->stdio_file
[STDIN_FILENO
]);
534 rw
= (context
->std_output
== EXEC_OUTPUT_FILE
&& streq_ptr(context
->stdio_file
[STDIN_FILENO
], context
->stdio_file
[STDOUT_FILENO
])) ||
535 (context
->std_error
== EXEC_OUTPUT_FILE
&& streq_ptr(context
->stdio_file
[STDIN_FILENO
], context
->stdio_file
[STDERR_FILENO
]));
537 fd
= acquire_path(context
->stdio_file
[STDIN_FILENO
], rw
? O_RDWR
: O_RDONLY
, 0666 & ~context
->umask
);
541 return move_fd(fd
, STDIN_FILENO
, false);
545 assert_not_reached("Unknown input type");
549 static bool can_inherit_stderr_from_stdout(
550 const ExecContext
*context
,
556 /* Returns true, if given the specified STDERR and STDOUT output we can directly dup() the stdout fd to the
559 if (e
== EXEC_OUTPUT_INHERIT
)
564 if (e
== EXEC_OUTPUT_NAMED_FD
)
565 return streq_ptr(context
->stdio_fdname
[STDOUT_FILENO
], context
->stdio_fdname
[STDERR_FILENO
]);
567 if (IN_SET(e
, EXEC_OUTPUT_FILE
, EXEC_OUTPUT_FILE_APPEND
, EXEC_OUTPUT_FILE_TRUNCATE
))
568 return streq_ptr(context
->stdio_file
[STDOUT_FILENO
], context
->stdio_file
[STDERR_FILENO
]);
573 static int setup_output(
575 const ExecContext
*context
,
576 const ExecParameters
*params
,
579 const int named_iofds
[static 3],
583 dev_t
*journal_stream_dev
,
584 ino_t
*journal_stream_ino
) {
594 assert(journal_stream_dev
);
595 assert(journal_stream_ino
);
597 if (fileno
== STDOUT_FILENO
&& params
->stdout_fd
>= 0) {
599 if (dup2(params
->stdout_fd
, STDOUT_FILENO
) < 0)
602 return STDOUT_FILENO
;
605 if (fileno
== STDERR_FILENO
&& params
->stderr_fd
>= 0) {
606 if (dup2(params
->stderr_fd
, STDERR_FILENO
) < 0)
609 return STDERR_FILENO
;
612 i
= fixup_input(context
, socket_fd
, params
->flags
& EXEC_APPLY_TTY_STDIN
);
613 o
= fixup_output(context
->std_output
, socket_fd
);
615 if (fileno
== STDERR_FILENO
) {
617 e
= fixup_output(context
->std_error
, socket_fd
);
619 /* This expects the input and output are already set up */
621 /* Don't change the stderr file descriptor if we inherit all
622 * the way and are not on a tty */
623 if (e
== EXEC_OUTPUT_INHERIT
&&
624 o
== EXEC_OUTPUT_INHERIT
&&
625 i
== EXEC_INPUT_NULL
&&
626 !is_terminal_input(context
->std_input
) &&
630 /* Duplicate from stdout if possible */
631 if (can_inherit_stderr_from_stdout(context
, o
, e
))
632 return dup2(STDOUT_FILENO
, fileno
) < 0 ? -errno
: fileno
;
636 } else if (o
== EXEC_OUTPUT_INHERIT
) {
637 /* If input got downgraded, inherit the original value */
638 if (i
== EXEC_INPUT_NULL
&& is_terminal_input(context
->std_input
))
639 return open_terminal_as(exec_context_tty_path(context
), O_WRONLY
, fileno
);
641 /* If the input is connected to anything that's not a /dev/null or a data fd, inherit that... */
642 if (!IN_SET(i
, EXEC_INPUT_NULL
, EXEC_INPUT_DATA
))
643 return dup2(STDIN_FILENO
, fileno
) < 0 ? -errno
: fileno
;
645 /* If we are not started from PID 1 we just inherit STDOUT from our parent process. */
649 /* We need to open /dev/null here anew, to get the right access mode. */
650 return open_null_as(O_WRONLY
, fileno
);
655 case EXEC_OUTPUT_NULL
:
656 return open_null_as(O_WRONLY
, fileno
);
658 case EXEC_OUTPUT_TTY
:
659 if (is_terminal_input(i
))
660 return dup2(STDIN_FILENO
, fileno
) < 0 ? -errno
: fileno
;
662 /* We don't reset the terminal if this is just about output */
663 return open_terminal_as(exec_context_tty_path(context
), O_WRONLY
, fileno
);
665 case EXEC_OUTPUT_KMSG
:
666 case EXEC_OUTPUT_KMSG_AND_CONSOLE
:
667 case EXEC_OUTPUT_JOURNAL
:
668 case EXEC_OUTPUT_JOURNAL_AND_CONSOLE
:
669 r
= connect_logger_as(unit
, context
, params
, o
, ident
, fileno
, uid
, gid
);
671 log_unit_warning_errno(unit
, r
, "Failed to connect %s to the journal socket, ignoring: %m",
672 fileno
== STDOUT_FILENO
? "stdout" : "stderr");
673 r
= open_null_as(O_WRONLY
, fileno
);
677 /* If we connected this fd to the journal via a stream, patch the device/inode into the passed
678 * parameters, but only then. This is useful so that we can set $JOURNAL_STREAM that permits
679 * services to detect whether they are connected to the journal or not.
681 * If both stdout and stderr are connected to a stream then let's make sure to store the data
682 * about STDERR as that's usually the best way to do logging. */
684 if (fstat(fileno
, &st
) >= 0 &&
685 (*journal_stream_ino
== 0 || fileno
== STDERR_FILENO
)) {
686 *journal_stream_dev
= st
.st_dev
;
687 *journal_stream_ino
= st
.st_ino
;
692 case EXEC_OUTPUT_SOCKET
:
693 assert(socket_fd
>= 0);
695 return dup2(socket_fd
, fileno
) < 0 ? -errno
: fileno
;
697 case EXEC_OUTPUT_NAMED_FD
:
698 assert(named_iofds
[fileno
] >= 0);
700 (void) fd_nonblock(named_iofds
[fileno
], false);
701 return dup2(named_iofds
[fileno
], fileno
) < 0 ? -errno
: fileno
;
703 case EXEC_OUTPUT_FILE
:
704 case EXEC_OUTPUT_FILE_APPEND
:
705 case EXEC_OUTPUT_FILE_TRUNCATE
: {
709 assert(context
->stdio_file
[fileno
]);
711 rw
= context
->std_input
== EXEC_INPUT_FILE
&&
712 streq_ptr(context
->stdio_file
[fileno
], context
->stdio_file
[STDIN_FILENO
]);
715 return dup2(STDIN_FILENO
, fileno
) < 0 ? -errno
: fileno
;
718 if (o
== EXEC_OUTPUT_FILE_APPEND
)
720 else if (o
== EXEC_OUTPUT_FILE_TRUNCATE
)
723 fd
= acquire_path(context
->stdio_file
[fileno
], flags
, 0666 & ~context
->umask
);
727 return move_fd(fd
, fileno
, 0);
731 assert_not_reached("Unknown error type");
735 static int chown_terminal(int fd
, uid_t uid
) {
740 /* Before we chown/chmod the TTY, let's ensure this is actually a tty */
741 if (isatty(fd
) < 1) {
742 if (IN_SET(errno
, EINVAL
, ENOTTY
))
743 return 0; /* not a tty */
748 /* This might fail. What matters are the results. */
749 r
= fchmod_and_chown(fd
, TTY_MODE
, uid
, GID_INVALID
);
756 static int setup_confirm_stdio(const char *vc
, int *_saved_stdin
, int *_saved_stdout
) {
757 _cleanup_close_
int fd
= -1, saved_stdin
= -1, saved_stdout
= -1;
760 assert(_saved_stdin
);
761 assert(_saved_stdout
);
763 saved_stdin
= fcntl(STDIN_FILENO
, F_DUPFD
, 3);
767 saved_stdout
= fcntl(STDOUT_FILENO
, F_DUPFD
, 3);
768 if (saved_stdout
< 0)
771 fd
= acquire_terminal(vc
, ACQUIRE_TERMINAL_WAIT
, DEFAULT_CONFIRM_USEC
);
775 r
= chown_terminal(fd
, getuid());
779 r
= reset_terminal_fd(fd
, true);
783 r
= rearrange_stdio(fd
, fd
, STDERR_FILENO
);
788 *_saved_stdin
= saved_stdin
;
789 *_saved_stdout
= saved_stdout
;
791 saved_stdin
= saved_stdout
= -1;
796 static void write_confirm_error_fd(int err
, int fd
, const Unit
*u
) {
799 if (err
== -ETIMEDOUT
)
800 dprintf(fd
, "Confirmation question timed out for %s, assuming positive response.\n", u
->id
);
803 dprintf(fd
, "Couldn't ask confirmation for %s: %m, assuming positive response.\n", u
->id
);
807 static void write_confirm_error(int err
, const char *vc
, const Unit
*u
) {
808 _cleanup_close_
int fd
= -1;
812 fd
= open_terminal(vc
, O_WRONLY
|O_NOCTTY
|O_CLOEXEC
);
816 write_confirm_error_fd(err
, fd
, u
);
819 static int restore_confirm_stdio(int *saved_stdin
, int *saved_stdout
) {
823 assert(saved_stdout
);
827 if (*saved_stdin
>= 0)
828 if (dup2(*saved_stdin
, STDIN_FILENO
) < 0)
831 if (*saved_stdout
>= 0)
832 if (dup2(*saved_stdout
, STDOUT_FILENO
) < 0)
835 *saved_stdin
= safe_close(*saved_stdin
);
836 *saved_stdout
= safe_close(*saved_stdout
);
842 CONFIRM_PRETEND_FAILURE
= -1,
843 CONFIRM_PRETEND_SUCCESS
= 0,
847 static int ask_for_confirmation(const char *vc
, Unit
*u
, const char *cmdline
) {
848 int saved_stdout
= -1, saved_stdin
= -1, r
;
849 _cleanup_free_
char *e
= NULL
;
852 /* For any internal errors, assume a positive response. */
853 r
= setup_confirm_stdio(vc
, &saved_stdin
, &saved_stdout
);
855 write_confirm_error(r
, vc
, u
);
856 return CONFIRM_EXECUTE
;
859 /* confirm_spawn might have been disabled while we were sleeping. */
860 if (manager_is_confirm_spawn_disabled(u
->manager
)) {
865 e
= ellipsize(cmdline
, 60, 100);
873 r
= ask_char(&c
, "yfshiDjcn", "Execute %s? [y, f, s – h for help] ", e
);
875 write_confirm_error_fd(r
, STDOUT_FILENO
, u
);
882 printf("Resuming normal execution.\n");
883 manager_disable_confirm_spawn();
887 unit_dump(u
, stdout
, " ");
888 continue; /* ask again */
890 printf("Failing execution.\n");
891 r
= CONFIRM_PRETEND_FAILURE
;
894 printf(" c - continue, proceed without asking anymore\n"
895 " D - dump, show the state of the unit\n"
896 " f - fail, don't execute the command and pretend it failed\n"
898 " i - info, show a short summary of the unit\n"
899 " j - jobs, show jobs that are in progress\n"
900 " s - skip, don't execute the command and pretend it succeeded\n"
901 " y - yes, execute the command\n");
902 continue; /* ask again */
904 printf(" Description: %s\n"
907 u
->id
, u
->description
, cmdline
);
908 continue; /* ask again */
910 manager_dump_jobs(u
->manager
, stdout
, " ");
911 continue; /* ask again */
913 /* 'n' was removed in favor of 'f'. */
914 printf("Didn't understand 'n', did you mean 'f'?\n");
915 continue; /* ask again */
917 printf("Skipping execution.\n");
918 r
= CONFIRM_PRETEND_SUCCESS
;
924 assert_not_reached("Unhandled choice");
930 restore_confirm_stdio(&saved_stdin
, &saved_stdout
);
934 static int get_fixed_user(const ExecContext
*c
, const char **user
,
935 uid_t
*uid
, gid_t
*gid
,
936 const char **home
, const char **shell
) {
945 /* Note that we don't set $HOME or $SHELL if they are not particularly enlightening anyway
946 * (i.e. are "/" or "/bin/nologin"). */
949 r
= get_user_creds(&name
, uid
, gid
, home
, shell
, USER_CREDS_CLEAN
);
957 static int get_fixed_group(const ExecContext
*c
, const char **group
, gid_t
*gid
) {
967 r
= get_group_creds(&name
, gid
, 0);
975 static int get_supplementary_groups(const ExecContext
*c
, const char *user
,
976 const char *group
, gid_t gid
,
977 gid_t
**supplementary_gids
, int *ngids
) {
981 bool keep_groups
= false;
982 gid_t
*groups
= NULL
;
983 _cleanup_free_ gid_t
*l_gids
= NULL
;
988 * If user is given, then lookup GID and supplementary groups list.
989 * We avoid NSS lookups for gid=0. Also we have to initialize groups
990 * here and as early as possible so we keep the list of supplementary
991 * groups of the caller.
993 if (user
&& gid_is_valid(gid
) && gid
!= 0) {
994 /* First step, initialize groups from /etc/groups */
995 if (initgroups(user
, gid
) < 0)
1001 if (strv_isempty(c
->supplementary_groups
))
1005 * If SupplementaryGroups= was passed then NGROUPS_MAX has to
1006 * be positive, otherwise fail.
1009 ngroups_max
= (int) sysconf(_SC_NGROUPS_MAX
);
1010 if (ngroups_max
<= 0)
1011 return errno_or_else(EOPNOTSUPP
);
1013 l_gids
= new(gid_t
, ngroups_max
);
1019 * Lookup the list of groups that the user belongs to, we
1020 * avoid NSS lookups here too for gid=0.
1023 if (getgrouplist(user
, gid
, l_gids
, &k
) < 0)
1028 STRV_FOREACH(i
, c
->supplementary_groups
) {
1031 if (k
>= ngroups_max
)
1035 r
= get_group_creds(&g
, l_gids
+k
, 0);
1043 * Sets ngids to zero to drop all supplementary groups, happens
1044 * when we are under root and SupplementaryGroups= is empty.
1051 /* Otherwise get the final list of supplementary groups */
1052 groups
= memdup(l_gids
, sizeof(gid_t
) * k
);
1056 *supplementary_gids
= groups
;
1064 static int enforce_groups(gid_t gid
, const gid_t
*supplementary_gids
, int ngids
) {
1067 /* Handle SupplementaryGroups= if it is not empty */
1069 r
= maybe_setgroups(ngids
, supplementary_gids
);
1074 if (gid_is_valid(gid
)) {
1075 /* Then set our gids */
1076 if (setresgid(gid
, gid
, gid
) < 0)
1083 static int set_securebits(int bits
, int mask
) {
1084 int current
, applied
;
1085 current
= prctl(PR_GET_SECUREBITS
);
1088 /* Clear all securebits defined in mask and set bits */
1089 applied
= (current
& ~mask
) | bits
;
1090 if (current
== applied
)
1092 if (prctl(PR_SET_SECUREBITS
, applied
) < 0)
1097 static int enforce_user(const ExecContext
*context
, uid_t uid
) {
1101 if (!uid_is_valid(uid
))
1104 /* Sets (but doesn't look up) the uid and make sure we keep the
1105 * capabilities while doing so. For setting secure bits the capability CAP_SETPCAP is
1106 * required, so we also need keep-caps in this case.
1109 if (context
->capability_ambient_set
!= 0 || context
->secure_bits
!= 0) {
1111 /* First step: If we need to keep capabilities but
1112 * drop privileges we need to make sure we keep our
1113 * caps, while we drop privileges. */
1115 /* Add KEEP_CAPS to the securebits */
1116 r
= set_securebits(1<<SECURE_KEEP_CAPS
, 0);
1122 /* Second step: actually set the uids */
1123 if (setresuid(uid
, uid
, uid
) < 0)
1126 /* At this point we should have all necessary capabilities but
1127 are otherwise a normal user. However, the caps might got
1128 corrupted due to the setresuid() so we need clean them up
1129 later. This is done outside of this call. */
1136 static int null_conv(
1138 const struct pam_message
**msg
,
1139 struct pam_response
**resp
,
1140 void *appdata_ptr
) {
1142 /* We don't support conversations */
1144 return PAM_CONV_ERR
;
1149 static int setup_pam(
1156 const int fds
[], size_t n_fds
) {
1160 static const struct pam_conv conv
= {
1165 _cleanup_(barrier_destroy
) Barrier barrier
= BARRIER_NULL
;
1166 pam_handle_t
*handle
= NULL
;
1168 int pam_code
= PAM_SUCCESS
, r
;
1169 char **nv
, **e
= NULL
;
1170 bool close_session
= false;
1171 pid_t pam_pid
= 0, parent_pid
;
1178 /* We set up PAM in the parent process, then fork. The child
1179 * will then stay around until killed via PR_GET_PDEATHSIG or
1180 * systemd via the cgroup logic. It will then remove the PAM
1181 * session again. The parent process will exec() the actual
1182 * daemon. We do things this way to ensure that the main PID
1183 * of the daemon is the one we initially fork()ed. */
1185 r
= barrier_create(&barrier
);
1189 if (log_get_max_level() < LOG_DEBUG
)
1190 flags
|= PAM_SILENT
;
1192 pam_code
= pam_start(name
, user
, &conv
, &handle
);
1193 if (pam_code
!= PAM_SUCCESS
) {
1199 _cleanup_free_
char *q
= NULL
;
1201 /* Hmm, so no TTY was explicitly passed, but an fd passed to us directly might be a TTY. Let's figure
1202 * out if that's the case, and read the TTY off it. */
1204 if (getttyname_malloc(STDIN_FILENO
, &q
) >= 0)
1205 tty
= strjoina("/dev/", q
);
1209 pam_code
= pam_set_item(handle
, PAM_TTY
, tty
);
1210 if (pam_code
!= PAM_SUCCESS
)
1214 STRV_FOREACH(nv
, *env
) {
1215 pam_code
= pam_putenv(handle
, *nv
);
1216 if (pam_code
!= PAM_SUCCESS
)
1220 pam_code
= pam_acct_mgmt(handle
, flags
);
1221 if (pam_code
!= PAM_SUCCESS
)
1224 pam_code
= pam_setcred(handle
, PAM_ESTABLISH_CRED
| flags
);
1225 if (pam_code
!= PAM_SUCCESS
)
1226 log_debug("pam_setcred() failed, ignoring: %s", pam_strerror(handle
, pam_code
));
1228 pam_code
= pam_open_session(handle
, flags
);
1229 if (pam_code
!= PAM_SUCCESS
)
1232 close_session
= true;
1234 e
= pam_getenvlist(handle
);
1236 pam_code
= PAM_BUF_ERR
;
1240 /* Block SIGTERM, so that we know that it won't get lost in
1243 assert_se(sigprocmask_many(SIG_BLOCK
, &old_ss
, SIGTERM
, -1) >= 0);
1245 parent_pid
= getpid_cached();
1247 r
= safe_fork("(sd-pam)", 0, &pam_pid
);
1251 int sig
, ret
= EXIT_PAM
;
1253 /* The child's job is to reset the PAM session on
1255 barrier_set_role(&barrier
, BARRIER_CHILD
);
1257 /* Make sure we don't keep open the passed fds in this child. We assume that otherwise only
1258 * those fds are open here that have been opened by PAM. */
1259 (void) close_many(fds
, n_fds
);
1261 /* Drop privileges - we don't need any to pam_close_session
1262 * and this will make PR_SET_PDEATHSIG work in most cases.
1263 * If this fails, ignore the error - but expect sd-pam threads
1264 * to fail to exit normally */
1266 r
= maybe_setgroups(0, NULL
);
1268 log_warning_errno(r
, "Failed to setgroups() in sd-pam: %m");
1269 if (setresgid(gid
, gid
, gid
) < 0)
1270 log_warning_errno(errno
, "Failed to setresgid() in sd-pam: %m");
1271 if (setresuid(uid
, uid
, uid
) < 0)
1272 log_warning_errno(errno
, "Failed to setresuid() in sd-pam: %m");
1274 (void) ignore_signals(SIGPIPE
);
1276 /* Wait until our parent died. This will only work if
1277 * the above setresuid() succeeds, otherwise the kernel
1278 * will not allow unprivileged parents kill their privileged
1279 * children this way. We rely on the control groups kill logic
1280 * to do the rest for us. */
1281 if (prctl(PR_SET_PDEATHSIG
, SIGTERM
) < 0)
1284 /* Tell the parent that our setup is done. This is especially
1285 * important regarding dropping privileges. Otherwise, unit
1286 * setup might race against our setresuid(2) call.
1288 * If the parent aborted, we'll detect this below, hence ignore
1289 * return failure here. */
1290 (void) barrier_place(&barrier
);
1292 /* Check if our parent process might already have died? */
1293 if (getppid() == parent_pid
) {
1296 assert_se(sigemptyset(&ss
) >= 0);
1297 assert_se(sigaddset(&ss
, SIGTERM
) >= 0);
1300 if (sigwait(&ss
, &sig
) < 0) {
1307 assert(sig
== SIGTERM
);
1312 pam_code
= pam_setcred(handle
, PAM_DELETE_CRED
| flags
);
1313 if (pam_code
!= PAM_SUCCESS
)
1316 /* If our parent died we'll end the session */
1317 if (getppid() != parent_pid
) {
1318 pam_code
= pam_close_session(handle
, flags
);
1319 if (pam_code
!= PAM_SUCCESS
)
1326 pam_end(handle
, pam_code
| flags
);
1330 barrier_set_role(&barrier
, BARRIER_PARENT
);
1332 /* If the child was forked off successfully it will do all the
1333 * cleanups, so forget about the handle here. */
1336 /* Unblock SIGTERM again in the parent */
1337 assert_se(sigprocmask(SIG_SETMASK
, &old_ss
, NULL
) >= 0);
1339 /* We close the log explicitly here, since the PAM modules
1340 * might have opened it, but we don't want this fd around. */
1343 /* Synchronously wait for the child to initialize. We don't care for
1344 * errors as we cannot recover. However, warn loudly if it happens. */
1345 if (!barrier_place_and_sync(&barrier
))
1346 log_error("PAM initialization failed");
1348 return strv_free_and_replace(*env
, e
);
1351 if (pam_code
!= PAM_SUCCESS
) {
1352 log_error("PAM failed: %s", pam_strerror(handle
, pam_code
));
1353 r
= -EPERM
; /* PAM errors do not map to errno */
1355 log_error_errno(r
, "PAM failed: %m");
1359 pam_code
= pam_close_session(handle
, flags
);
1361 pam_end(handle
, pam_code
| flags
);
1373 static void rename_process_from_path(const char *path
) {
1374 char process_name
[11];
1378 /* This resulting string must fit in 10 chars (i.e. the length
1379 * of "/sbin/init") to look pretty in /bin/ps */
1383 rename_process("(...)");
1389 /* The end of the process name is usually more
1390 * interesting, since the first bit might just be
1396 process_name
[0] = '(';
1397 memcpy(process_name
+1, p
, l
);
1398 process_name
[1+l
] = ')';
1399 process_name
[1+l
+1] = 0;
1401 rename_process(process_name
);
1404 static bool context_has_address_families(const ExecContext
*c
) {
1407 return c
->address_families_allow_list
||
1408 !set_isempty(c
->address_families
);
1411 static bool context_has_syscall_filters(const ExecContext
*c
) {
1414 return c
->syscall_allow_list
||
1415 !hashmap_isempty(c
->syscall_filter
);
1418 static bool context_has_syscall_logs(const ExecContext
*c
) {
1421 return c
->syscall_log_allow_list
||
1422 !hashmap_isempty(c
->syscall_log
);
1425 static bool context_has_no_new_privileges(const ExecContext
*c
) {
1428 if (c
->no_new_privileges
)
1431 if (have_effective_cap(CAP_SYS_ADMIN
)) /* if we are privileged, we don't need NNP */
1434 /* We need NNP if we have any form of seccomp and are unprivileged */
1435 return c
->lock_personality
||
1436 c
->memory_deny_write_execute
||
1437 c
->private_devices
||
1439 c
->protect_hostname
||
1440 c
->protect_kernel_tunables
||
1441 c
->protect_kernel_modules
||
1442 c
->protect_kernel_logs
||
1443 context_has_address_families(c
) ||
1444 exec_context_restrict_namespaces_set(c
) ||
1445 c
->restrict_realtime
||
1446 c
->restrict_suid_sgid
||
1447 !set_isempty(c
->syscall_archs
) ||
1448 context_has_syscall_filters(c
) ||
1449 context_has_syscall_logs(c
);
1452 static bool exec_context_has_credentials(const ExecContext
*context
) {
1456 return !hashmap_isempty(context
->set_credentials
) ||
1457 context
->load_credentials
;
1462 static bool skip_seccomp_unavailable(const Unit
* u
, const char* msg
) {
1464 if (is_seccomp_available())
1467 log_unit_debug(u
, "SECCOMP features not detected in the kernel, skipping %s", msg
);
1471 static int apply_syscall_filter(const Unit
* u
, const ExecContext
*c
, bool needs_ambient_hack
) {
1472 uint32_t negative_action
, default_action
, action
;
1478 if (!context_has_syscall_filters(c
))
1481 if (skip_seccomp_unavailable(u
, "SystemCallFilter="))
1484 negative_action
= c
->syscall_errno
== SECCOMP_ERROR_NUMBER_KILL
? scmp_act_kill_process() : SCMP_ACT_ERRNO(c
->syscall_errno
);
1486 if (c
->syscall_allow_list
) {
1487 default_action
= negative_action
;
1488 action
= SCMP_ACT_ALLOW
;
1490 default_action
= SCMP_ACT_ALLOW
;
1491 action
= negative_action
;
1494 if (needs_ambient_hack
) {
1495 r
= seccomp_filter_set_add(c
->syscall_filter
, c
->syscall_allow_list
, syscall_filter_sets
+ SYSCALL_FILTER_SET_SETUID
);
1500 return seccomp_load_syscall_filter_set_raw(default_action
, c
->syscall_filter
, action
, false);
1503 static int apply_syscall_log(const Unit
* u
, const ExecContext
*c
) {
1505 uint32_t default_action
, action
;
1511 if (!context_has_syscall_logs(c
))
1515 if (skip_seccomp_unavailable(u
, "SystemCallLog="))
1518 if (c
->syscall_log_allow_list
) {
1519 /* Log nothing but the ones listed */
1520 default_action
= SCMP_ACT_ALLOW
;
1521 action
= SCMP_ACT_LOG
;
1523 /* Log everything but the ones listed */
1524 default_action
= SCMP_ACT_LOG
;
1525 action
= SCMP_ACT_ALLOW
;
1528 return seccomp_load_syscall_filter_set_raw(default_action
, c
->syscall_log
, action
, false);
1530 /* old libseccomp */
1531 log_unit_debug(u
, "SECCOMP feature SCMP_ACT_LOG not available, skipping SystemCallLog=");
1536 static int apply_syscall_archs(const Unit
*u
, const ExecContext
*c
) {
1540 if (set_isempty(c
->syscall_archs
))
1543 if (skip_seccomp_unavailable(u
, "SystemCallArchitectures="))
1546 return seccomp_restrict_archs(c
->syscall_archs
);
1549 static int apply_address_families(const Unit
* u
, const ExecContext
*c
) {
1553 if (!context_has_address_families(c
))
1556 if (skip_seccomp_unavailable(u
, "RestrictAddressFamilies="))
1559 return seccomp_restrict_address_families(c
->address_families
, c
->address_families_allow_list
);
1562 static int apply_memory_deny_write_execute(const Unit
* u
, const ExecContext
*c
) {
1566 if (!c
->memory_deny_write_execute
)
1569 if (skip_seccomp_unavailable(u
, "MemoryDenyWriteExecute="))
1572 return seccomp_memory_deny_write_execute();
1575 static int apply_restrict_realtime(const Unit
* u
, const ExecContext
*c
) {
1579 if (!c
->restrict_realtime
)
1582 if (skip_seccomp_unavailable(u
, "RestrictRealtime="))
1585 return seccomp_restrict_realtime();
1588 static int apply_restrict_suid_sgid(const Unit
* u
, const ExecContext
*c
) {
1592 if (!c
->restrict_suid_sgid
)
1595 if (skip_seccomp_unavailable(u
, "RestrictSUIDSGID="))
1598 return seccomp_restrict_suid_sgid();
1601 static int apply_protect_sysctl(const Unit
*u
, const ExecContext
*c
) {
1605 /* Turn off the legacy sysctl() system call. Many distributions turn this off while building the kernel, but
1606 * let's protect even those systems where this is left on in the kernel. */
1608 if (!c
->protect_kernel_tunables
)
1611 if (skip_seccomp_unavailable(u
, "ProtectKernelTunables="))
1614 return seccomp_protect_sysctl();
1617 static int apply_protect_kernel_modules(const Unit
*u
, const ExecContext
*c
) {
1621 /* Turn off module syscalls on ProtectKernelModules=yes */
1623 if (!c
->protect_kernel_modules
)
1626 if (skip_seccomp_unavailable(u
, "ProtectKernelModules="))
1629 return seccomp_load_syscall_filter_set(SCMP_ACT_ALLOW
, syscall_filter_sets
+ SYSCALL_FILTER_SET_MODULE
, SCMP_ACT_ERRNO(EPERM
), false);
1632 static int apply_protect_kernel_logs(const Unit
*u
, const ExecContext
*c
) {
1636 if (!c
->protect_kernel_logs
)
1639 if (skip_seccomp_unavailable(u
, "ProtectKernelLogs="))
1642 return seccomp_protect_syslog();
1645 static int apply_protect_clock(const Unit
*u
, const ExecContext
*c
) {
1649 if (!c
->protect_clock
)
1652 if (skip_seccomp_unavailable(u
, "ProtectClock="))
1655 return seccomp_load_syscall_filter_set(SCMP_ACT_ALLOW
, syscall_filter_sets
+ SYSCALL_FILTER_SET_CLOCK
, SCMP_ACT_ERRNO(EPERM
), false);
1658 static int apply_private_devices(const Unit
*u
, const ExecContext
*c
) {
1662 /* If PrivateDevices= is set, also turn off iopl and all @raw-io syscalls. */
1664 if (!c
->private_devices
)
1667 if (skip_seccomp_unavailable(u
, "PrivateDevices="))
1670 return seccomp_load_syscall_filter_set(SCMP_ACT_ALLOW
, syscall_filter_sets
+ SYSCALL_FILTER_SET_RAW_IO
, SCMP_ACT_ERRNO(EPERM
), false);
1673 static int apply_restrict_namespaces(const Unit
*u
, const ExecContext
*c
) {
1677 if (!exec_context_restrict_namespaces_set(c
))
1680 if (skip_seccomp_unavailable(u
, "RestrictNamespaces="))
1683 return seccomp_restrict_namespaces(c
->restrict_namespaces
);
1686 static int apply_lock_personality(const Unit
* u
, const ExecContext
*c
) {
1687 unsigned long personality
;
1693 if (!c
->lock_personality
)
1696 if (skip_seccomp_unavailable(u
, "LockPersonality="))
1699 personality
= c
->personality
;
1701 /* If personality is not specified, use either PER_LINUX or PER_LINUX32 depending on what is currently set. */
1702 if (personality
== PERSONALITY_INVALID
) {
1704 r
= opinionated_personality(&personality
);
1709 return seccomp_lock_personality(personality
);
1714 static int apply_protect_hostname(const Unit
*u
, const ExecContext
*c
, int *ret_exit_status
) {
1718 if (!c
->protect_hostname
)
1721 if (ns_type_supported(NAMESPACE_UTS
)) {
1722 if (unshare(CLONE_NEWUTS
) < 0) {
1723 if (!ERRNO_IS_NOT_SUPPORTED(errno
) && !ERRNO_IS_PRIVILEGE(errno
)) {
1724 *ret_exit_status
= EXIT_NAMESPACE
;
1725 return log_unit_error_errno(u
, errno
, "Failed to set up UTS namespacing: %m");
1728 log_unit_warning(u
, "ProtectHostname=yes is configured, but UTS namespace setup is prohibited (container manager?), ignoring namespace setup.");
1731 log_unit_warning(u
, "ProtectHostname=yes is configured, but the kernel does not support UTS namespaces, ignoring namespace setup.");
1736 if (skip_seccomp_unavailable(u
, "ProtectHostname="))
1739 r
= seccomp_protect_hostname();
1741 *ret_exit_status
= EXIT_SECCOMP
;
1742 return log_unit_error_errno(u
, r
, "Failed to apply hostname restrictions: %m");
1749 static void do_idle_pipe_dance(int idle_pipe
[static 4]) {
1752 idle_pipe
[1] = safe_close(idle_pipe
[1]);
1753 idle_pipe
[2] = safe_close(idle_pipe
[2]);
1755 if (idle_pipe
[0] >= 0) {
1758 r
= fd_wait_for_event(idle_pipe
[0], POLLHUP
, IDLE_TIMEOUT_USEC
);
1760 if (idle_pipe
[3] >= 0 && r
== 0 /* timeout */) {
1763 /* Signal systemd that we are bored and want to continue. */
1764 n
= write(idle_pipe
[3], "x", 1);
1766 /* Wait for systemd to react to the signal above. */
1767 (void) fd_wait_for_event(idle_pipe
[0], POLLHUP
, IDLE_TIMEOUT2_USEC
);
1770 idle_pipe
[0] = safe_close(idle_pipe
[0]);
1774 idle_pipe
[3] = safe_close(idle_pipe
[3]);
1777 static const char *exec_directory_env_name_to_string(ExecDirectoryType t
);
1779 static int build_environment(
1781 const ExecContext
*c
,
1782 const ExecParameters
*p
,
1785 const char *username
,
1787 dev_t journal_stream_dev
,
1788 ino_t journal_stream_ino
,
1791 _cleanup_strv_free_
char **our_env
= NULL
;
1800 #define N_ENV_VARS 17
1801 our_env
= new0(char*, N_ENV_VARS
+ _EXEC_DIRECTORY_TYPE_MAX
);
1806 _cleanup_free_
char *joined
= NULL
;
1808 if (asprintf(&x
, "LISTEN_PID="PID_FMT
, getpid_cached()) < 0)
1810 our_env
[n_env
++] = x
;
1812 if (asprintf(&x
, "LISTEN_FDS=%zu", n_fds
) < 0)
1814 our_env
[n_env
++] = x
;
1816 joined
= strv_join(p
->fd_names
, ":");
1820 x
= strjoin("LISTEN_FDNAMES=", joined
);
1823 our_env
[n_env
++] = x
;
1826 if ((p
->flags
& EXEC_SET_WATCHDOG
) && p
->watchdog_usec
> 0) {
1827 if (asprintf(&x
, "WATCHDOG_PID="PID_FMT
, getpid_cached()) < 0)
1829 our_env
[n_env
++] = x
;
1831 if (asprintf(&x
, "WATCHDOG_USEC="USEC_FMT
, p
->watchdog_usec
) < 0)
1833 our_env
[n_env
++] = x
;
1836 /* If this is D-Bus, tell the nss-systemd module, since it relies on being able to use D-Bus look up dynamic
1837 * users via PID 1, possibly dead-locking the dbus daemon. This way it will not use D-Bus to resolve names, but
1838 * check the database directly. */
1839 if (p
->flags
& EXEC_NSS_BYPASS_BUS
) {
1840 x
= strdup("SYSTEMD_NSS_BYPASS_BUS=1");
1843 our_env
[n_env
++] = x
;
1847 x
= strjoin("HOME=", home
);
1851 path_simplify(x
+ 5);
1852 our_env
[n_env
++] = x
;
1856 x
= strjoin("LOGNAME=", username
);
1859 our_env
[n_env
++] = x
;
1861 x
= strjoin("USER=", username
);
1864 our_env
[n_env
++] = x
;
1868 x
= strjoin("SHELL=", shell
);
1872 path_simplify(x
+ 6);
1873 our_env
[n_env
++] = x
;
1876 if (!sd_id128_is_null(u
->invocation_id
)) {
1877 if (asprintf(&x
, "INVOCATION_ID=" SD_ID128_FORMAT_STR
, SD_ID128_FORMAT_VAL(u
->invocation_id
)) < 0)
1880 our_env
[n_env
++] = x
;
1883 if (exec_context_needs_term(c
)) {
1884 const char *tty_path
, *term
= NULL
;
1886 tty_path
= exec_context_tty_path(c
);
1888 /* If we are forked off PID 1 and we are supposed to operate on /dev/console, then let's try
1889 * to inherit the $TERM set for PID 1. This is useful for containers so that the $TERM the
1890 * container manager passes to PID 1 ends up all the way in the console login shown. */
1892 if (path_equal_ptr(tty_path
, "/dev/console") && getppid() == 1)
1893 term
= getenv("TERM");
1896 term
= default_term_for_tty(tty_path
);
1898 x
= strjoin("TERM=", term
);
1901 our_env
[n_env
++] = x
;
1904 if (journal_stream_dev
!= 0 && journal_stream_ino
!= 0) {
1905 if (asprintf(&x
, "JOURNAL_STREAM=" DEV_FMT
":" INO_FMT
, journal_stream_dev
, journal_stream_ino
) < 0)
1908 our_env
[n_env
++] = x
;
1911 if (c
->log_namespace
) {
1912 x
= strjoin("LOG_NAMESPACE=", c
->log_namespace
);
1916 our_env
[n_env
++] = x
;
1919 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
1920 _cleanup_free_
char *pre
= NULL
, *joined
= NULL
;
1926 if (strv_isempty(c
->directories
[t
].paths
))
1929 n
= exec_directory_env_name_to_string(t
);
1933 pre
= strjoin(p
->prefix
[t
], "/");
1937 joined
= strv_join_full(c
->directories
[t
].paths
, ":", pre
, true);
1941 x
= strjoin(n
, "=", joined
);
1945 our_env
[n_env
++] = x
;
1948 if (exec_context_has_credentials(c
) && p
->prefix
[EXEC_DIRECTORY_RUNTIME
]) {
1949 x
= strjoin("CREDENTIALS_DIRECTORY=", p
->prefix
[EXEC_DIRECTORY_RUNTIME
], "/credentials/", u
->id
);
1953 our_env
[n_env
++] = x
;
1956 if (asprintf(&x
, "SYSTEMD_EXEC_PID=" PID_FMT
, getpid_cached()) < 0)
1959 our_env
[n_env
++] = x
;
1961 our_env
[n_env
++] = NULL
;
1962 assert(n_env
<= N_ENV_VARS
+ _EXEC_DIRECTORY_TYPE_MAX
);
1965 *ret
= TAKE_PTR(our_env
);
1970 static int build_pass_environment(const ExecContext
*c
, char ***ret
) {
1971 _cleanup_strv_free_
char **pass_env
= NULL
;
1975 STRV_FOREACH(i
, c
->pass_environment
) {
1976 _cleanup_free_
char *x
= NULL
;
1982 x
= strjoin(*i
, "=", v
);
1986 if (!GREEDY_REALLOC(pass_env
, n_env
+ 2))
1989 pass_env
[n_env
++] = TAKE_PTR(x
);
1990 pass_env
[n_env
] = NULL
;
1993 *ret
= TAKE_PTR(pass_env
);
1998 bool exec_needs_mount_namespace(
1999 const ExecContext
*context
,
2000 const ExecParameters
*params
,
2001 const ExecRuntime
*runtime
) {
2005 if (context
->root_image
)
2008 if (!strv_isempty(context
->read_write_paths
) ||
2009 !strv_isempty(context
->read_only_paths
) ||
2010 !strv_isempty(context
->inaccessible_paths
) ||
2011 !strv_isempty(context
->exec_paths
) ||
2012 !strv_isempty(context
->no_exec_paths
))
2015 if (context
->n_bind_mounts
> 0)
2018 if (context
->n_temporary_filesystems
> 0)
2021 if (context
->n_mount_images
> 0)
2024 if (context
->n_extension_images
> 0)
2027 if (!IN_SET(context
->mount_flags
, 0, MS_SHARED
))
2030 if (context
->private_tmp
&& runtime
&& (runtime
->tmp_dir
|| runtime
->var_tmp_dir
))
2033 if (context
->private_devices
||
2034 context
->private_mounts
||
2035 context
->protect_system
!= PROTECT_SYSTEM_NO
||
2036 context
->protect_home
!= PROTECT_HOME_NO
||
2037 context
->protect_kernel_tunables
||
2038 context
->protect_kernel_modules
||
2039 context
->protect_kernel_logs
||
2040 context
->protect_control_groups
||
2041 context
->protect_proc
!= PROTECT_PROC_DEFAULT
||
2042 context
->proc_subset
!= PROC_SUBSET_ALL
||
2043 context
->private_ipc
||
2044 context
->ipc_namespace_path
)
2047 if (context
->root_directory
) {
2048 if (exec_context_get_effective_mount_apivfs(context
))
2051 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
2052 if (params
&& !params
->prefix
[t
])
2055 if (!strv_isempty(context
->directories
[t
].paths
))
2060 if (context
->dynamic_user
&&
2061 (!strv_isempty(context
->directories
[EXEC_DIRECTORY_STATE
].paths
) ||
2062 !strv_isempty(context
->directories
[EXEC_DIRECTORY_CACHE
].paths
) ||
2063 !strv_isempty(context
->directories
[EXEC_DIRECTORY_LOGS
].paths
)))
2066 if (context
->log_namespace
)
2072 static int setup_private_users(uid_t ouid
, gid_t ogid
, uid_t uid
, gid_t gid
) {
2073 _cleanup_free_
char *uid_map
= NULL
, *gid_map
= NULL
;
2074 _cleanup_close_pair_
int errno_pipe
[2] = { -1, -1 };
2075 _cleanup_close_
int unshare_ready_fd
= -1;
2076 _cleanup_(sigkill_waitp
) pid_t pid
= 0;
2081 /* Set up a user namespace and map the original UID/GID (IDs from before any user or group changes, i.e.
2082 * the IDs from the user or system manager(s)) to itself, the selected UID/GID to itself, and everything else to
2083 * nobody. In order to be able to write this mapping we need CAP_SETUID in the original user namespace, which
2084 * we however lack after opening the user namespace. To work around this we fork() a temporary child process,
2085 * which waits for the parent to create the new user namespace while staying in the original namespace. The
2086 * child then writes the UID mapping, under full privileges. The parent waits for the child to finish and
2087 * continues execution normally.
2088 * For unprivileged users (i.e. without capabilities), the root to root mapping is excluded. As such, it
2089 * does not need CAP_SETUID to write the single line mapping to itself. */
2091 /* Can only set up multiple mappings with CAP_SETUID. */
2092 if (have_effective_cap(CAP_SETUID
) && uid
!= ouid
&& uid_is_valid(uid
))
2093 r
= asprintf(&uid_map
,
2094 UID_FMT
" " UID_FMT
" 1\n" /* Map $OUID → $OUID */
2095 UID_FMT
" " UID_FMT
" 1\n", /* Map $UID → $UID */
2096 ouid
, ouid
, uid
, uid
);
2098 r
= asprintf(&uid_map
,
2099 UID_FMT
" " UID_FMT
" 1\n", /* Map $OUID → $OUID */
2105 /* Can only set up multiple mappings with CAP_SETGID. */
2106 if (have_effective_cap(CAP_SETGID
) && gid
!= ogid
&& gid_is_valid(gid
))
2107 r
= asprintf(&gid_map
,
2108 GID_FMT
" " GID_FMT
" 1\n" /* Map $OGID → $OGID */
2109 GID_FMT
" " GID_FMT
" 1\n", /* Map $GID → $GID */
2110 ogid
, ogid
, gid
, gid
);
2112 r
= asprintf(&gid_map
,
2113 GID_FMT
" " GID_FMT
" 1\n", /* Map $OGID -> $OGID */
2119 /* Create a communication channel so that the parent can tell the child when it finished creating the user
2121 unshare_ready_fd
= eventfd(0, EFD_CLOEXEC
);
2122 if (unshare_ready_fd
< 0)
2125 /* Create a communication channel so that the child can tell the parent a proper error code in case it
2127 if (pipe2(errno_pipe
, O_CLOEXEC
) < 0)
2130 r
= safe_fork("(sd-userns)", FORK_RESET_SIGNALS
|FORK_DEATHSIG
, &pid
);
2134 _cleanup_close_
int fd
= -1;
2138 /* Child process, running in the original user namespace. Let's update the parent's UID/GID map from
2139 * here, after the parent opened its own user namespace. */
2142 errno_pipe
[0] = safe_close(errno_pipe
[0]);
2144 /* Wait until the parent unshared the user namespace */
2145 if (read(unshare_ready_fd
, &c
, sizeof(c
)) < 0) {
2150 /* Disable the setgroups() system call in the child user namespace, for good. */
2151 a
= procfs_file_alloca(ppid
, "setgroups");
2152 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
2154 if (errno
!= ENOENT
) {
2159 /* If the file is missing the kernel is too old, let's continue anyway. */
2161 if (write(fd
, "deny\n", 5) < 0) {
2166 fd
= safe_close(fd
);
2169 /* First write the GID map */
2170 a
= procfs_file_alloca(ppid
, "gid_map");
2171 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
2176 if (write(fd
, gid_map
, strlen(gid_map
)) < 0) {
2180 fd
= safe_close(fd
);
2182 /* The write the UID map */
2183 a
= procfs_file_alloca(ppid
, "uid_map");
2184 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
2189 if (write(fd
, uid_map
, strlen(uid_map
)) < 0) {
2194 _exit(EXIT_SUCCESS
);
2197 (void) write(errno_pipe
[1], &r
, sizeof(r
));
2198 _exit(EXIT_FAILURE
);
2201 errno_pipe
[1] = safe_close(errno_pipe
[1]);
2203 if (unshare(CLONE_NEWUSER
) < 0)
2206 /* Let the child know that the namespace is ready now */
2207 if (write(unshare_ready_fd
, &c
, sizeof(c
)) < 0)
2210 /* Try to read an error code from the child */
2211 n
= read(errno_pipe
[0], &r
, sizeof(r
));
2214 if (n
== sizeof(r
)) { /* an error code was sent to us */
2219 if (n
!= 0) /* on success we should have read 0 bytes */
2222 r
= wait_for_terminate_and_check("(sd-userns)", pid
, 0);
2226 if (r
!= EXIT_SUCCESS
) /* If something strange happened with the child, let's consider this fatal, too */
2232 static bool exec_directory_is_private(const ExecContext
*context
, ExecDirectoryType type
) {
2233 if (!context
->dynamic_user
)
2236 if (type
== EXEC_DIRECTORY_CONFIGURATION
)
2239 if (type
== EXEC_DIRECTORY_RUNTIME
&& context
->runtime_directory_preserve_mode
== EXEC_PRESERVE_NO
)
2245 static int setup_exec_directory(
2246 const ExecContext
*context
,
2247 const ExecParameters
*params
,
2250 ExecDirectoryType type
,
2253 static const int exit_status_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
2254 [EXEC_DIRECTORY_RUNTIME
] = EXIT_RUNTIME_DIRECTORY
,
2255 [EXEC_DIRECTORY_STATE
] = EXIT_STATE_DIRECTORY
,
2256 [EXEC_DIRECTORY_CACHE
] = EXIT_CACHE_DIRECTORY
,
2257 [EXEC_DIRECTORY_LOGS
] = EXIT_LOGS_DIRECTORY
,
2258 [EXEC_DIRECTORY_CONFIGURATION
] = EXIT_CONFIGURATION_DIRECTORY
,
2265 assert(type
>= 0 && type
< _EXEC_DIRECTORY_TYPE_MAX
);
2266 assert(exit_status
);
2268 if (!params
->prefix
[type
])
2271 if (params
->flags
& EXEC_CHOWN_DIRECTORIES
) {
2272 if (!uid_is_valid(uid
))
2274 if (!gid_is_valid(gid
))
2278 STRV_FOREACH(rt
, context
->directories
[type
].paths
) {
2279 _cleanup_free_
char *p
= NULL
, *pp
= NULL
;
2281 p
= path_join(params
->prefix
[type
], *rt
);
2287 r
= mkdir_parents_label(p
, 0755);
2291 if (exec_directory_is_private(context
, type
)) {
2292 /* So, here's one extra complication when dealing with DynamicUser=1 units. In that
2293 * case we want to avoid leaving a directory around fully accessible that is owned by
2294 * a dynamic user whose UID is later on reused. To lock this down we use the same
2295 * trick used by container managers to prohibit host users to get access to files of
2296 * the same UID in containers: we place everything inside a directory that has an
2297 * access mode of 0700 and is owned root:root, so that it acts as security boundary
2298 * for unprivileged host code. We then use fs namespacing to make this directory
2299 * permeable for the service itself.
2301 * Specifically: for a service which wants a special directory "foo/" we first create
2302 * a directory "private/" with access mode 0700 owned by root:root. Then we place
2303 * "foo" inside of that directory (i.e. "private/foo/"), and make "foo" a symlink to
2304 * "private/foo". This way, privileged host users can access "foo/" as usual, but
2305 * unprivileged host users can't look into it. Inside of the namespace of the unit
2306 * "private/" is replaced by a more liberally accessible tmpfs, into which the host's
2307 * "private/foo/" is mounted under the same name, thus disabling the access boundary
2308 * for the service and making sure it only gets access to the dirs it needs but no
2309 * others. Tricky? Yes, absolutely, but it works!
2311 * Note that we don't do this for EXEC_DIRECTORY_CONFIGURATION as that's assumed not
2312 * to be owned by the service itself.
2314 * Also, note that we don't do this for EXEC_DIRECTORY_RUNTIME as that's often used
2315 * for sharing files or sockets with other services. */
2317 pp
= path_join(params
->prefix
[type
], "private");
2323 /* First set up private root if it doesn't exist yet, with access mode 0700 and owned by root:root */
2324 r
= mkdir_safe_label(pp
, 0700, 0, 0, MKDIR_WARN_MODE
);
2328 if (!path_extend(&pp
, *rt
)) {
2333 /* Create all directories between the configured directory and this private root, and mark them 0755 */
2334 r
= mkdir_parents_label(pp
, 0755);
2338 if (is_dir(p
, false) > 0 &&
2339 (laccess(pp
, F_OK
) < 0 && errno
== ENOENT
)) {
2341 /* Hmm, the private directory doesn't exist yet, but the normal one exists? If so, move
2342 * it over. Most likely the service has been upgraded from one that didn't use
2343 * DynamicUser=1, to one that does. */
2345 log_info("Found pre-existing public %s= directory %s, migrating to %s.\n"
2346 "Apparently, service previously had DynamicUser= turned off, and has now turned it on.",
2347 exec_directory_type_to_string(type
), p
, pp
);
2349 if (rename(p
, pp
) < 0) {
2354 /* Otherwise, create the actual directory for the service */
2356 r
= mkdir_label(pp
, context
->directories
[type
].mode
);
2357 if (r
< 0 && r
!= -EEXIST
)
2361 /* And link it up from the original place */
2362 r
= symlink_idempotent(pp
, p
, true);
2367 _cleanup_free_
char *target
= NULL
;
2369 if (type
!= EXEC_DIRECTORY_CONFIGURATION
&&
2370 readlink_and_make_absolute(p
, &target
) >= 0) {
2371 _cleanup_free_
char *q
= NULL
, *q_resolved
= NULL
, *target_resolved
= NULL
;
2373 /* This already exists and is a symlink? Interesting. Maybe it's one created
2374 * by DynamicUser=1 (see above)?
2376 * We do this for all directory types except for ConfigurationDirectory=,
2377 * since they all support the private/ symlink logic at least in some
2378 * configurations, see above. */
2380 r
= chase_symlinks(target
, NULL
, 0, &target_resolved
, NULL
);
2384 q
= path_join(params
->prefix
[type
], "private", *rt
);
2390 /* /var/lib or friends may be symlinks. So, let's chase them also. */
2391 r
= chase_symlinks(q
, NULL
, CHASE_NONEXISTENT
, &q_resolved
, NULL
);
2395 if (path_equal(q_resolved
, target_resolved
)) {
2397 /* Hmm, apparently DynamicUser= was once turned on for this service,
2398 * but is no longer. Let's move the directory back up. */
2400 log_info("Found pre-existing private %s= directory %s, migrating to %s.\n"
2401 "Apparently, service previously had DynamicUser= turned on, and has now turned it off.",
2402 exec_directory_type_to_string(type
), q
, p
);
2404 if (unlink(p
) < 0) {
2409 if (rename(q
, p
) < 0) {
2416 r
= mkdir_label(p
, context
->directories
[type
].mode
);
2421 if (type
== EXEC_DIRECTORY_CONFIGURATION
) {
2424 /* Don't change the owner/access mode of the configuration directory,
2425 * as in the common case it is not written to by a service, and shall
2426 * not be writable. */
2428 if (stat(p
, &st
) < 0) {
2433 /* Still complain if the access mode doesn't match */
2434 if (((st
.st_mode
^ context
->directories
[type
].mode
) & 07777) != 0)
2435 log_warning("%s \'%s\' already exists but the mode is different. "
2436 "(File system: %o %sMode: %o)",
2437 exec_directory_type_to_string(type
), *rt
,
2438 st
.st_mode
& 07777, exec_directory_type_to_string(type
), context
->directories
[type
].mode
& 07777);
2445 /* Lock down the access mode (we use chmod_and_chown() to make this idempotent. We don't
2446 * specify UID/GID here, so that path_chown_recursive() can optimize things depending on the
2447 * current UID/GID ownership.) */
2448 r
= chmod_and_chown(pp
?: p
, context
->directories
[type
].mode
, UID_INVALID
, GID_INVALID
);
2452 /* Then, change the ownership of the whole tree, if necessary. When dynamic users are used we
2453 * drop the suid/sgid bits, since we really don't want SUID/SGID files for dynamic UID/GID
2454 * assignments to exist. */
2455 r
= path_chown_recursive(pp
?: p
, uid
, gid
, context
->dynamic_user
? 01777 : 07777);
2463 *exit_status
= exit_status_table
[type
];
2467 static int write_credential(
2473 bool ownership_ok
) {
2475 _cleanup_(unlink_and_freep
) char *tmp
= NULL
;
2476 _cleanup_close_
int fd
= -1;
2479 r
= tempfn_random_child("", "cred", &tmp
);
2483 fd
= openat(dfd
, tmp
, O_CREAT
|O_RDWR
|O_CLOEXEC
|O_EXCL
|O_NOFOLLOW
|O_NOCTTY
, 0600);
2489 r
= loop_write(fd
, data
, size
, /* do_pool = */ false);
2493 if (fchmod(fd
, 0400) < 0) /* Take away "w" bit */
2496 if (uid_is_valid(uid
) && uid
!= getuid()) {
2497 r
= fd_add_uid_acl_permission(fd
, uid
, ACL_READ
);
2499 if (!ERRNO_IS_NOT_SUPPORTED(r
) && !ERRNO_IS_PRIVILEGE(r
))
2502 if (!ownership_ok
) /* Ideally we use ACLs, since we can neatly express what we want
2503 * to express: that the user gets read access and nothing
2504 * else. But if the backing fs can't support that (e.g. ramfs)
2505 * then we can use file ownership instead. But that's only safe if
2506 * we can then re-mount the whole thing read-only, so that the
2507 * user can no longer chmod() the file to gain write access. */
2510 if (fchown(fd
, uid
, GID_INVALID
) < 0)
2515 if (renameat(dfd
, tmp
, dfd
, id
) < 0)
2522 #define CREDENTIALS_BYTES_MAX (1024LU * 1024LU) /* Refuse to pass more than 1M, after all this is unswappable memory */
2524 static int acquire_credentials(
2525 const ExecContext
*context
,
2526 const ExecParameters
*params
,
2530 bool ownership_ok
) {
2532 uint64_t left
= CREDENTIALS_BYTES_MAX
;
2533 _cleanup_close_
int dfd
= -1;
2534 ExecSetCredential
*sc
;
2541 dfd
= open(p
, O_DIRECTORY
|O_CLOEXEC
);
2545 /* First we use the literally specified credentials. Note that they might be overridden again below,
2546 * and thus act as a "default" if the same credential is specified multiple times */
2547 HASHMAP_FOREACH(sc
, context
->set_credentials
) {
2550 add
= strlen(sc
->id
) + sc
->size
;
2554 r
= write_credential(dfd
, sc
->id
, sc
->data
, sc
->size
, uid
, ownership_ok
);
2561 /* Then, load credential off disk (or acquire via AF_UNIX socket) */
2562 STRV_FOREACH_PAIR(id
, fn
, context
->load_credentials
) {
2563 ReadFullFileFlags flags
= READ_FULL_FILE_SECURE
;
2564 _cleanup_(erase_and_freep
) char *data
= NULL
;
2565 _cleanup_free_
char *j
= NULL
, *bindname
= NULL
;
2566 bool missing_ok
= true;
2570 if (path_is_absolute(*fn
)) {
2571 /* If this is an absolute path, read the data directly from it, and support AF_UNIX sockets */
2573 flags
|= READ_FULL_FILE_CONNECT_SOCKET
;
2575 /* Pass some minimal info about the unit and the credential name we are looking to acquire
2576 * via the source socket address in case we read off an AF_UNIX socket. */
2577 if (asprintf(&bindname
, "@%" PRIx64
"/unit/%s/%s", random_u64(), unit
, *id
) < 0)
2582 } else if (params
->received_credentials
) {
2583 /* If this is a relative path, take it relative to the credentials we received
2584 * ourselves. We don't support the AF_UNIX stuff in this mode, since we are operating
2585 * on a credential store, i.e. this is guaranteed to be regular files. */
2586 j
= path_join(params
->received_credentials
, *fn
);
2595 r
= read_full_file_full(AT_FDCWD
, source
, UINT64_MAX
, SIZE_MAX
, flags
, bindname
, &data
, &size
);
2598 if (r
== -ENOENT
&& (missing_ok
|| faccessat(dfd
, *id
, F_OK
, AT_SYMLINK_NOFOLLOW
) >= 0)) {
2599 /* Make a missing inherited credential non-fatal, let's just continue. After all apps
2600 * will get clear errors if we don't pass such a missing credential on as they
2601 * themselves will get ENOENT when trying to read them, which should not be much
2602 * worse than when we handle the error here and make it fatal.
2604 * Also, if the source file doesn't exist, but we already acquired the key otherwise,
2605 * then don't fail either. */
2606 log_debug_errno(r
, "Couldn't read inherited credential '%s', skipping: %m", *fn
);
2610 return log_debug_errno(r
, "Failed to read credential '%s': %m", *fn
);
2612 add
= strlen(*id
) + size
;
2616 r
= write_credential(dfd
, *id
, data
, size
, uid
, ownership_ok
);
2623 if (fchmod(dfd
, 0500) < 0) /* Now take away the "w" bit */
2626 /* After we created all keys with the right perms, also make sure the credential store as a whole is
2629 if (uid_is_valid(uid
) && uid
!= getuid()) {
2630 r
= fd_add_uid_acl_permission(dfd
, uid
, ACL_READ
| ACL_EXECUTE
);
2632 if (!ERRNO_IS_NOT_SUPPORTED(r
) && !ERRNO_IS_PRIVILEGE(r
))
2638 if (fchown(dfd
, uid
, GID_INVALID
) < 0)
2646 static int setup_credentials_internal(
2647 const ExecContext
*context
,
2648 const ExecParameters
*params
,
2650 const char *final
, /* This is where the credential store shall eventually end up at */
2651 const char *workspace
, /* This is where we can prepare it before moving it to the final place */
2652 bool reuse_workspace
, /* Whether to reuse any existing workspace mount if it already is a mount */
2653 bool must_mount
, /* Whether to require that we mount something, it's not OK to use the plain directory fall back */
2656 int r
, workspace_mounted
; /* negative if we don't know yet whether we have/can mount something; true
2657 * if we mounted something; false if we definitely can't mount anything */
2665 if (reuse_workspace
) {
2666 r
= path_is_mount_point(workspace
, NULL
, 0);
2670 workspace_mounted
= true; /* If this is already a mount, and we are supposed to reuse it, let's keep this in mind */
2672 workspace_mounted
= -1; /* We need to figure out if we can mount something to the workspace */
2674 workspace_mounted
= -1; /* ditto */
2676 r
= path_is_mount_point(final
, NULL
, 0);
2680 /* If the final place already has something mounted, we use that. If the workspace also has
2681 * something mounted we assume it's actually the same mount (but with MS_RDONLY
2683 final_mounted
= true;
2685 if (workspace_mounted
< 0) {
2686 /* If the final place is mounted, but the workspace we isn't, then let's bind mount
2687 * the final version to the workspace, and make it writable, so that we can make
2690 r
= mount_nofollow_verbose(LOG_DEBUG
, final
, workspace
, NULL
, MS_BIND
|MS_REC
, NULL
);
2694 r
= mount_nofollow_verbose(LOG_DEBUG
, NULL
, workspace
, NULL
, MS_BIND
|MS_REMOUNT
|MS_NODEV
|MS_NOEXEC
|MS_NOSUID
, NULL
);
2698 workspace_mounted
= true;
2701 final_mounted
= false;
2703 if (workspace_mounted
< 0) {
2704 /* Nothing is mounted on the workspace yet, let's try to mount something now */
2705 for (int try = 0;; try++) {
2708 /* Try "ramfs" first, since it's not swap backed */
2709 r
= mount_nofollow_verbose(LOG_DEBUG
, "ramfs", workspace
, "ramfs", MS_NODEV
|MS_NOEXEC
|MS_NOSUID
, "mode=0700");
2711 workspace_mounted
= true;
2715 } else if (try == 1) {
2716 _cleanup_free_
char *opts
= NULL
;
2718 if (asprintf(&opts
, "mode=0700,nr_inodes=1024,size=%lu", CREDENTIALS_BYTES_MAX
) < 0)
2721 /* Fall back to "tmpfs" otherwise */
2722 r
= mount_nofollow_verbose(LOG_DEBUG
, "tmpfs", workspace
, "tmpfs", MS_NODEV
|MS_NOEXEC
|MS_NOSUID
, opts
);
2724 workspace_mounted
= true;
2729 /* 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. */
2730 r
= mount_nofollow_verbose(LOG_DEBUG
, final
, workspace
, NULL
, MS_BIND
|MS_REC
, NULL
);
2732 if (!ERRNO_IS_PRIVILEGE(r
)) /* Propagate anything that isn't a permission problem */
2735 if (must_mount
) /* If we it's not OK to use the plain directory
2736 * fallback, propagate all errors too */
2739 /* If we lack privileges to bind mount stuff, then let's gracefully
2740 * proceed for compat with container envs, and just use the final dir
2743 workspace_mounted
= false;
2747 /* Make the new bind mount writable (i.e. drop MS_RDONLY) */
2748 r
= mount_nofollow_verbose(LOG_DEBUG
, NULL
, workspace
, NULL
, MS_BIND
|MS_REMOUNT
|MS_NODEV
|MS_NOEXEC
|MS_NOSUID
, NULL
);
2752 workspace_mounted
= true;
2758 assert(!must_mount
|| workspace_mounted
> 0);
2759 where
= workspace_mounted
? workspace
: final
;
2761 r
= acquire_credentials(context
, params
, unit
, where
, uid
, workspace_mounted
);
2765 if (workspace_mounted
) {
2766 /* Make workspace read-only now, so that any bind mount we make from it defaults to read-only too */
2767 r
= mount_nofollow_verbose(LOG_DEBUG
, NULL
, workspace
, NULL
, MS_BIND
|MS_REMOUNT
|MS_RDONLY
|MS_NODEV
|MS_NOEXEC
|MS_NOSUID
, NULL
);
2771 /* And mount it to the final place, read-only */
2773 r
= umount_verbose(LOG_DEBUG
, workspace
, MNT_DETACH
|UMOUNT_NOFOLLOW
);
2775 r
= mount_nofollow_verbose(LOG_DEBUG
, workspace
, final
, NULL
, MS_MOVE
, NULL
);
2779 _cleanup_free_
char *parent
= NULL
;
2781 /* If we do not have our own mount put used the plain directory fallback, then we need to
2782 * open access to the top-level credential directory and the per-service directory now */
2784 parent
= dirname_malloc(final
);
2787 if (chmod(parent
, 0755) < 0)
2794 static int setup_credentials(
2795 const ExecContext
*context
,
2796 const ExecParameters
*params
,
2800 _cleanup_free_
char *p
= NULL
, *q
= NULL
;
2807 if (!exec_context_has_credentials(context
))
2810 if (!params
->prefix
[EXEC_DIRECTORY_RUNTIME
])
2813 /* This where we'll place stuff when we are done; this main credentials directory is world-readable,
2814 * and the subdir we mount over with a read-only file system readable by the service's user */
2815 q
= path_join(params
->prefix
[EXEC_DIRECTORY_RUNTIME
], "credentials");
2819 r
= mkdir_label(q
, 0755); /* top-level dir: world readable/searchable */
2820 if (r
< 0 && r
!= -EEXIST
)
2823 p
= path_join(q
, unit
);
2827 r
= mkdir_label(p
, 0700); /* per-unit dir: private to user */
2828 if (r
< 0 && r
!= -EEXIST
)
2831 r
= safe_fork("(sd-mkdcreds)", FORK_DEATHSIG
|FORK_WAIT
|FORK_NEW_MOUNTNS
, NULL
);
2833 _cleanup_free_
char *t
= NULL
, *u
= NULL
;
2835 /* If this is not a privilege or support issue then propagate the error */
2836 if (!ERRNO_IS_NOT_SUPPORTED(r
) && !ERRNO_IS_PRIVILEGE(r
))
2839 /* Temporary workspace, that remains inaccessible all the time. We prepare stuff there before moving
2840 * it into place, so that users can't access half-initialized credential stores. */
2841 t
= path_join(params
->prefix
[EXEC_DIRECTORY_RUNTIME
], "systemd/temporary-credentials");
2845 /* We can't set up a mount namespace. In that case operate on a fixed, inaccessible per-unit
2846 * directory outside of /run/credentials/ first, and then move it over to /run/credentials/
2847 * after it is fully set up */
2848 u
= path_join(t
, unit
);
2852 FOREACH_STRING(i
, t
, u
) {
2853 r
= mkdir_label(i
, 0700);
2854 if (r
< 0 && r
!= -EEXIST
)
2858 r
= setup_credentials_internal(
2862 p
, /* final mount point */
2863 u
, /* temporary workspace to overmount */
2864 true, /* reuse the workspace if it is already a mount */
2865 false, /* it's OK to fall back to a plain directory if we can't mount anything */
2868 (void) rmdir(u
); /* remove the workspace again if we can. */
2873 } else if (r
== 0) {
2875 /* We managed to set up a mount namespace, and are now in a child. That's great. In this case
2876 * we can use the same directory for all cases, after turning off propagation. Question
2877 * though is: where do we turn off propagation exactly, and where do we place the workspace
2878 * directory? We need some place that is guaranteed to be a mount point in the host, and
2879 * which is guaranteed to have a subdir we can mount over. /run/ is not suitable for this,
2880 * since we ultimately want to move the resulting file system there, i.e. we need propagation
2881 * for /run/ eventually. We could use our own /run/systemd/bind mount on itself, but that
2882 * would be visible in the host mount table all the time, which we want to avoid. Hence, what
2883 * we do here instead we use /dev/ and /dev/shm/ for our purposes. We know for sure that
2884 * /dev/ is a mount point and we now for sure that /dev/shm/ exists. Hence we can turn off
2885 * propagation on the former, and then overmount the latter.
2887 * Yes it's nasty playing games with /dev/ and /dev/shm/ like this, since it does not exist
2888 * for this purpose, but there are few other candidates that work equally well for us, and
2889 * given that the we do this in a privately namespaced short-lived single-threaded process
2890 * that no one else sees this should be OK to do. */
2892 r
= mount_nofollow_verbose(LOG_DEBUG
, NULL
, "/dev", NULL
, MS_SLAVE
|MS_REC
, NULL
); /* Turn off propagation from our namespace to host */
2896 r
= setup_credentials_internal(
2900 p
, /* final mount point */
2901 "/dev/shm", /* temporary workspace to overmount */
2902 false, /* do not reuse /dev/shm if it is already a mount, under no circumstances */
2903 true, /* insist that something is mounted, do not allow fallback to plain directory */
2908 _exit(EXIT_SUCCESS
);
2911 _exit(EXIT_FAILURE
);
2918 static int setup_smack(
2919 const ExecContext
*context
,
2920 int executable_fd
) {
2924 assert(executable_fd
>= 0);
2926 if (context
->smack_process_label
) {
2927 r
= mac_smack_apply_pid(0, context
->smack_process_label
);
2931 #ifdef SMACK_DEFAULT_PROCESS_LABEL
2933 _cleanup_free_
char *exec_label
= NULL
;
2935 r
= mac_smack_read_fd(executable_fd
, SMACK_ATTR_EXEC
, &exec_label
);
2936 if (r
< 0 && !IN_SET(r
, -ENODATA
, -EOPNOTSUPP
))
2939 r
= mac_smack_apply_pid(0, exec_label
? : SMACK_DEFAULT_PROCESS_LABEL
);
2949 static int compile_bind_mounts(
2950 const ExecContext
*context
,
2951 const ExecParameters
*params
,
2952 BindMount
**ret_bind_mounts
,
2953 size_t *ret_n_bind_mounts
,
2954 char ***ret_empty_directories
) {
2956 _cleanup_strv_free_
char **empty_directories
= NULL
;
2957 BindMount
*bind_mounts
;
2963 assert(ret_bind_mounts
);
2964 assert(ret_n_bind_mounts
);
2965 assert(ret_empty_directories
);
2967 n
= context
->n_bind_mounts
;
2968 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
2969 if (!params
->prefix
[t
])
2972 n
+= strv_length(context
->directories
[t
].paths
);
2976 *ret_bind_mounts
= NULL
;
2977 *ret_n_bind_mounts
= 0;
2978 *ret_empty_directories
= NULL
;
2982 bind_mounts
= new(BindMount
, n
);
2986 for (size_t i
= 0; i
< context
->n_bind_mounts
; i
++) {
2987 BindMount
*item
= context
->bind_mounts
+ i
;
2990 s
= strdup(item
->source
);
2996 d
= strdup(item
->destination
);
3003 bind_mounts
[h
++] = (BindMount
) {
3006 .read_only
= item
->read_only
,
3007 .recursive
= item
->recursive
,
3008 .ignore_enoent
= item
->ignore_enoent
,
3012 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
3015 if (!params
->prefix
[t
])
3018 if (strv_isempty(context
->directories
[t
].paths
))
3021 if (exec_directory_is_private(context
, t
) &&
3022 !exec_context_with_rootfs(context
)) {
3025 /* So this is for a dynamic user, and we need to make sure the process can access its own
3026 * directory. For that we overmount the usually inaccessible "private" subdirectory with a
3027 * tmpfs that makes it accessible and is empty except for the submounts we do this for. */
3029 private_root
= path_join(params
->prefix
[t
], "private");
3030 if (!private_root
) {
3035 r
= strv_consume(&empty_directories
, private_root
);
3040 STRV_FOREACH(suffix
, context
->directories
[t
].paths
) {
3043 if (exec_directory_is_private(context
, t
))
3044 s
= path_join(params
->prefix
[t
], "private", *suffix
);
3046 s
= path_join(params
->prefix
[t
], *suffix
);
3052 if (exec_directory_is_private(context
, t
) &&
3053 exec_context_with_rootfs(context
))
3054 /* When RootDirectory= or RootImage= are set, then the symbolic link to the private
3055 * directory is not created on the root directory. So, let's bind-mount the directory
3056 * on the 'non-private' place. */
3057 d
= path_join(params
->prefix
[t
], *suffix
);
3066 bind_mounts
[h
++] = (BindMount
) {
3070 .nosuid
= context
->dynamic_user
, /* don't allow suid/sgid when DynamicUser= is on */
3072 .ignore_enoent
= false,
3079 *ret_bind_mounts
= bind_mounts
;
3080 *ret_n_bind_mounts
= n
;
3081 *ret_empty_directories
= TAKE_PTR(empty_directories
);
3086 bind_mount_free_many(bind_mounts
, h
);
3090 static bool insist_on_sandboxing(
3091 const ExecContext
*context
,
3092 const char *root_dir
,
3093 const char *root_image
,
3094 const BindMount
*bind_mounts
,
3095 size_t n_bind_mounts
) {
3098 assert(n_bind_mounts
== 0 || bind_mounts
);
3100 /* Checks whether we need to insist on fs namespacing. i.e. whether we have settings configured that
3101 * would alter the view on the file system beyond making things read-only or invisible, i.e. would
3102 * rearrange stuff in a way we cannot ignore gracefully. */
3104 if (context
->n_temporary_filesystems
> 0)
3107 if (root_dir
|| root_image
)
3110 if (context
->n_mount_images
> 0)
3113 if (context
->dynamic_user
)
3116 /* If there are any bind mounts set that don't map back onto themselves, fs namespacing becomes
3118 for (size_t i
= 0; i
< n_bind_mounts
; i
++)
3119 if (!path_equal(bind_mounts
[i
].source
, bind_mounts
[i
].destination
))
3122 if (context
->log_namespace
)
3128 static int apply_mount_namespace(
3130 ExecCommandFlags command_flags
,
3131 const ExecContext
*context
,
3132 const ExecParameters
*params
,
3133 const ExecRuntime
*runtime
,
3134 char **error_path
) {
3136 _cleanup_strv_free_
char **empty_directories
= NULL
;
3137 const char *tmp_dir
= NULL
, *var_tmp_dir
= NULL
;
3138 const char *root_dir
= NULL
, *root_image
= NULL
;
3139 _cleanup_free_
char *creds_path
= NULL
, *incoming_dir
= NULL
, *propagate_dir
= NULL
;
3140 NamespaceInfo ns_info
;
3141 bool needs_sandboxing
;
3142 BindMount
*bind_mounts
= NULL
;
3143 size_t n_bind_mounts
= 0;
3148 if (params
->flags
& EXEC_APPLY_CHROOT
) {
3149 root_image
= context
->root_image
;
3152 root_dir
= context
->root_directory
;
3155 r
= compile_bind_mounts(context
, params
, &bind_mounts
, &n_bind_mounts
, &empty_directories
);
3159 needs_sandboxing
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && !(command_flags
& EXEC_COMMAND_FULLY_PRIVILEGED
);
3160 if (needs_sandboxing
) {
3161 /* The runtime struct only contains the parent of the private /tmp,
3162 * which is non-accessible to world users. Inside of it there's a /tmp
3163 * that is sticky, and that's the one we want to use here.
3164 * This does not apply when we are using /run/systemd/empty as fallback. */
3166 if (context
->private_tmp
&& runtime
) {
3167 if (streq_ptr(runtime
->tmp_dir
, RUN_SYSTEMD_EMPTY
))
3168 tmp_dir
= runtime
->tmp_dir
;
3169 else if (runtime
->tmp_dir
)
3170 tmp_dir
= strjoina(runtime
->tmp_dir
, "/tmp");
3172 if (streq_ptr(runtime
->var_tmp_dir
, RUN_SYSTEMD_EMPTY
))
3173 var_tmp_dir
= runtime
->var_tmp_dir
;
3174 else if (runtime
->var_tmp_dir
)
3175 var_tmp_dir
= strjoina(runtime
->var_tmp_dir
, "/tmp");
3178 ns_info
= (NamespaceInfo
) {
3179 .ignore_protect_paths
= false,
3180 .private_dev
= context
->private_devices
,
3181 .protect_control_groups
= context
->protect_control_groups
,
3182 .protect_kernel_tunables
= context
->protect_kernel_tunables
,
3183 .protect_kernel_modules
= context
->protect_kernel_modules
,
3184 .protect_kernel_logs
= context
->protect_kernel_logs
,
3185 .protect_hostname
= context
->protect_hostname
,
3186 .mount_apivfs
= exec_context_get_effective_mount_apivfs(context
),
3187 .private_mounts
= context
->private_mounts
,
3188 .protect_home
= context
->protect_home
,
3189 .protect_system
= context
->protect_system
,
3190 .protect_proc
= context
->protect_proc
,
3191 .proc_subset
= context
->proc_subset
,
3192 .private_ipc
= context
->private_ipc
|| context
->ipc_namespace_path
,
3194 } else if (!context
->dynamic_user
&& root_dir
)
3196 * If DynamicUser=no and RootDirectory= is set then lets pass a relaxed
3197 * sandbox info, otherwise enforce it, don't ignore protected paths and
3198 * fail if we are enable to apply the sandbox inside the mount namespace.
3200 ns_info
= (NamespaceInfo
) {
3201 .ignore_protect_paths
= true,
3204 ns_info
= (NamespaceInfo
) {};
3206 if (context
->mount_flags
== MS_SHARED
)
3207 log_unit_debug(u
, "shared mount propagation hidden by other fs namespacing unit settings: ignoring");
3209 if (exec_context_has_credentials(context
) &&
3210 params
->prefix
[EXEC_DIRECTORY_RUNTIME
] &&
3211 FLAGS_SET(params
->flags
, EXEC_WRITE_CREDENTIALS
)) {
3212 creds_path
= path_join(params
->prefix
[EXEC_DIRECTORY_RUNTIME
], "credentials", u
->id
);
3219 if (MANAGER_IS_SYSTEM(u
->manager
)) {
3220 propagate_dir
= path_join("/run/systemd/propagate/", u
->id
);
3221 if (!propagate_dir
) {
3226 incoming_dir
= strdup("/run/systemd/incoming");
3227 if (!incoming_dir
) {
3233 r
= setup_namespace(root_dir
, root_image
, context
->root_image_options
,
3234 &ns_info
, context
->read_write_paths
,
3235 needs_sandboxing
? context
->read_only_paths
: NULL
,
3236 needs_sandboxing
? context
->inaccessible_paths
: NULL
,
3237 needs_sandboxing
? context
->exec_paths
: NULL
,
3238 needs_sandboxing
? context
->no_exec_paths
: NULL
,
3242 context
->temporary_filesystems
,
3243 context
->n_temporary_filesystems
,
3244 context
->mount_images
,
3245 context
->n_mount_images
,
3249 context
->log_namespace
,
3250 context
->mount_flags
,
3251 context
->root_hash
, context
->root_hash_size
, context
->root_hash_path
,
3252 context
->root_hash_sig
, context
->root_hash_sig_size
, context
->root_hash_sig_path
,
3253 context
->root_verity
,
3254 context
->extension_images
,
3255 context
->n_extension_images
,
3258 root_dir
|| root_image
? params
->notify_socket
: NULL
,
3261 /* If we couldn't set up the namespace this is probably due to a missing capability. setup_namespace() reports
3262 * that with a special, recognizable error ENOANO. In this case, silently proceed, but only if exclusively
3263 * sandboxing options were used, i.e. nothing such as RootDirectory= or BindMount= that would result in a
3264 * completely different execution environment. */
3266 if (insist_on_sandboxing(
3268 root_dir
, root_image
,
3271 log_unit_debug(u
, "Failed to set up namespace, and refusing to continue since the selected namespacing options alter mount environment non-trivially.\n"
3272 "Bind mounts: %zu, temporary filesystems: %zu, root directory: %s, root image: %s, dynamic user: %s",
3273 n_bind_mounts
, context
->n_temporary_filesystems
, yes_no(root_dir
), yes_no(root_image
), yes_no(context
->dynamic_user
));
3277 log_unit_debug(u
, "Failed to set up namespace, assuming containerized execution and ignoring.");
3283 bind_mount_free_many(bind_mounts
, n_bind_mounts
);
3287 static int apply_working_directory(
3288 const ExecContext
*context
,
3289 const ExecParameters
*params
,
3296 assert(exit_status
);
3298 if (context
->working_directory_home
) {
3301 *exit_status
= EXIT_CHDIR
;
3308 wd
= empty_to_root(context
->working_directory
);
3310 if (params
->flags
& EXEC_APPLY_CHROOT
)
3313 d
= prefix_roota(context
->root_directory
, wd
);
3315 if (chdir(d
) < 0 && !context
->working_directory_missing_ok
) {
3316 *exit_status
= EXIT_CHDIR
;
3323 static int apply_root_directory(
3324 const ExecContext
*context
,
3325 const ExecParameters
*params
,
3326 const bool needs_mount_ns
,
3330 assert(exit_status
);
3332 if (params
->flags
& EXEC_APPLY_CHROOT
)
3333 if (!needs_mount_ns
&& context
->root_directory
)
3334 if (chroot(context
->root_directory
) < 0) {
3335 *exit_status
= EXIT_CHROOT
;
3342 static int setup_keyring(
3344 const ExecContext
*context
,
3345 const ExecParameters
*p
,
3346 uid_t uid
, gid_t gid
) {
3348 key_serial_t keyring
;
3357 /* Let's set up a new per-service "session" kernel keyring for each system service. This has the benefit that
3358 * each service runs with its own keyring shared among all processes of the service, but with no hook-up beyond
3359 * that scope, and in particular no link to the per-UID keyring. If we don't do this the keyring will be
3360 * automatically created on-demand and then linked to the per-UID keyring, by the kernel. The kernel's built-in
3361 * on-demand behaviour is very appropriate for login users, but probably not so much for system services, where
3362 * UIDs are not necessarily specific to a service but reused (at least in the case of UID 0). */
3364 if (context
->keyring_mode
== EXEC_KEYRING_INHERIT
)
3367 /* Acquiring a reference to the user keyring is nasty. We briefly change identity in order to get things set up
3368 * properly by the kernel. If we don't do that then we can't create it atomically, and that sucks for parallel
3369 * execution. This mimics what pam_keyinit does, too. Setting up session keyring, to be owned by the right user
3370 * & group is just as nasty as acquiring a reference to the user keyring. */
3372 saved_uid
= getuid();
3373 saved_gid
= getgid();
3375 if (gid_is_valid(gid
) && gid
!= saved_gid
) {
3376 if (setregid(gid
, -1) < 0)
3377 return log_unit_error_errno(u
, errno
, "Failed to change GID for user keyring: %m");
3380 if (uid_is_valid(uid
) && uid
!= saved_uid
) {
3381 if (setreuid(uid
, -1) < 0) {
3382 r
= log_unit_error_errno(u
, errno
, "Failed to change UID for user keyring: %m");
3387 keyring
= keyctl(KEYCTL_JOIN_SESSION_KEYRING
, 0, 0, 0, 0);
3388 if (keyring
== -1) {
3389 if (errno
== ENOSYS
)
3390 log_unit_debug_errno(u
, errno
, "Kernel keyring not supported, ignoring.");
3391 else if (ERRNO_IS_PRIVILEGE(errno
))
3392 log_unit_debug_errno(u
, errno
, "Kernel keyring access prohibited, ignoring.");
3393 else if (errno
== EDQUOT
)
3394 log_unit_debug_errno(u
, errno
, "Out of kernel keyrings to allocate, ignoring.");
3396 r
= log_unit_error_errno(u
, errno
, "Setting up kernel keyring failed: %m");
3401 /* When requested link the user keyring into the session keyring. */
3402 if (context
->keyring_mode
== EXEC_KEYRING_SHARED
) {
3404 if (keyctl(KEYCTL_LINK
,
3405 KEY_SPEC_USER_KEYRING
,
3406 KEY_SPEC_SESSION_KEYRING
, 0, 0) < 0) {
3407 r
= log_unit_error_errno(u
, errno
, "Failed to link user keyring into session keyring: %m");
3412 /* Restore uid/gid back */
3413 if (uid_is_valid(uid
) && uid
!= saved_uid
) {
3414 if (setreuid(saved_uid
, -1) < 0) {
3415 r
= log_unit_error_errno(u
, errno
, "Failed to change UID back for user keyring: %m");
3420 if (gid_is_valid(gid
) && gid
!= saved_gid
) {
3421 if (setregid(saved_gid
, -1) < 0)
3422 return log_unit_error_errno(u
, errno
, "Failed to change GID back for user keyring: %m");
3425 /* Populate they keyring with the invocation ID by default, as original saved_uid. */
3426 if (!sd_id128_is_null(u
->invocation_id
)) {
3429 key
= add_key("user", "invocation_id", &u
->invocation_id
, sizeof(u
->invocation_id
), KEY_SPEC_SESSION_KEYRING
);
3431 log_unit_debug_errno(u
, errno
, "Failed to add invocation ID to keyring, ignoring: %m");
3433 if (keyctl(KEYCTL_SETPERM
, key
,
3434 KEY_POS_VIEW
|KEY_POS_READ
|KEY_POS_SEARCH
|
3435 KEY_USR_VIEW
|KEY_USR_READ
|KEY_USR_SEARCH
, 0, 0) < 0)
3436 r
= log_unit_error_errno(u
, errno
, "Failed to restrict invocation ID permission: %m");
3441 /* Revert back uid & gid for the last time, and exit */
3442 /* no extra logging, as only the first already reported error matters */
3443 if (getuid() != saved_uid
)
3444 (void) setreuid(saved_uid
, -1);
3446 if (getgid() != saved_gid
)
3447 (void) setregid(saved_gid
, -1);
3452 static void append_socket_pair(int *array
, size_t *n
, const int pair
[static 2]) {
3458 array
[(*n
)++] = pair
[0];
3460 array
[(*n
)++] = pair
[1];
3463 static int close_remaining_fds(
3464 const ExecParameters
*params
,
3465 const ExecRuntime
*runtime
,
3466 const DynamicCreds
*dcreds
,
3469 const int *fds
, size_t n_fds
) {
3471 size_t n_dont_close
= 0;
3472 int dont_close
[n_fds
+ 12];
3476 if (params
->stdin_fd
>= 0)
3477 dont_close
[n_dont_close
++] = params
->stdin_fd
;
3478 if (params
->stdout_fd
>= 0)
3479 dont_close
[n_dont_close
++] = params
->stdout_fd
;
3480 if (params
->stderr_fd
>= 0)
3481 dont_close
[n_dont_close
++] = params
->stderr_fd
;
3484 dont_close
[n_dont_close
++] = socket_fd
;
3486 memcpy(dont_close
+ n_dont_close
, fds
, sizeof(int) * n_fds
);
3487 n_dont_close
+= n_fds
;
3491 append_socket_pair(dont_close
, &n_dont_close
, runtime
->netns_storage_socket
);
3492 append_socket_pair(dont_close
, &n_dont_close
, runtime
->ipcns_storage_socket
);
3497 append_socket_pair(dont_close
, &n_dont_close
, dcreds
->user
->storage_socket
);
3499 append_socket_pair(dont_close
, &n_dont_close
, dcreds
->group
->storage_socket
);
3502 if (user_lookup_fd
>= 0)
3503 dont_close
[n_dont_close
++] = user_lookup_fd
;
3505 return close_all_fds(dont_close
, n_dont_close
);
3508 static int send_user_lookup(
3516 /* Send the resolved UID/GID to PID 1 after we learnt it. We send a single datagram, containing the UID/GID
3517 * data as well as the unit name. Note that we suppress sending this if no user/group to resolve was
3520 if (user_lookup_fd
< 0)
3523 if (!uid_is_valid(uid
) && !gid_is_valid(gid
))
3526 if (writev(user_lookup_fd
,
3528 IOVEC_INIT(&uid
, sizeof(uid
)),
3529 IOVEC_INIT(&gid
, sizeof(gid
)),
3530 IOVEC_INIT_STRING(unit
->id
) }, 3) < 0)
3536 static int acquire_home(const ExecContext
*c
, uid_t uid
, const char** home
, char **buf
) {
3543 /* If WorkingDirectory=~ is set, try to acquire a usable home directory. */
3548 if (!c
->working_directory_home
)
3551 r
= get_home_dir(buf
);
3559 static int compile_suggested_paths(const ExecContext
*c
, const ExecParameters
*p
, char ***ret
) {
3560 _cleanup_strv_free_
char ** list
= NULL
;
3567 assert(c
->dynamic_user
);
3569 /* Compile a list of paths that it might make sense to read the owning UID from to use as initial candidate for
3570 * dynamic UID allocation, in order to save us from doing costly recursive chown()s of the special
3573 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
3576 if (t
== EXEC_DIRECTORY_CONFIGURATION
)
3582 STRV_FOREACH(i
, c
->directories
[t
].paths
) {
3585 if (exec_directory_is_private(c
, t
))
3586 e
= path_join(p
->prefix
[t
], "private", *i
);
3588 e
= path_join(p
->prefix
[t
], *i
);
3592 r
= strv_consume(&list
, e
);
3598 *ret
= TAKE_PTR(list
);
3603 static char *exec_command_line(char **argv
);
3605 static int exec_parameters_get_cgroup_path(const ExecParameters
*params
, char **ret
) {
3606 bool using_subcgroup
;
3612 if (!params
->cgroup_path
)
3615 /* If we are called for a unit where cgroup delegation is on, and the payload created its own populated
3616 * subcgroup (which we expect it to do, after all it asked for delegation), then we cannot place the control
3617 * processes started after the main unit's process in the unit's main cgroup because it is now an inner one,
3618 * and inner cgroups may not contain processes. Hence, if delegation is on, and this is a control process,
3619 * let's use ".control" as subcgroup instead. Note that we do so only for ExecStartPost=, ExecReload=,
3620 * ExecStop=, ExecStopPost=, i.e. for the commands where the main process is already forked. For ExecStartPre=
3621 * this is not necessary, the cgroup is still empty. We distinguish these cases with the EXEC_CONTROL_CGROUP
3622 * flag, which is only passed for the former statements, not for the latter. */
3624 using_subcgroup
= FLAGS_SET(params
->flags
, EXEC_CONTROL_CGROUP
|EXEC_CGROUP_DELEGATE
|EXEC_IS_CONTROL
);
3625 if (using_subcgroup
)
3626 p
= path_join(params
->cgroup_path
, ".control");
3628 p
= strdup(params
->cgroup_path
);
3633 return using_subcgroup
;
3636 static int exec_context_cpu_affinity_from_numa(const ExecContext
*c
, CPUSet
*ret
) {
3637 _cleanup_(cpu_set_reset
) CPUSet s
= {};
3643 if (!c
->numa_policy
.nodes
.set
) {
3644 log_debug("Can't derive CPU affinity mask from NUMA mask because NUMA mask is not set, ignoring");
3648 r
= numa_to_cpu_set(&c
->numa_policy
, &s
);
3654 return cpu_set_add_all(ret
, &s
);
3657 bool exec_context_get_cpu_affinity_from_numa(const ExecContext
*c
) {
3660 return c
->cpu_affinity_from_numa
;
3663 static int add_shifted_fd(int *fds
, size_t fds_size
, size_t *n_fds
, int fd
, int *ret_fd
) {
3668 assert(*n_fds
< fds_size
);
3676 if (fd
< 3 + (int) *n_fds
) {
3677 /* Let's move the fd up, so that it's outside of the fd range we will use to store
3678 * the fds we pass to the process (or which are closed only during execve). */
3680 r
= fcntl(fd
, F_DUPFD_CLOEXEC
, 3 + (int) *n_fds
);
3684 CLOSE_AND_REPLACE(fd
, r
);
3687 *ret_fd
= fds
[*n_fds
] = fd
;
3692 static int exec_child(
3694 const ExecCommand
*command
,
3695 const ExecContext
*context
,
3696 const ExecParameters
*params
,
3697 ExecRuntime
*runtime
,
3698 DynamicCreds
*dcreds
,
3700 const int named_iofds
[static 3],
3702 size_t n_socket_fds
,
3703 size_t n_storage_fds
,
3708 _cleanup_strv_free_
char **our_env
= NULL
, **pass_env
= NULL
, **accum_env
= NULL
, **replaced_argv
= NULL
;
3709 int r
, ngids
= 0, exec_fd
;
3710 _cleanup_free_ gid_t
*supplementary_gids
= NULL
;
3711 const char *username
= NULL
, *groupname
= NULL
;
3712 _cleanup_free_
char *home_buffer
= NULL
;
3713 const char *home
= NULL
, *shell
= NULL
;
3714 char **final_argv
= NULL
;
3715 dev_t journal_stream_dev
= 0;
3716 ino_t journal_stream_ino
= 0;
3717 bool userns_set_up
= false;
3718 bool needs_sandboxing
, /* Do we need to set up full sandboxing? (i.e. all namespacing, all MAC stuff, caps, yadda yadda */
3719 needs_setuid
, /* Do we need to do the actual setresuid()/setresgid() calls? */
3720 needs_mount_namespace
, /* Do we need to set up a mount namespace for this kernel? */
3721 needs_ambient_hack
; /* Do we need to apply the ambient capabilities hack? */
3723 _cleanup_free_
char *mac_selinux_context_net
= NULL
;
3724 bool use_selinux
= false;
3727 bool use_smack
= false;
3730 bool use_apparmor
= false;
3732 uid_t saved_uid
= getuid();
3733 gid_t saved_gid
= getgid();
3734 uid_t uid
= UID_INVALID
;
3735 gid_t gid
= GID_INVALID
;
3736 size_t n_fds
= n_socket_fds
+ n_storage_fds
, /* fds to pass to the child */
3737 n_keep_fds
; /* total number of fds not to close */
3739 _cleanup_free_ gid_t
*gids_after_pam
= NULL
;
3740 int ngids_after_pam
= 0;
3746 assert(exit_status
);
3748 rename_process_from_path(command
->path
);
3750 /* We reset exactly these signals, since they are the only ones we set to SIG_IGN in the main
3751 * daemon. All others we leave untouched because we set them to SIG_DFL or a valid handler initially,
3752 * both of which will be demoted to SIG_DFL. */
3753 (void) default_signals(SIGNALS_CRASH_HANDLER
,
3756 if (context
->ignore_sigpipe
)
3757 (void) ignore_signals(SIGPIPE
);
3759 r
= reset_signal_mask();
3761 *exit_status
= EXIT_SIGNAL_MASK
;
3762 return log_unit_error_errno(unit
, r
, "Failed to set process signal mask: %m");
3765 if (params
->idle_pipe
)
3766 do_idle_pipe_dance(params
->idle_pipe
);
3768 /* Close fds we don't need very early to make sure we don't block init reexecution because it cannot bind its
3769 * sockets. Among the fds we close are the logging fds, and we want to keep them closed, so that we don't have
3770 * any fds open we don't really want open during the transition. In order to make logging work, we switch the
3771 * log subsystem into open_when_needed mode, so that it reopens the logs on every single log call. */
3774 log_set_open_when_needed(true);
3776 /* In case anything used libc syslog(), close this here, too */
3779 int keep_fds
[n_fds
+ 2];
3780 memcpy_safe(keep_fds
, fds
, n_fds
* sizeof(int));
3783 r
= add_shifted_fd(keep_fds
, ELEMENTSOF(keep_fds
), &n_keep_fds
, params
->exec_fd
, &exec_fd
);
3785 *exit_status
= EXIT_FDS
;
3786 return log_unit_error_errno(unit
, r
, "Failed to shift fd and set FD_CLOEXEC: %m");
3789 r
= close_remaining_fds(params
, runtime
, dcreds
, user_lookup_fd
, socket_fd
, keep_fds
, n_keep_fds
);
3791 *exit_status
= EXIT_FDS
;
3792 return log_unit_error_errno(unit
, r
, "Failed to close unwanted file descriptors: %m");
3795 if (!context
->same_pgrp
&&
3797 *exit_status
= EXIT_SETSID
;
3798 return log_unit_error_errno(unit
, errno
, "Failed to create new process session: %m");
3801 exec_context_tty_reset(context
, params
);
3803 if (unit_shall_confirm_spawn(unit
)) {
3804 const char *vc
= params
->confirm_spawn
;
3805 _cleanup_free_
char *cmdline
= NULL
;
3807 cmdline
= exec_command_line(command
->argv
);
3809 *exit_status
= EXIT_MEMORY
;
3813 r
= ask_for_confirmation(vc
, unit
, cmdline
);
3814 if (r
!= CONFIRM_EXECUTE
) {
3815 if (r
== CONFIRM_PRETEND_SUCCESS
) {
3816 *exit_status
= EXIT_SUCCESS
;
3819 *exit_status
= EXIT_CONFIRM
;
3820 return log_unit_error_errno(unit
, SYNTHETIC_ERRNO(ECANCELED
),
3821 "Execution cancelled by the user");
3825 /* We are about to invoke NSS and PAM modules. Let's tell them what we are doing here, maybe they care. This is
3826 * used by nss-resolve to disable itself when we are about to start systemd-resolved, to avoid deadlocks. Note
3827 * that these env vars do not survive the execve(), which means they really only apply to the PAM and NSS
3828 * invocations themselves. Also note that while we'll only invoke NSS modules involved in user management they
3829 * might internally call into other NSS modules that are involved in hostname resolution, we never know. */
3830 if (setenv("SYSTEMD_ACTIVATION_UNIT", unit
->id
, true) != 0 ||
3831 setenv("SYSTEMD_ACTIVATION_SCOPE", MANAGER_IS_SYSTEM(unit
->manager
) ? "system" : "user", true) != 0) {
3832 *exit_status
= EXIT_MEMORY
;
3833 return log_unit_error_errno(unit
, errno
, "Failed to update environment: %m");
3836 if (context
->dynamic_user
&& dcreds
) {
3837 _cleanup_strv_free_
char **suggested_paths
= NULL
;
3839 /* On top of that, make sure we bypass our own NSS module nss-systemd comprehensively for any NSS
3840 * checks, if DynamicUser=1 is used, as we shouldn't create a feedback loop with ourselves here. */
3841 if (putenv((char*) "SYSTEMD_NSS_DYNAMIC_BYPASS=1") != 0) {
3842 *exit_status
= EXIT_USER
;
3843 return log_unit_error_errno(unit
, errno
, "Failed to update environment: %m");
3846 r
= compile_suggested_paths(context
, params
, &suggested_paths
);
3848 *exit_status
= EXIT_MEMORY
;
3852 r
= dynamic_creds_realize(dcreds
, suggested_paths
, &uid
, &gid
);
3854 *exit_status
= EXIT_USER
;
3856 return log_unit_error_errno(unit
, SYNTHETIC_ERRNO(EOPNOTSUPP
),
3857 "Failed to update dynamic user credentials: User or group with specified name already exists.");
3858 return log_unit_error_errno(unit
, r
, "Failed to update dynamic user credentials: %m");
3861 if (!uid_is_valid(uid
)) {
3862 *exit_status
= EXIT_USER
;
3863 return log_unit_error_errno(unit
, SYNTHETIC_ERRNO(ESRCH
), "UID validation failed for \""UID_FMT
"\"", uid
);
3866 if (!gid_is_valid(gid
)) {
3867 *exit_status
= EXIT_USER
;
3868 return log_unit_error_errno(unit
, SYNTHETIC_ERRNO(ESRCH
), "GID validation failed for \""GID_FMT
"\"", gid
);
3872 username
= dcreds
->user
->name
;
3875 r
= get_fixed_user(context
, &username
, &uid
, &gid
, &home
, &shell
);
3877 *exit_status
= EXIT_USER
;
3878 return log_unit_error_errno(unit
, r
, "Failed to determine user credentials: %m");
3881 r
= get_fixed_group(context
, &groupname
, &gid
);
3883 *exit_status
= EXIT_GROUP
;
3884 return log_unit_error_errno(unit
, r
, "Failed to determine group credentials: %m");
3888 /* Initialize user supplementary groups and get SupplementaryGroups= ones */
3889 r
= get_supplementary_groups(context
, username
, groupname
, gid
,
3890 &supplementary_gids
, &ngids
);
3892 *exit_status
= EXIT_GROUP
;
3893 return log_unit_error_errno(unit
, r
, "Failed to determine supplementary groups: %m");
3896 r
= send_user_lookup(unit
, user_lookup_fd
, uid
, gid
);
3898 *exit_status
= EXIT_USER
;
3899 return log_unit_error_errno(unit
, r
, "Failed to send user credentials to PID1: %m");
3902 user_lookup_fd
= safe_close(user_lookup_fd
);
3904 r
= acquire_home(context
, uid
, &home
, &home_buffer
);
3906 *exit_status
= EXIT_CHDIR
;
3907 return log_unit_error_errno(unit
, r
, "Failed to determine $HOME for user: %m");
3910 /* If a socket is connected to STDIN/STDOUT/STDERR, we
3911 * must sure to drop O_NONBLOCK */
3913 (void) fd_nonblock(socket_fd
, false);
3915 /* Journald will try to look-up our cgroup in order to populate _SYSTEMD_CGROUP and _SYSTEMD_UNIT fields.
3916 * Hence we need to migrate to the target cgroup from init.scope before connecting to journald */
3917 if (params
->cgroup_path
) {
3918 _cleanup_free_
char *p
= NULL
;
3920 r
= exec_parameters_get_cgroup_path(params
, &p
);
3922 *exit_status
= EXIT_CGROUP
;
3923 return log_unit_error_errno(unit
, r
, "Failed to acquire cgroup path: %m");
3926 r
= cg_attach_everywhere(params
->cgroup_supported
, p
, 0, NULL
, NULL
);
3928 *exit_status
= EXIT_CGROUP
;
3929 return log_unit_error_errno(unit
, r
, "Failed to attach to cgroup %s: %m", p
);
3933 if (context
->network_namespace_path
&& runtime
&& runtime
->netns_storage_socket
[0] >= 0) {
3934 r
= open_shareable_ns_path(runtime
->netns_storage_socket
, context
->network_namespace_path
, CLONE_NEWNET
);
3936 *exit_status
= EXIT_NETWORK
;
3937 return log_unit_error_errno(unit
, r
, "Failed to open network namespace path %s: %m", context
->network_namespace_path
);
3941 if (context
->ipc_namespace_path
&& runtime
&& runtime
->ipcns_storage_socket
[0] >= 0) {
3942 r
= open_shareable_ns_path(runtime
->ipcns_storage_socket
, context
->ipc_namespace_path
, CLONE_NEWIPC
);
3944 *exit_status
= EXIT_NAMESPACE
;
3945 return log_unit_error_errno(unit
, r
, "Failed to open IPC namespace path %s: %m", context
->ipc_namespace_path
);
3949 r
= setup_input(context
, params
, socket_fd
, named_iofds
);
3951 *exit_status
= EXIT_STDIN
;
3952 return log_unit_error_errno(unit
, r
, "Failed to set up standard input: %m");
3955 r
= setup_output(unit
, context
, params
, STDOUT_FILENO
, socket_fd
, named_iofds
, basename(command
->path
), uid
, gid
, &journal_stream_dev
, &journal_stream_ino
);
3957 *exit_status
= EXIT_STDOUT
;
3958 return log_unit_error_errno(unit
, r
, "Failed to set up standard output: %m");
3961 r
= setup_output(unit
, context
, params
, STDERR_FILENO
, socket_fd
, named_iofds
, basename(command
->path
), uid
, gid
, &journal_stream_dev
, &journal_stream_ino
);
3963 *exit_status
= EXIT_STDERR
;
3964 return log_unit_error_errno(unit
, r
, "Failed to set up standard error output: %m");
3967 if (context
->oom_score_adjust_set
) {
3968 /* When we can't make this change due to EPERM, then let's silently skip over it. User namespaces
3969 * prohibit write access to this file, and we shouldn't trip up over that. */
3970 r
= set_oom_score_adjust(context
->oom_score_adjust
);
3971 if (ERRNO_IS_PRIVILEGE(r
))
3972 log_unit_debug_errno(unit
, r
, "Failed to adjust OOM setting, assuming containerized execution, ignoring: %m");
3974 *exit_status
= EXIT_OOM_ADJUST
;
3975 return log_unit_error_errno(unit
, r
, "Failed to adjust OOM setting: %m");
3979 if (context
->coredump_filter_set
) {
3980 r
= set_coredump_filter(context
->coredump_filter
);
3981 if (ERRNO_IS_PRIVILEGE(r
))
3982 log_unit_debug_errno(unit
, r
, "Failed to adjust coredump_filter, ignoring: %m");
3984 return log_unit_error_errno(unit
, r
, "Failed to adjust coredump_filter: %m");
3987 if (context
->nice_set
) {
3988 r
= setpriority_closest(context
->nice
);
3990 return log_unit_error_errno(unit
, r
, "Failed to set up process scheduling priority (nice level): %m");
3993 if (context
->cpu_sched_set
) {
3994 struct sched_param param
= {
3995 .sched_priority
= context
->cpu_sched_priority
,
3998 r
= sched_setscheduler(0,
3999 context
->cpu_sched_policy
|
4000 (context
->cpu_sched_reset_on_fork
?
4001 SCHED_RESET_ON_FORK
: 0),
4004 *exit_status
= EXIT_SETSCHEDULER
;
4005 return log_unit_error_errno(unit
, errno
, "Failed to set up CPU scheduling: %m");
4009 if (context
->cpu_affinity_from_numa
|| context
->cpu_set
.set
) {
4010 _cleanup_(cpu_set_reset
) CPUSet converted_cpu_set
= {};
4011 const CPUSet
*cpu_set
;
4013 if (context
->cpu_affinity_from_numa
) {
4014 r
= exec_context_cpu_affinity_from_numa(context
, &converted_cpu_set
);
4016 *exit_status
= EXIT_CPUAFFINITY
;
4017 return log_unit_error_errno(unit
, r
, "Failed to derive CPU affinity mask from NUMA mask: %m");
4020 cpu_set
= &converted_cpu_set
;
4022 cpu_set
= &context
->cpu_set
;
4024 if (sched_setaffinity(0, cpu_set
->allocated
, cpu_set
->set
) < 0) {
4025 *exit_status
= EXIT_CPUAFFINITY
;
4026 return log_unit_error_errno(unit
, errno
, "Failed to set up CPU affinity: %m");
4030 if (mpol_is_valid(numa_policy_get_type(&context
->numa_policy
))) {
4031 r
= apply_numa_policy(&context
->numa_policy
);
4032 if (r
== -EOPNOTSUPP
)
4033 log_unit_debug_errno(unit
, r
, "NUMA support not available, ignoring.");
4035 *exit_status
= EXIT_NUMA_POLICY
;
4036 return log_unit_error_errno(unit
, r
, "Failed to set NUMA memory policy: %m");
4040 if (context
->ioprio_set
)
4041 if (ioprio_set(IOPRIO_WHO_PROCESS
, 0, context
->ioprio
) < 0) {
4042 *exit_status
= EXIT_IOPRIO
;
4043 return log_unit_error_errno(unit
, errno
, "Failed to set up IO scheduling priority: %m");
4046 if (context
->timer_slack_nsec
!= NSEC_INFINITY
)
4047 if (prctl(PR_SET_TIMERSLACK
, context
->timer_slack_nsec
) < 0) {
4048 *exit_status
= EXIT_TIMERSLACK
;
4049 return log_unit_error_errno(unit
, errno
, "Failed to set up timer slack: %m");
4052 if (context
->personality
!= PERSONALITY_INVALID
) {
4053 r
= safe_personality(context
->personality
);
4055 *exit_status
= EXIT_PERSONALITY
;
4056 return log_unit_error_errno(unit
, r
, "Failed to set up execution domain (personality): %m");
4060 if (context
->utmp_id
)
4061 utmp_put_init_process(context
->utmp_id
, getpid_cached(), getsid(0),
4063 context
->utmp_mode
== EXEC_UTMP_INIT
? INIT_PROCESS
:
4064 context
->utmp_mode
== EXEC_UTMP_LOGIN
? LOGIN_PROCESS
:
4068 if (uid_is_valid(uid
)) {
4069 r
= chown_terminal(STDIN_FILENO
, uid
);
4071 *exit_status
= EXIT_STDIN
;
4072 return log_unit_error_errno(unit
, r
, "Failed to change ownership of terminal: %m");
4076 /* If delegation is enabled we'll pass ownership of the cgroup to the user of the new process. On cgroup v1
4077 * this is only about systemd's own hierarchy, i.e. not the controller hierarchies, simply because that's not
4078 * safe. On cgroup v2 there's only one hierarchy anyway, and delegation is safe there, hence in that case only
4079 * touch a single hierarchy too. */
4080 if (params
->cgroup_path
&& context
->user
&& (params
->flags
& EXEC_CGROUP_DELEGATE
)) {
4081 r
= cg_set_access(SYSTEMD_CGROUP_CONTROLLER
, params
->cgroup_path
, uid
, gid
);
4083 *exit_status
= EXIT_CGROUP
;
4084 return log_unit_error_errno(unit
, r
, "Failed to adjust control group access: %m");
4088 for (ExecDirectoryType dt
= 0; dt
< _EXEC_DIRECTORY_TYPE_MAX
; dt
++) {
4089 r
= setup_exec_directory(context
, params
, uid
, gid
, dt
, exit_status
);
4091 return log_unit_error_errno(unit
, r
, "Failed to set up special execution directory in %s: %m", params
->prefix
[dt
]);
4094 if (FLAGS_SET(params
->flags
, EXEC_WRITE_CREDENTIALS
)) {
4095 r
= setup_credentials(context
, params
, unit
->id
, uid
);
4097 *exit_status
= EXIT_CREDENTIALS
;
4098 return log_unit_error_errno(unit
, r
, "Failed to set up credentials: %m");
4102 r
= build_environment(
4114 *exit_status
= EXIT_MEMORY
;
4118 r
= build_pass_environment(context
, &pass_env
);
4120 *exit_status
= EXIT_MEMORY
;
4124 accum_env
= strv_env_merge(5,
4125 params
->environment
,
4128 context
->environment
,
4131 *exit_status
= EXIT_MEMORY
;
4134 accum_env
= strv_env_clean(accum_env
);
4136 (void) umask(context
->umask
);
4138 r
= setup_keyring(unit
, context
, params
, uid
, gid
);
4140 *exit_status
= EXIT_KEYRING
;
4141 return log_unit_error_errno(unit
, r
, "Failed to set up kernel keyring: %m");
4144 /* We need sandboxing if the caller asked us to apply it and the command isn't explicitly excepted from it */
4145 needs_sandboxing
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && !(command
->flags
& EXEC_COMMAND_FULLY_PRIVILEGED
);
4147 /* 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 */
4148 needs_ambient_hack
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && (command
->flags
& EXEC_COMMAND_AMBIENT_MAGIC
) && !ambient_capabilities_supported();
4150 /* 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 */
4151 if (needs_ambient_hack
)
4152 needs_setuid
= false;
4154 needs_setuid
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && !(command
->flags
& (EXEC_COMMAND_FULLY_PRIVILEGED
|EXEC_COMMAND_NO_SETUID
));
4156 if (needs_sandboxing
) {
4157 /* MAC enablement checks need to be done before a new mount ns is created, as they rely on /sys being
4158 * present. The actual MAC context application will happen later, as late as possible, to avoid
4159 * impacting our own code paths. */
4162 use_selinux
= mac_selinux_use();
4165 use_smack
= mac_smack_use();
4168 use_apparmor
= mac_apparmor_use();
4172 if (needs_sandboxing
) {
4175 /* Let's set the resource limits before we call into PAM, so that pam_limits wins over what
4176 * is set here. (See below.) */
4178 r
= setrlimit_closest_all((const struct rlimit
* const *) context
->rlimit
, &which_failed
);
4180 *exit_status
= EXIT_LIMITS
;
4181 return log_unit_error_errno(unit
, r
, "Failed to adjust resource limit RLIMIT_%s: %m", rlimit_to_string(which_failed
));
4185 if (needs_setuid
&& context
->pam_name
&& username
) {
4186 /* Let's call into PAM after we set up our own idea of resource limits to that pam_limits
4187 * wins here. (See above.) */
4189 /* All fds passed in the fds array will be closed in the pam child process. */
4190 r
= setup_pam(context
->pam_name
, username
, uid
, gid
, context
->tty_path
, &accum_env
, fds
, n_fds
);
4192 *exit_status
= EXIT_PAM
;
4193 return log_unit_error_errno(unit
, r
, "Failed to set up PAM session: %m");
4196 ngids_after_pam
= getgroups_alloc(&gids_after_pam
);
4197 if (ngids_after_pam
< 0) {
4198 *exit_status
= EXIT_MEMORY
;
4199 return log_unit_error_errno(unit
, ngids_after_pam
, "Failed to obtain groups after setting up PAM: %m");
4203 if (needs_sandboxing
&& context
->private_users
&& !have_effective_cap(CAP_SYS_ADMIN
)) {
4204 /* If we're unprivileged, set up the user namespace first to enable use of the other namespaces.
4205 * Users with CAP_SYS_ADMIN can set up user namespaces last because they will be able to
4206 * set up the all of the other namespaces (i.e. network, mount, UTS) without a user namespace. */
4208 userns_set_up
= true;
4209 r
= setup_private_users(saved_uid
, saved_gid
, uid
, gid
);
4211 *exit_status
= EXIT_USER
;
4212 return log_unit_error_errno(unit
, r
, "Failed to set up user namespacing for unprivileged user: %m");
4216 if ((context
->private_network
|| context
->network_namespace_path
) && runtime
&& runtime
->netns_storage_socket
[0] >= 0) {
4218 if (ns_type_supported(NAMESPACE_NET
)) {
4219 r
= setup_shareable_ns(runtime
->netns_storage_socket
, CLONE_NEWNET
);
4221 log_unit_warning_errno(unit
, r
,
4222 "PrivateNetwork=yes is configured, but network namespace setup failed, ignoring: %m");
4224 *exit_status
= EXIT_NETWORK
;
4225 return log_unit_error_errno(unit
, r
, "Failed to set up network namespacing: %m");
4227 } else if (context
->network_namespace_path
) {
4228 *exit_status
= EXIT_NETWORK
;
4229 return log_unit_error_errno(unit
, SYNTHETIC_ERRNO(EOPNOTSUPP
),
4230 "NetworkNamespacePath= is not supported, refusing.");
4232 log_unit_warning(unit
, "PrivateNetwork=yes is configured, but the kernel does not support network namespaces, ignoring.");
4235 if ((context
->private_ipc
|| context
->ipc_namespace_path
) && runtime
&& runtime
->ipcns_storage_socket
[0] >= 0) {
4237 if (ns_type_supported(NAMESPACE_IPC
)) {
4238 r
= setup_shareable_ns(runtime
->ipcns_storage_socket
, CLONE_NEWIPC
);
4240 log_unit_warning_errno(unit
, r
,
4241 "PrivateIPC=yes is configured, but IPC namespace setup failed, ignoring: %m");
4243 *exit_status
= EXIT_NAMESPACE
;
4244 return log_unit_error_errno(unit
, r
, "Failed to set up IPC namespacing: %m");
4246 } else if (context
->ipc_namespace_path
) {
4247 *exit_status
= EXIT_NAMESPACE
;
4248 return log_unit_error_errno(unit
, SYNTHETIC_ERRNO(EOPNOTSUPP
),
4249 "IPCNamespacePath= is not supported, refusing.");
4251 log_unit_warning(unit
, "PrivateIPC=yes is configured, but the kernel does not support IPC namespaces, ignoring.");
4254 needs_mount_namespace
= exec_needs_mount_namespace(context
, params
, runtime
);
4255 if (needs_mount_namespace
) {
4256 _cleanup_free_
char *error_path
= NULL
;
4258 r
= apply_mount_namespace(unit
, command
->flags
, context
, params
, runtime
, &error_path
);
4260 *exit_status
= EXIT_NAMESPACE
;
4261 return log_unit_error_errno(unit
, r
, "Failed to set up mount namespacing%s%s: %m",
4262 error_path
? ": " : "", strempty(error_path
));
4266 if (needs_sandboxing
) {
4267 r
= apply_protect_hostname(unit
, context
, exit_status
);
4272 /* Drop groups as early as possible.
4273 * This needs to be done after PrivateDevices=y setup as device nodes should be owned by the host's root.
4274 * For non-root in a userns, devices will be owned by the user/group before the group change, and nobody. */
4276 _cleanup_free_ gid_t
*gids_to_enforce
= NULL
;
4277 int ngids_to_enforce
= 0;
4279 ngids_to_enforce
= merge_gid_lists(supplementary_gids
,
4284 if (ngids_to_enforce
< 0) {
4285 *exit_status
= EXIT_MEMORY
;
4286 return log_unit_error_errno(unit
,
4288 "Failed to merge group lists. Group membership might be incorrect: %m");
4291 r
= enforce_groups(gid
, gids_to_enforce
, ngids_to_enforce
);
4293 *exit_status
= EXIT_GROUP
;
4294 return log_unit_error_errno(unit
, r
, "Changing group credentials failed: %m");
4298 /* If the user namespace was not set up above, try to do it now.
4299 * It's preferred to set up the user namespace later (after all other namespaces) so as not to be
4300 * restricted by rules pertaining to combining user namspaces with other namespaces (e.g. in the
4301 * case of mount namespaces being less privileged when the mount point list is copied from a
4302 * different user namespace). */
4304 if (needs_sandboxing
&& context
->private_users
&& !userns_set_up
) {
4305 r
= setup_private_users(saved_uid
, saved_gid
, uid
, gid
);
4307 *exit_status
= EXIT_USER
;
4308 return log_unit_error_errno(unit
, r
, "Failed to set up user namespacing: %m");
4312 /* Now that the mount namespace has been set up and privileges adjusted, let's look for the thing we
4315 _cleanup_free_
char *executable
= NULL
;
4316 _cleanup_close_
int executable_fd
= -1;
4317 r
= find_executable_full(command
->path
, false, &executable
, &executable_fd
);
4319 if (r
!= -ENOMEM
&& (command
->flags
& EXEC_COMMAND_IGNORE_FAILURE
)) {
4320 log_unit_struct_errno(unit
, LOG_INFO
, r
,
4321 "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR
,
4322 LOG_UNIT_INVOCATION_ID(unit
),
4323 LOG_UNIT_MESSAGE(unit
, "Executable %s missing, skipping: %m",
4325 "EXECUTABLE=%s", command
->path
);
4329 *exit_status
= EXIT_EXEC
;
4331 return log_unit_struct_errno(unit
, LOG_INFO
, r
,
4332 "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR
,
4333 LOG_UNIT_INVOCATION_ID(unit
),
4334 LOG_UNIT_MESSAGE(unit
, "Failed to locate executable %s: %m",
4336 "EXECUTABLE=%s", command
->path
);
4339 r
= add_shifted_fd(keep_fds
, ELEMENTSOF(keep_fds
), &n_keep_fds
, executable_fd
, &executable_fd
);
4341 *exit_status
= EXIT_FDS
;
4342 return log_unit_error_errno(unit
, r
, "Failed to shift fd and set FD_CLOEXEC: %m");
4346 if (needs_sandboxing
&& use_selinux
&& params
->selinux_context_net
&& socket_fd
>= 0) {
4347 r
= mac_selinux_get_child_mls_label(socket_fd
, executable
, context
->selinux_context
, &mac_selinux_context_net
);
4349 *exit_status
= EXIT_SELINUX_CONTEXT
;
4350 return log_unit_error_errno(unit
, r
, "Failed to determine SELinux context: %m");
4355 /* We repeat the fd closing here, to make sure that nothing is leaked from the PAM modules. Note that we are
4356 * more aggressive this time since socket_fd and the netns and ipcns fds we don't need anymore. We do keep the exec_fd
4357 * however if we have it as we want to keep it open until the final execve(). */
4359 r
= close_all_fds(keep_fds
, n_keep_fds
);
4361 r
= shift_fds(fds
, n_fds
);
4363 r
= flags_fds(fds
, n_socket_fds
, n_storage_fds
, context
->non_blocking
);
4365 *exit_status
= EXIT_FDS
;
4366 return log_unit_error_errno(unit
, r
, "Failed to adjust passed file descriptors: %m");
4369 /* At this point, the fds we want to pass to the program are all ready and set up, with O_CLOEXEC turned off
4370 * and at the right fd numbers. The are no other fds open, with one exception: the exec_fd if it is defined,
4371 * and it has O_CLOEXEC set, after all we want it to be closed by the execve(), so that our parent knows we
4374 secure_bits
= context
->secure_bits
;
4376 if (needs_sandboxing
) {
4379 /* Set the RTPRIO resource limit to 0, but only if nothing else was explicitly
4380 * requested. (Note this is placed after the general resource limit initialization, see
4381 * above, in order to take precedence.) */
4382 if (context
->restrict_realtime
&& !context
->rlimit
[RLIMIT_RTPRIO
]) {
4383 if (setrlimit(RLIMIT_RTPRIO
, &RLIMIT_MAKE_CONST(0)) < 0) {
4384 *exit_status
= EXIT_LIMITS
;
4385 return log_unit_error_errno(unit
, errno
, "Failed to adjust RLIMIT_RTPRIO resource limit: %m");
4390 /* LSM Smack needs the capability CAP_MAC_ADMIN to change the current execution security context of the
4391 * process. This is the latest place before dropping capabilities. Other MAC context are set later. */
4393 r
= setup_smack(context
, executable_fd
);
4395 *exit_status
= EXIT_SMACK_PROCESS_LABEL
;
4396 return log_unit_error_errno(unit
, r
, "Failed to set SMACK process label: %m");
4401 bset
= context
->capability_bounding_set
;
4402 /* If the ambient caps hack is enabled (which means the kernel can't do them, and the user asked for
4403 * our magic fallback), then let's add some extra caps, so that the service can drop privs of its own,
4404 * instead of us doing that */
4405 if (needs_ambient_hack
)
4406 bset
|= (UINT64_C(1) << CAP_SETPCAP
) |
4407 (UINT64_C(1) << CAP_SETUID
) |
4408 (UINT64_C(1) << CAP_SETGID
);
4410 if (!cap_test_all(bset
)) {
4411 r
= capability_bounding_set_drop(bset
, false);
4413 *exit_status
= EXIT_CAPABILITIES
;
4414 return log_unit_error_errno(unit
, r
, "Failed to drop capabilities: %m");
4418 /* Ambient capabilities are cleared during setresuid() (in enforce_user()) even with
4420 * To be able to raise the ambient capabilities after setresuid() they have to be
4421 * added to the inherited set and keep caps has to be set (done in enforce_user()).
4422 * After setresuid() the ambient capabilities can be raised as they are present in
4423 * the permitted and inhertiable set. However it is possible that someone wants to
4424 * set ambient capabilities without changing the user, so we also set the ambient
4425 * capabilities here.
4426 * The requested ambient capabilities are raised in the inheritable set if the
4427 * second argument is true. */
4428 if (!needs_ambient_hack
) {
4429 r
= capability_ambient_set_apply(context
->capability_ambient_set
, true);
4431 *exit_status
= EXIT_CAPABILITIES
;
4432 return log_unit_error_errno(unit
, r
, "Failed to apply ambient capabilities (before UID change): %m");
4437 /* chroot to root directory first, before we lose the ability to chroot */
4438 r
= apply_root_directory(context
, params
, needs_mount_namespace
, exit_status
);
4440 return log_unit_error_errno(unit
, r
, "Chrooting to the requested root directory failed: %m");
4443 if (uid_is_valid(uid
)) {
4444 r
= enforce_user(context
, uid
);
4446 *exit_status
= EXIT_USER
;
4447 return log_unit_error_errno(unit
, r
, "Failed to change UID to " UID_FMT
": %m", uid
);
4450 if (!needs_ambient_hack
&&
4451 context
->capability_ambient_set
!= 0) {
4453 /* Raise the ambient capabilities after user change. */
4454 r
= capability_ambient_set_apply(context
->capability_ambient_set
, false);
4456 *exit_status
= EXIT_CAPABILITIES
;
4457 return log_unit_error_errno(unit
, r
, "Failed to apply ambient capabilities (after UID change): %m");
4463 /* Apply working directory here, because the working directory might be on NFS and only the user running
4464 * this service might have the correct privilege to change to the working directory */
4465 r
= apply_working_directory(context
, params
, home
, exit_status
);
4467 return log_unit_error_errno(unit
, r
, "Changing to the requested working directory failed: %m");
4469 if (needs_sandboxing
) {
4470 /* Apply other MAC contexts late, but before seccomp syscall filtering, as those should really be last to
4471 * influence our own codepaths as little as possible. Moreover, applying MAC contexts usually requires
4472 * syscalls that are subject to seccomp filtering, hence should probably be applied before the syscalls
4473 * are restricted. */
4477 char *exec_context
= mac_selinux_context_net
?: context
->selinux_context
;
4480 r
= setexeccon(exec_context
);
4482 *exit_status
= EXIT_SELINUX_CONTEXT
;
4483 return log_unit_error_errno(unit
, r
, "Failed to change SELinux context to %s: %m", exec_context
);
4490 if (use_apparmor
&& context
->apparmor_profile
) {
4491 r
= aa_change_onexec(context
->apparmor_profile
);
4492 if (r
< 0 && !context
->apparmor_profile_ignore
) {
4493 *exit_status
= EXIT_APPARMOR_PROFILE
;
4494 return log_unit_error_errno(unit
, errno
, "Failed to prepare AppArmor profile change to %s: %m", context
->apparmor_profile
);
4499 /* PR_GET_SECUREBITS is not privileged, while PR_SET_SECUREBITS is. So to suppress potential EPERMs
4500 * we'll try not to call PR_SET_SECUREBITS unless necessary. Setting securebits requires
4502 if (prctl(PR_GET_SECUREBITS
) != secure_bits
) {
4503 /* CAP_SETPCAP is required to set securebits. This capability is raised into the
4504 * effective set here.
4505 * The effective set is overwritten during execve with the following values:
4506 * - ambient set (for non-root processes)
4507 * - (inheritable | bounding) set for root processes)
4509 * Hence there is no security impact to raise it in the effective set before execve
4511 r
= capability_gain_cap_setpcap(NULL
);
4513 *exit_status
= EXIT_CAPABILITIES
;
4514 return log_unit_error_errno(unit
, r
, "Failed to gain CAP_SETPCAP for setting secure bits");
4516 if (prctl(PR_SET_SECUREBITS
, secure_bits
) < 0) {
4517 *exit_status
= EXIT_SECUREBITS
;
4518 return log_unit_error_errno(unit
, errno
, "Failed to set process secure bits: %m");
4522 if (context_has_no_new_privileges(context
))
4523 if (prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0) < 0) {
4524 *exit_status
= EXIT_NO_NEW_PRIVILEGES
;
4525 return log_unit_error_errno(unit
, errno
, "Failed to disable new privileges: %m");
4529 r
= apply_address_families(unit
, context
);
4531 *exit_status
= EXIT_ADDRESS_FAMILIES
;
4532 return log_unit_error_errno(unit
, r
, "Failed to restrict address families: %m");
4535 r
= apply_memory_deny_write_execute(unit
, context
);
4537 *exit_status
= EXIT_SECCOMP
;
4538 return log_unit_error_errno(unit
, r
, "Failed to disable writing to executable memory: %m");
4541 r
= apply_restrict_realtime(unit
, context
);
4543 *exit_status
= EXIT_SECCOMP
;
4544 return log_unit_error_errno(unit
, r
, "Failed to apply realtime restrictions: %m");
4547 r
= apply_restrict_suid_sgid(unit
, context
);
4549 *exit_status
= EXIT_SECCOMP
;
4550 return log_unit_error_errno(unit
, r
, "Failed to apply SUID/SGID restrictions: %m");
4553 r
= apply_restrict_namespaces(unit
, context
);
4555 *exit_status
= EXIT_SECCOMP
;
4556 return log_unit_error_errno(unit
, r
, "Failed to apply namespace restrictions: %m");
4559 r
= apply_protect_sysctl(unit
, context
);
4561 *exit_status
= EXIT_SECCOMP
;
4562 return log_unit_error_errno(unit
, r
, "Failed to apply sysctl restrictions: %m");
4565 r
= apply_protect_kernel_modules(unit
, context
);
4567 *exit_status
= EXIT_SECCOMP
;
4568 return log_unit_error_errno(unit
, r
, "Failed to apply module loading restrictions: %m");
4571 r
= apply_protect_kernel_logs(unit
, context
);
4573 *exit_status
= EXIT_SECCOMP
;
4574 return log_unit_error_errno(unit
, r
, "Failed to apply kernel log restrictions: %m");
4577 r
= apply_protect_clock(unit
, context
);
4579 *exit_status
= EXIT_SECCOMP
;
4580 return log_unit_error_errno(unit
, r
, "Failed to apply clock restrictions: %m");
4583 r
= apply_private_devices(unit
, context
);
4585 *exit_status
= EXIT_SECCOMP
;
4586 return log_unit_error_errno(unit
, r
, "Failed to set up private devices: %m");
4589 r
= apply_syscall_archs(unit
, context
);
4591 *exit_status
= EXIT_SECCOMP
;
4592 return log_unit_error_errno(unit
, r
, "Failed to apply syscall architecture restrictions: %m");
4595 r
= apply_lock_personality(unit
, context
);
4597 *exit_status
= EXIT_SECCOMP
;
4598 return log_unit_error_errno(unit
, r
, "Failed to lock personalities: %m");
4601 r
= apply_syscall_log(unit
, context
);
4603 *exit_status
= EXIT_SECCOMP
;
4604 return log_unit_error_errno(unit
, r
, "Failed to apply system call log filters: %m");
4607 /* This really should remain the last step before the execve(), to make sure our own code is unaffected
4608 * by the filter as little as possible. */
4609 r
= apply_syscall_filter(unit
, context
, needs_ambient_hack
);
4611 *exit_status
= EXIT_SECCOMP
;
4612 return log_unit_error_errno(unit
, r
, "Failed to apply system call filters: %m");
4617 if (!strv_isempty(context
->unset_environment
)) {
4620 ee
= strv_env_delete(accum_env
, 1, context
->unset_environment
);
4622 *exit_status
= EXIT_MEMORY
;
4626 strv_free_and_replace(accum_env
, ee
);
4629 if (!FLAGS_SET(command
->flags
, EXEC_COMMAND_NO_ENV_EXPAND
)) {
4630 replaced_argv
= replace_env_argv(command
->argv
, accum_env
);
4631 if (!replaced_argv
) {
4632 *exit_status
= EXIT_MEMORY
;
4635 final_argv
= replaced_argv
;
4637 final_argv
= command
->argv
;
4639 if (DEBUG_LOGGING
) {
4640 _cleanup_free_
char *line
= NULL
;
4642 line
= exec_command_line(final_argv
);
4644 log_unit_struct(unit
, LOG_DEBUG
,
4645 "EXECUTABLE=%s", executable
,
4646 LOG_UNIT_MESSAGE(unit
, "Executing: %s", line
),
4647 LOG_UNIT_INVOCATION_ID(unit
));
4653 /* We have finished with all our initializations. Let's now let the manager know that. From this point
4654 * on, if the manager sees POLLHUP on the exec_fd, then execve() was successful. */
4656 if (write(exec_fd
, &hot
, sizeof(hot
)) < 0) {
4657 *exit_status
= EXIT_EXEC
;
4658 return log_unit_error_errno(unit
, errno
, "Failed to enable exec_fd: %m");
4662 r
= fexecve_or_execve(executable_fd
, executable
, final_argv
, accum_env
);
4667 /* The execve() failed. This means the exec_fd is still open. Which means we need to tell the manager
4668 * that POLLHUP on it no longer means execve() succeeded. */
4670 if (write(exec_fd
, &hot
, sizeof(hot
)) < 0) {
4671 *exit_status
= EXIT_EXEC
;
4672 return log_unit_error_errno(unit
, errno
, "Failed to disable exec_fd: %m");
4676 *exit_status
= EXIT_EXEC
;
4677 return log_unit_error_errno(unit
, r
, "Failed to execute %s: %m", executable
);
4680 static int exec_context_load_environment(const Unit
*unit
, const ExecContext
*c
, char ***l
);
4681 static int exec_context_named_iofds(const ExecContext
*c
, const ExecParameters
*p
, int named_iofds
[static 3]);
4683 int exec_spawn(Unit
*unit
,
4684 ExecCommand
*command
,
4685 const ExecContext
*context
,
4686 const ExecParameters
*params
,
4687 ExecRuntime
*runtime
,
4688 DynamicCreds
*dcreds
,
4691 int socket_fd
, r
, named_iofds
[3] = { -1, -1, -1 }, *fds
= NULL
;
4692 _cleanup_free_
char *subcgroup_path
= NULL
;
4693 _cleanup_strv_free_
char **files_env
= NULL
;
4694 size_t n_storage_fds
= 0, n_socket_fds
= 0;
4695 _cleanup_free_
char *line
= NULL
;
4703 assert(params
->fds
|| (params
->n_socket_fds
+ params
->n_storage_fds
<= 0));
4705 if (context
->std_input
== EXEC_INPUT_SOCKET
||
4706 context
->std_output
== EXEC_OUTPUT_SOCKET
||
4707 context
->std_error
== EXEC_OUTPUT_SOCKET
) {
4709 if (params
->n_socket_fds
> 1)
4710 return log_unit_error_errno(unit
, SYNTHETIC_ERRNO(EINVAL
), "Got more than one socket.");
4712 if (params
->n_socket_fds
== 0)
4713 return log_unit_error_errno(unit
, SYNTHETIC_ERRNO(EINVAL
), "Got no socket.");
4715 socket_fd
= params
->fds
[0];
4719 n_socket_fds
= params
->n_socket_fds
;
4720 n_storage_fds
= params
->n_storage_fds
;
4723 r
= exec_context_named_iofds(context
, params
, named_iofds
);
4725 return log_unit_error_errno(unit
, r
, "Failed to load a named file descriptor: %m");
4727 r
= exec_context_load_environment(unit
, context
, &files_env
);
4729 return log_unit_error_errno(unit
, r
, "Failed to load environment files: %m");
4731 line
= exec_command_line(command
->argv
);
4735 /* Fork with up-to-date SELinux label database, so the child inherits the up-to-date db
4736 and, until the next SELinux policy changes, we save further reloads in future children. */
4737 mac_selinux_maybe_reload();
4739 log_unit_struct(unit
, LOG_DEBUG
,
4740 LOG_UNIT_MESSAGE(unit
, "About to execute %s", line
),
4741 "EXECUTABLE=%s", command
->path
, /* We won't know the real executable path until we create
4742 the mount namespace in the child, but we want to log
4743 from the parent, so we need to use the (possibly
4744 inaccurate) path here. */
4745 LOG_UNIT_INVOCATION_ID(unit
));
4747 if (params
->cgroup_path
) {
4748 r
= exec_parameters_get_cgroup_path(params
, &subcgroup_path
);
4750 return log_unit_error_errno(unit
, r
, "Failed to acquire subcgroup path: %m");
4751 if (r
> 0) { /* We are using a child cgroup */
4752 r
= cg_create(SYSTEMD_CGROUP_CONTROLLER
, subcgroup_path
);
4754 return log_unit_error_errno(unit
, r
, "Failed to create control group '%s': %m", subcgroup_path
);
4756 /* Normally we would not propagate the oomd xattrs to children but since we created this
4757 * sub-cgroup internally we should do it. */
4758 cgroup_oomd_xattr_apply(unit
, subcgroup_path
);
4764 return log_unit_error_errno(unit
, errno
, "Failed to fork: %m");
4767 int exit_status
= EXIT_SUCCESS
;
4769 r
= exec_child(unit
,
4781 unit
->manager
->user_lookup_fds
[1],
4785 const char *status
=
4786 exit_status_to_string(exit_status
,
4787 EXIT_STATUS_LIBC
| EXIT_STATUS_SYSTEMD
);
4789 log_unit_struct_errno(unit
, LOG_ERR
, r
,
4790 "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR
,
4791 LOG_UNIT_INVOCATION_ID(unit
),
4792 LOG_UNIT_MESSAGE(unit
, "Failed at step %s spawning %s: %m",
4793 status
, command
->path
),
4794 "EXECUTABLE=%s", command
->path
);
4800 log_unit_debug(unit
, "Forked %s as "PID_FMT
, command
->path
, pid
);
4802 /* We add the new process to the cgroup both in the child (so that we can be sure that no user code is ever
4803 * executed outside of the cgroup) and in the parent (so that we can be sure that when we kill the cgroup the
4804 * process will be killed too). */
4806 (void) cg_attach(SYSTEMD_CGROUP_CONTROLLER
, subcgroup_path
, pid
);
4808 exec_status_start(&command
->exec_status
, pid
);
4814 void exec_context_init(ExecContext
*c
) {
4818 c
->ioprio
= IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE
, 0);
4819 c
->cpu_sched_policy
= SCHED_OTHER
;
4820 c
->syslog_priority
= LOG_DAEMON
|LOG_INFO
;
4821 c
->syslog_level_prefix
= true;
4822 c
->ignore_sigpipe
= true;
4823 c
->timer_slack_nsec
= NSEC_INFINITY
;
4824 c
->personality
= PERSONALITY_INVALID
;
4825 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++)
4826 c
->directories
[t
].mode
= 0755;
4827 c
->timeout_clean_usec
= USEC_INFINITY
;
4828 c
->capability_bounding_set
= CAP_ALL
;
4829 assert_cc(NAMESPACE_FLAGS_INITIAL
!= NAMESPACE_FLAGS_ALL
);
4830 c
->restrict_namespaces
= NAMESPACE_FLAGS_INITIAL
;
4831 c
->log_level_max
= -1;
4833 c
->syscall_errno
= SECCOMP_ERROR_NUMBER_KILL
;
4835 numa_policy_reset(&c
->numa_policy
);
4838 void exec_context_done(ExecContext
*c
) {
4841 c
->environment
= strv_free(c
->environment
);
4842 c
->environment_files
= strv_free(c
->environment_files
);
4843 c
->pass_environment
= strv_free(c
->pass_environment
);
4844 c
->unset_environment
= strv_free(c
->unset_environment
);
4846 rlimit_free_all(c
->rlimit
);
4848 for (size_t l
= 0; l
< 3; l
++) {
4849 c
->stdio_fdname
[l
] = mfree(c
->stdio_fdname
[l
]);
4850 c
->stdio_file
[l
] = mfree(c
->stdio_file
[l
]);
4853 c
->working_directory
= mfree(c
->working_directory
);
4854 c
->root_directory
= mfree(c
->root_directory
);
4855 c
->root_image
= mfree(c
->root_image
);
4856 c
->root_image_options
= mount_options_free_all(c
->root_image_options
);
4857 c
->root_hash
= mfree(c
->root_hash
);
4858 c
->root_hash_size
= 0;
4859 c
->root_hash_path
= mfree(c
->root_hash_path
);
4860 c
->root_hash_sig
= mfree(c
->root_hash_sig
);
4861 c
->root_hash_sig_size
= 0;
4862 c
->root_hash_sig_path
= mfree(c
->root_hash_sig_path
);
4863 c
->root_verity
= mfree(c
->root_verity
);
4864 c
->extension_images
= mount_image_free_many(c
->extension_images
, &c
->n_extension_images
);
4865 c
->tty_path
= mfree(c
->tty_path
);
4866 c
->syslog_identifier
= mfree(c
->syslog_identifier
);
4867 c
->user
= mfree(c
->user
);
4868 c
->group
= mfree(c
->group
);
4870 c
->supplementary_groups
= strv_free(c
->supplementary_groups
);
4872 c
->pam_name
= mfree(c
->pam_name
);
4874 c
->read_only_paths
= strv_free(c
->read_only_paths
);
4875 c
->read_write_paths
= strv_free(c
->read_write_paths
);
4876 c
->inaccessible_paths
= strv_free(c
->inaccessible_paths
);
4877 c
->exec_paths
= strv_free(c
->exec_paths
);
4878 c
->no_exec_paths
= strv_free(c
->no_exec_paths
);
4880 bind_mount_free_many(c
->bind_mounts
, c
->n_bind_mounts
);
4881 c
->bind_mounts
= NULL
;
4882 c
->n_bind_mounts
= 0;
4883 temporary_filesystem_free_many(c
->temporary_filesystems
, c
->n_temporary_filesystems
);
4884 c
->temporary_filesystems
= NULL
;
4885 c
->n_temporary_filesystems
= 0;
4886 c
->mount_images
= mount_image_free_many(c
->mount_images
, &c
->n_mount_images
);
4888 cpu_set_reset(&c
->cpu_set
);
4889 numa_policy_reset(&c
->numa_policy
);
4891 c
->utmp_id
= mfree(c
->utmp_id
);
4892 c
->selinux_context
= mfree(c
->selinux_context
);
4893 c
->apparmor_profile
= mfree(c
->apparmor_profile
);
4894 c
->smack_process_label
= mfree(c
->smack_process_label
);
4896 c
->syscall_filter
= hashmap_free(c
->syscall_filter
);
4897 c
->syscall_archs
= set_free(c
->syscall_archs
);
4898 c
->address_families
= set_free(c
->address_families
);
4900 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++)
4901 c
->directories
[t
].paths
= strv_free(c
->directories
[t
].paths
);
4903 c
->log_level_max
= -1;
4905 exec_context_free_log_extra_fields(c
);
4907 c
->log_ratelimit_interval_usec
= 0;
4908 c
->log_ratelimit_burst
= 0;
4910 c
->stdin_data
= mfree(c
->stdin_data
);
4911 c
->stdin_data_size
= 0;
4913 c
->network_namespace_path
= mfree(c
->network_namespace_path
);
4914 c
->ipc_namespace_path
= mfree(c
->ipc_namespace_path
);
4916 c
->log_namespace
= mfree(c
->log_namespace
);
4918 c
->load_credentials
= strv_free(c
->load_credentials
);
4919 c
->set_credentials
= hashmap_free(c
->set_credentials
);
4922 int exec_context_destroy_runtime_directory(const ExecContext
*c
, const char *runtime_prefix
) {
4927 if (!runtime_prefix
)
4930 STRV_FOREACH(i
, c
->directories
[EXEC_DIRECTORY_RUNTIME
].paths
) {
4931 _cleanup_free_
char *p
= NULL
;
4933 if (exec_directory_is_private(c
, EXEC_DIRECTORY_RUNTIME
))
4934 p
= path_join(runtime_prefix
, "private", *i
);
4936 p
= path_join(runtime_prefix
, *i
);
4940 /* We execute this synchronously, since we need to be sure this is gone when we start the
4942 (void) rm_rf(p
, REMOVE_ROOT
);
4948 int exec_context_destroy_credentials(const ExecContext
*c
, const char *runtime_prefix
, const char *unit
) {
4949 _cleanup_free_
char *p
= NULL
;
4953 if (!runtime_prefix
|| !unit
)
4956 p
= path_join(runtime_prefix
, "credentials", unit
);
4960 /* This is either a tmpfs/ramfs of its own, or a plain directory. Either way, let's first try to
4961 * unmount it, and afterwards remove the mount point */
4962 (void) umount2(p
, MNT_DETACH
|UMOUNT_NOFOLLOW
);
4963 (void) rm_rf(p
, REMOVE_ROOT
|REMOVE_CHMOD
);
4968 static void exec_command_done(ExecCommand
*c
) {
4971 c
->path
= mfree(c
->path
);
4972 c
->argv
= strv_free(c
->argv
);
4975 void exec_command_done_array(ExecCommand
*c
, size_t n
) {
4976 for (size_t i
= 0; i
< n
; i
++)
4977 exec_command_done(c
+i
);
4980 ExecCommand
* exec_command_free_list(ExecCommand
*c
) {
4984 LIST_REMOVE(command
, c
, i
);
4985 exec_command_done(i
);
4992 void exec_command_free_array(ExecCommand
**c
, size_t n
) {
4993 for (size_t i
= 0; i
< n
; i
++)
4994 c
[i
] = exec_command_free_list(c
[i
]);
4997 void exec_command_reset_status_array(ExecCommand
*c
, size_t n
) {
4998 for (size_t i
= 0; i
< n
; i
++)
4999 exec_status_reset(&c
[i
].exec_status
);
5002 void exec_command_reset_status_list_array(ExecCommand
**c
, size_t n
) {
5003 for (size_t i
= 0; i
< n
; i
++) {
5006 LIST_FOREACH(command
, z
, c
[i
])
5007 exec_status_reset(&z
->exec_status
);
5011 typedef struct InvalidEnvInfo
{
5016 static void invalid_env(const char *p
, void *userdata
) {
5017 InvalidEnvInfo
*info
= userdata
;
5019 log_unit_error(info
->unit
, "Ignoring invalid environment assignment '%s': %s", p
, info
->path
);
5022 const char* exec_context_fdname(const ExecContext
*c
, int fd_index
) {
5028 if (c
->std_input
!= EXEC_INPUT_NAMED_FD
)
5031 return c
->stdio_fdname
[STDIN_FILENO
] ?: "stdin";
5034 if (c
->std_output
!= EXEC_OUTPUT_NAMED_FD
)
5037 return c
->stdio_fdname
[STDOUT_FILENO
] ?: "stdout";
5040 if (c
->std_error
!= EXEC_OUTPUT_NAMED_FD
)
5043 return c
->stdio_fdname
[STDERR_FILENO
] ?: "stderr";
5050 static int exec_context_named_iofds(
5051 const ExecContext
*c
,
5052 const ExecParameters
*p
,
5053 int named_iofds
[static 3]) {
5056 const char* stdio_fdname
[3];
5061 assert(named_iofds
);
5063 targets
= (c
->std_input
== EXEC_INPUT_NAMED_FD
) +
5064 (c
->std_output
== EXEC_OUTPUT_NAMED_FD
) +
5065 (c
->std_error
== EXEC_OUTPUT_NAMED_FD
);
5067 for (size_t i
= 0; i
< 3; i
++)
5068 stdio_fdname
[i
] = exec_context_fdname(c
, i
);
5070 n_fds
= p
->n_storage_fds
+ p
->n_socket_fds
;
5072 for (size_t i
= 0; i
< n_fds
&& targets
> 0; i
++)
5073 if (named_iofds
[STDIN_FILENO
] < 0 &&
5074 c
->std_input
== EXEC_INPUT_NAMED_FD
&&
5075 stdio_fdname
[STDIN_FILENO
] &&
5076 streq(p
->fd_names
[i
], stdio_fdname
[STDIN_FILENO
])) {
5078 named_iofds
[STDIN_FILENO
] = p
->fds
[i
];
5081 } else if (named_iofds
[STDOUT_FILENO
] < 0 &&
5082 c
->std_output
== EXEC_OUTPUT_NAMED_FD
&&
5083 stdio_fdname
[STDOUT_FILENO
] &&
5084 streq(p
->fd_names
[i
], stdio_fdname
[STDOUT_FILENO
])) {
5086 named_iofds
[STDOUT_FILENO
] = p
->fds
[i
];
5089 } else if (named_iofds
[STDERR_FILENO
] < 0 &&
5090 c
->std_error
== EXEC_OUTPUT_NAMED_FD
&&
5091 stdio_fdname
[STDERR_FILENO
] &&
5092 streq(p
->fd_names
[i
], stdio_fdname
[STDERR_FILENO
])) {
5094 named_iofds
[STDERR_FILENO
] = p
->fds
[i
];
5098 return targets
== 0 ? 0 : -ENOENT
;
5101 static int exec_context_load_environment(const Unit
*unit
, const ExecContext
*c
, char ***l
) {
5102 char **i
, **r
= NULL
;
5107 STRV_FOREACH(i
, c
->environment_files
) {
5110 bool ignore
= false;
5112 _cleanup_globfree_ glob_t pglob
= {};
5121 if (!path_is_absolute(fn
)) {
5129 /* Filename supports globbing, take all matching files */
5130 k
= safe_glob(fn
, 0, &pglob
);
5139 /* When we don't match anything, -ENOENT should be returned */
5140 assert(pglob
.gl_pathc
> 0);
5142 for (unsigned n
= 0; n
< pglob
.gl_pathc
; n
++) {
5143 k
= load_env_file(NULL
, pglob
.gl_pathv
[n
], &p
);
5151 /* Log invalid environment variables with filename */
5153 InvalidEnvInfo info
= {
5155 .path
= pglob
.gl_pathv
[n
]
5158 p
= strv_env_clean_with_callback(p
, invalid_env
, &info
);
5166 m
= strv_env_merge(2, r
, p
);
5182 static bool tty_may_match_dev_console(const char *tty
) {
5183 _cleanup_free_
char *resolved
= NULL
;
5188 tty
= skip_dev_prefix(tty
);
5190 /* trivial identity? */
5191 if (streq(tty
, "console"))
5194 if (resolve_dev_console(&resolved
) < 0)
5195 return true; /* if we could not resolve, assume it may */
5197 /* "tty0" means the active VC, so it may be the same sometimes */
5198 return path_equal(resolved
, tty
) || (streq(resolved
, "tty0") && tty_is_vc(tty
));
5201 static bool exec_context_may_touch_tty(const ExecContext
*ec
) {
5204 return ec
->tty_reset
||
5206 ec
->tty_vt_disallocate
||
5207 is_terminal_input(ec
->std_input
) ||
5208 is_terminal_output(ec
->std_output
) ||
5209 is_terminal_output(ec
->std_error
);
5212 bool exec_context_may_touch_console(const ExecContext
*ec
) {
5214 return exec_context_may_touch_tty(ec
) &&
5215 tty_may_match_dev_console(exec_context_tty_path(ec
));
5218 static void strv_fprintf(FILE *f
, char **l
) {
5224 fprintf(f
, " %s", *g
);
5227 static void strv_dump(FILE* f
, const char *prefix
, const char *name
, char **strv
) {
5232 if (!strv_isempty(strv
)) {
5233 fprintf(f
, "%s%s:", prefix
, name
);
5234 strv_fprintf(f
, strv
);
5239 void exec_context_dump(const ExecContext
*c
, FILE* f
, const char *prefix
) {
5240 char **e
, **d
, buf_clean
[FORMAT_TIMESPAN_MAX
];
5246 prefix
= strempty(prefix
);
5250 "%sWorkingDirectory: %s\n"
5251 "%sRootDirectory: %s\n"
5252 "%sNonBlocking: %s\n"
5253 "%sPrivateTmp: %s\n"
5254 "%sPrivateDevices: %s\n"
5255 "%sProtectKernelTunables: %s\n"
5256 "%sProtectKernelModules: %s\n"
5257 "%sProtectKernelLogs: %s\n"
5258 "%sProtectClock: %s\n"
5259 "%sProtectControlGroups: %s\n"
5260 "%sPrivateNetwork: %s\n"
5261 "%sPrivateUsers: %s\n"
5262 "%sProtectHome: %s\n"
5263 "%sProtectSystem: %s\n"
5264 "%sMountAPIVFS: %s\n"
5265 "%sIgnoreSIGPIPE: %s\n"
5266 "%sMemoryDenyWriteExecute: %s\n"
5267 "%sRestrictRealtime: %s\n"
5268 "%sRestrictSUIDSGID: %s\n"
5269 "%sKeyringMode: %s\n"
5270 "%sProtectHostname: %s\n"
5271 "%sProtectProc: %s\n"
5272 "%sProcSubset: %s\n",
5274 prefix
, empty_to_root(c
->working_directory
),
5275 prefix
, empty_to_root(c
->root_directory
),
5276 prefix
, yes_no(c
->non_blocking
),
5277 prefix
, yes_no(c
->private_tmp
),
5278 prefix
, yes_no(c
->private_devices
),
5279 prefix
, yes_no(c
->protect_kernel_tunables
),
5280 prefix
, yes_no(c
->protect_kernel_modules
),
5281 prefix
, yes_no(c
->protect_kernel_logs
),
5282 prefix
, yes_no(c
->protect_clock
),
5283 prefix
, yes_no(c
->protect_control_groups
),
5284 prefix
, yes_no(c
->private_network
),
5285 prefix
, yes_no(c
->private_users
),
5286 prefix
, protect_home_to_string(c
->protect_home
),
5287 prefix
, protect_system_to_string(c
->protect_system
),
5288 prefix
, yes_no(exec_context_get_effective_mount_apivfs(c
)),
5289 prefix
, yes_no(c
->ignore_sigpipe
),
5290 prefix
, yes_no(c
->memory_deny_write_execute
),
5291 prefix
, yes_no(c
->restrict_realtime
),
5292 prefix
, yes_no(c
->restrict_suid_sgid
),
5293 prefix
, exec_keyring_mode_to_string(c
->keyring_mode
),
5294 prefix
, yes_no(c
->protect_hostname
),
5295 prefix
, protect_proc_to_string(c
->protect_proc
),
5296 prefix
, proc_subset_to_string(c
->proc_subset
));
5299 fprintf(f
, "%sRootImage: %s\n", prefix
, c
->root_image
);
5301 if (c
->root_image_options
) {
5304 fprintf(f
, "%sRootImageOptions:", prefix
);
5305 LIST_FOREACH(mount_options
, o
, c
->root_image_options
)
5306 if (!isempty(o
->options
))
5307 fprintf(f
, " %s:%s",
5308 partition_designator_to_string(o
->partition_designator
),
5314 _cleanup_free_
char *encoded
= NULL
;
5315 encoded
= hexmem(c
->root_hash
, c
->root_hash_size
);
5317 fprintf(f
, "%sRootHash: %s\n", prefix
, encoded
);
5320 if (c
->root_hash_path
)
5321 fprintf(f
, "%sRootHash: %s\n", prefix
, c
->root_hash_path
);
5323 if (c
->root_hash_sig
) {
5324 _cleanup_free_
char *encoded
= NULL
;
5326 len
= base64mem(c
->root_hash_sig
, c
->root_hash_sig_size
, &encoded
);
5328 fprintf(f
, "%sRootHashSignature: base64:%s\n", prefix
, encoded
);
5331 if (c
->root_hash_sig_path
)
5332 fprintf(f
, "%sRootHashSignature: %s\n", prefix
, c
->root_hash_sig_path
);
5335 fprintf(f
, "%sRootVerity: %s\n", prefix
, c
->root_verity
);
5337 STRV_FOREACH(e
, c
->environment
)
5338 fprintf(f
, "%sEnvironment: %s\n", prefix
, *e
);
5340 STRV_FOREACH(e
, c
->environment_files
)
5341 fprintf(f
, "%sEnvironmentFile: %s\n", prefix
, *e
);
5343 STRV_FOREACH(e
, c
->pass_environment
)
5344 fprintf(f
, "%sPassEnvironment: %s\n", prefix
, *e
);
5346 STRV_FOREACH(e
, c
->unset_environment
)
5347 fprintf(f
, "%sUnsetEnvironment: %s\n", prefix
, *e
);
5349 fprintf(f
, "%sRuntimeDirectoryPreserve: %s\n", prefix
, exec_preserve_mode_to_string(c
->runtime_directory_preserve_mode
));
5351 for (ExecDirectoryType dt
= 0; dt
< _EXEC_DIRECTORY_TYPE_MAX
; dt
++) {
5352 fprintf(f
, "%s%sMode: %04o\n", prefix
, exec_directory_type_to_string(dt
), c
->directories
[dt
].mode
);
5354 STRV_FOREACH(d
, c
->directories
[dt
].paths
)
5355 fprintf(f
, "%s%s: %s\n", prefix
, exec_directory_type_to_string(dt
), *d
);
5359 "%sTimeoutCleanSec: %s\n",
5360 prefix
, format_timespan(buf_clean
, sizeof(buf_clean
), c
->timeout_clean_usec
, USEC_PER_SEC
));
5367 if (c
->oom_score_adjust_set
)
5369 "%sOOMScoreAdjust: %i\n",
5370 prefix
, c
->oom_score_adjust
);
5372 if (c
->coredump_filter_set
)
5374 "%sCoredumpFilter: 0x%"PRIx64
"\n",
5375 prefix
, c
->coredump_filter
);
5377 for (unsigned i
= 0; i
< RLIM_NLIMITS
; i
++)
5379 fprintf(f
, "%sLimit%s: " RLIM_FMT
"\n",
5380 prefix
, rlimit_to_string(i
), c
->rlimit
[i
]->rlim_max
);
5381 fprintf(f
, "%sLimit%sSoft: " RLIM_FMT
"\n",
5382 prefix
, rlimit_to_string(i
), c
->rlimit
[i
]->rlim_cur
);
5385 if (c
->ioprio_set
) {
5386 _cleanup_free_
char *class_str
= NULL
;
5388 r
= ioprio_class_to_string_alloc(IOPRIO_PRIO_CLASS(c
->ioprio
), &class_str
);
5390 fprintf(f
, "%sIOSchedulingClass: %s\n", prefix
, class_str
);
5392 fprintf(f
, "%sIOPriority: %lu\n", prefix
, IOPRIO_PRIO_DATA(c
->ioprio
));
5395 if (c
->cpu_sched_set
) {
5396 _cleanup_free_
char *policy_str
= NULL
;
5398 r
= sched_policy_to_string_alloc(c
->cpu_sched_policy
, &policy_str
);
5400 fprintf(f
, "%sCPUSchedulingPolicy: %s\n", prefix
, policy_str
);
5403 "%sCPUSchedulingPriority: %i\n"
5404 "%sCPUSchedulingResetOnFork: %s\n",
5405 prefix
, c
->cpu_sched_priority
,
5406 prefix
, yes_no(c
->cpu_sched_reset_on_fork
));
5409 if (c
->cpu_set
.set
) {
5410 _cleanup_free_
char *affinity
= NULL
;
5412 affinity
= cpu_set_to_range_string(&c
->cpu_set
);
5413 fprintf(f
, "%sCPUAffinity: %s\n", prefix
, affinity
);
5416 if (mpol_is_valid(numa_policy_get_type(&c
->numa_policy
))) {
5417 _cleanup_free_
char *nodes
= NULL
;
5419 nodes
= cpu_set_to_range_string(&c
->numa_policy
.nodes
);
5420 fprintf(f
, "%sNUMAPolicy: %s\n", prefix
, mpol_to_string(numa_policy_get_type(&c
->numa_policy
)));
5421 fprintf(f
, "%sNUMAMask: %s\n", prefix
, strnull(nodes
));
5424 if (c
->timer_slack_nsec
!= NSEC_INFINITY
)
5425 fprintf(f
, "%sTimerSlackNSec: "NSEC_FMT
"\n", prefix
, c
->timer_slack_nsec
);
5428 "%sStandardInput: %s\n"
5429 "%sStandardOutput: %s\n"
5430 "%sStandardError: %s\n",
5431 prefix
, exec_input_to_string(c
->std_input
),
5432 prefix
, exec_output_to_string(c
->std_output
),
5433 prefix
, exec_output_to_string(c
->std_error
));
5435 if (c
->std_input
== EXEC_INPUT_NAMED_FD
)
5436 fprintf(f
, "%sStandardInputFileDescriptorName: %s\n", prefix
, c
->stdio_fdname
[STDIN_FILENO
]);
5437 if (c
->std_output
== EXEC_OUTPUT_NAMED_FD
)
5438 fprintf(f
, "%sStandardOutputFileDescriptorName: %s\n", prefix
, c
->stdio_fdname
[STDOUT_FILENO
]);
5439 if (c
->std_error
== EXEC_OUTPUT_NAMED_FD
)
5440 fprintf(f
, "%sStandardErrorFileDescriptorName: %s\n", prefix
, c
->stdio_fdname
[STDERR_FILENO
]);
5442 if (c
->std_input
== EXEC_INPUT_FILE
)
5443 fprintf(f
, "%sStandardInputFile: %s\n", prefix
, c
->stdio_file
[STDIN_FILENO
]);
5444 if (c
->std_output
== EXEC_OUTPUT_FILE
)
5445 fprintf(f
, "%sStandardOutputFile: %s\n", prefix
, c
->stdio_file
[STDOUT_FILENO
]);
5446 if (c
->std_output
== EXEC_OUTPUT_FILE_APPEND
)
5447 fprintf(f
, "%sStandardOutputFileToAppend: %s\n", prefix
, c
->stdio_file
[STDOUT_FILENO
]);
5448 if (c
->std_output
== EXEC_OUTPUT_FILE_TRUNCATE
)
5449 fprintf(f
, "%sStandardOutputFileToTruncate: %s\n", prefix
, c
->stdio_file
[STDOUT_FILENO
]);
5450 if (c
->std_error
== EXEC_OUTPUT_FILE
)
5451 fprintf(f
, "%sStandardErrorFile: %s\n", prefix
, c
->stdio_file
[STDERR_FILENO
]);
5452 if (c
->std_error
== EXEC_OUTPUT_FILE_APPEND
)
5453 fprintf(f
, "%sStandardErrorFileToAppend: %s\n", prefix
, c
->stdio_file
[STDERR_FILENO
]);
5454 if (c
->std_error
== EXEC_OUTPUT_FILE_TRUNCATE
)
5455 fprintf(f
, "%sStandardErrorFileToTruncate: %s\n", prefix
, c
->stdio_file
[STDERR_FILENO
]);
5461 "%sTTYVHangup: %s\n"
5462 "%sTTYVTDisallocate: %s\n",
5463 prefix
, c
->tty_path
,
5464 prefix
, yes_no(c
->tty_reset
),
5465 prefix
, yes_no(c
->tty_vhangup
),
5466 prefix
, yes_no(c
->tty_vt_disallocate
));
5468 if (IN_SET(c
->std_output
,
5470 EXEC_OUTPUT_JOURNAL
,
5471 EXEC_OUTPUT_KMSG_AND_CONSOLE
,
5472 EXEC_OUTPUT_JOURNAL_AND_CONSOLE
) ||
5473 IN_SET(c
->std_error
,
5475 EXEC_OUTPUT_JOURNAL
,
5476 EXEC_OUTPUT_KMSG_AND_CONSOLE
,
5477 EXEC_OUTPUT_JOURNAL_AND_CONSOLE
)) {
5479 _cleanup_free_
char *fac_str
= NULL
, *lvl_str
= NULL
;
5481 r
= log_facility_unshifted_to_string_alloc(c
->syslog_priority
>> 3, &fac_str
);
5483 fprintf(f
, "%sSyslogFacility: %s\n", prefix
, fac_str
);
5485 r
= log_level_to_string_alloc(LOG_PRI(c
->syslog_priority
), &lvl_str
);
5487 fprintf(f
, "%sSyslogLevel: %s\n", prefix
, lvl_str
);
5490 if (c
->log_level_max
>= 0) {
5491 _cleanup_free_
char *t
= NULL
;
5493 (void) log_level_to_string_alloc(c
->log_level_max
, &t
);
5495 fprintf(f
, "%sLogLevelMax: %s\n", prefix
, strna(t
));
5498 if (c
->log_ratelimit_interval_usec
> 0) {
5499 char buf_timespan
[FORMAT_TIMESPAN_MAX
];
5502 "%sLogRateLimitIntervalSec: %s\n",
5503 prefix
, format_timespan(buf_timespan
, sizeof(buf_timespan
), c
->log_ratelimit_interval_usec
, USEC_PER_SEC
));
5506 if (c
->log_ratelimit_burst
> 0)
5507 fprintf(f
, "%sLogRateLimitBurst: %u\n", prefix
, c
->log_ratelimit_burst
);
5509 for (size_t j
= 0; j
< c
->n_log_extra_fields
; j
++) {
5510 fprintf(f
, "%sLogExtraFields: ", prefix
);
5511 fwrite(c
->log_extra_fields
[j
].iov_base
,
5512 1, c
->log_extra_fields
[j
].iov_len
,
5517 if (c
->log_namespace
)
5518 fprintf(f
, "%sLogNamespace: %s\n", prefix
, c
->log_namespace
);
5520 if (c
->secure_bits
) {
5521 _cleanup_free_
char *str
= NULL
;
5523 r
= secure_bits_to_string_alloc(c
->secure_bits
, &str
);
5525 fprintf(f
, "%sSecure Bits: %s\n", prefix
, str
);
5528 if (c
->capability_bounding_set
!= CAP_ALL
) {
5529 _cleanup_free_
char *str
= NULL
;
5531 r
= capability_set_to_string_alloc(c
->capability_bounding_set
, &str
);
5533 fprintf(f
, "%sCapabilityBoundingSet: %s\n", prefix
, str
);
5536 if (c
->capability_ambient_set
!= 0) {
5537 _cleanup_free_
char *str
= NULL
;
5539 r
= capability_set_to_string_alloc(c
->capability_ambient_set
, &str
);
5541 fprintf(f
, "%sAmbientCapabilities: %s\n", prefix
, str
);
5545 fprintf(f
, "%sUser: %s\n", prefix
, c
->user
);
5547 fprintf(f
, "%sGroup: %s\n", prefix
, c
->group
);
5549 fprintf(f
, "%sDynamicUser: %s\n", prefix
, yes_no(c
->dynamic_user
));
5551 strv_dump(f
, prefix
, "SupplementaryGroups", c
->supplementary_groups
);
5554 fprintf(f
, "%sPAMName: %s\n", prefix
, c
->pam_name
);
5556 strv_dump(f
, prefix
, "ReadWritePaths", c
->read_write_paths
);
5557 strv_dump(f
, prefix
, "ReadOnlyPaths", c
->read_only_paths
);
5558 strv_dump(f
, prefix
, "InaccessiblePaths", c
->inaccessible_paths
);
5559 strv_dump(f
, prefix
, "ExecPaths", c
->exec_paths
);
5560 strv_dump(f
, prefix
, "NoExecPaths", c
->no_exec_paths
);
5562 for (size_t i
= 0; i
< c
->n_bind_mounts
; i
++)
5563 fprintf(f
, "%s%s: %s%s:%s:%s\n", prefix
,
5564 c
->bind_mounts
[i
].read_only
? "BindReadOnlyPaths" : "BindPaths",
5565 c
->bind_mounts
[i
].ignore_enoent
? "-": "",
5566 c
->bind_mounts
[i
].source
,
5567 c
->bind_mounts
[i
].destination
,
5568 c
->bind_mounts
[i
].recursive
? "rbind" : "norbind");
5570 for (size_t i
= 0; i
< c
->n_temporary_filesystems
; i
++) {
5571 const TemporaryFileSystem
*t
= c
->temporary_filesystems
+ i
;
5573 fprintf(f
, "%sTemporaryFileSystem: %s%s%s\n", prefix
,
5575 isempty(t
->options
) ? "" : ":",
5576 strempty(t
->options
));
5581 "%sUtmpIdentifier: %s\n",
5582 prefix
, c
->utmp_id
);
5584 if (c
->selinux_context
)
5586 "%sSELinuxContext: %s%s\n",
5587 prefix
, c
->selinux_context_ignore
? "-" : "", c
->selinux_context
);
5589 if (c
->apparmor_profile
)
5591 "%sAppArmorProfile: %s%s\n",
5592 prefix
, c
->apparmor_profile_ignore
? "-" : "", c
->apparmor_profile
);
5594 if (c
->smack_process_label
)
5596 "%sSmackProcessLabel: %s%s\n",
5597 prefix
, c
->smack_process_label_ignore
? "-" : "", c
->smack_process_label
);
5599 if (c
->personality
!= PERSONALITY_INVALID
)
5601 "%sPersonality: %s\n",
5602 prefix
, strna(personality_to_string(c
->personality
)));
5605 "%sLockPersonality: %s\n",
5606 prefix
, yes_no(c
->lock_personality
));
5608 if (c
->syscall_filter
) {
5615 "%sSystemCallFilter: ",
5618 if (!c
->syscall_allow_list
)
5622 HASHMAP_FOREACH_KEY(val
, id
, c
->syscall_filter
) {
5623 _cleanup_free_
char *name
= NULL
;
5624 const char *errno_name
= NULL
;
5625 int num
= PTR_TO_INT(val
);
5632 name
= seccomp_syscall_resolve_num_arch(SCMP_ARCH_NATIVE
, PTR_TO_INT(id
) - 1);
5633 fputs(strna(name
), f
);
5636 errno_name
= seccomp_errno_or_action_to_string(num
);
5638 fprintf(f
, ":%s", errno_name
);
5640 fprintf(f
, ":%d", num
);
5648 if (c
->syscall_archs
) {
5654 "%sSystemCallArchitectures:",
5658 SET_FOREACH(id
, c
->syscall_archs
)
5659 fprintf(f
, " %s", strna(seccomp_arch_to_string(PTR_TO_UINT32(id
) - 1)));
5664 if (exec_context_restrict_namespaces_set(c
)) {
5665 _cleanup_free_
char *s
= NULL
;
5667 r
= namespace_flags_to_string(c
->restrict_namespaces
, &s
);
5669 fprintf(f
, "%sRestrictNamespaces: %s\n",
5673 if (c
->network_namespace_path
)
5675 "%sNetworkNamespacePath: %s\n",
5676 prefix
, c
->network_namespace_path
);
5678 if (c
->syscall_errno
> 0) {
5680 const char *errno_name
;
5683 fprintf(f
, "%sSystemCallErrorNumber: ", prefix
);
5686 errno_name
= seccomp_errno_or_action_to_string(c
->syscall_errno
);
5688 fputs(errno_name
, f
);
5690 fprintf(f
, "%d", c
->syscall_errno
);
5695 for (size_t i
= 0; i
< c
->n_mount_images
; i
++) {
5698 fprintf(f
, "%sMountImages: %s%s:%s", prefix
,
5699 c
->mount_images
[i
].ignore_enoent
? "-": "",
5700 c
->mount_images
[i
].source
,
5701 c
->mount_images
[i
].destination
);
5702 LIST_FOREACH(mount_options
, o
, c
->mount_images
[i
].mount_options
)
5703 fprintf(f
, ":%s:%s",
5704 partition_designator_to_string(o
->partition_designator
),
5705 strempty(o
->options
));
5709 for (size_t i
= 0; i
< c
->n_extension_images
; i
++) {
5712 fprintf(f
, "%sExtensionImages: %s%s", prefix
,
5713 c
->extension_images
[i
].ignore_enoent
? "-": "",
5714 c
->extension_images
[i
].source
);
5715 LIST_FOREACH(mount_options
, o
, c
->extension_images
[i
].mount_options
)
5716 fprintf(f
, ":%s:%s",
5717 partition_designator_to_string(o
->partition_designator
),
5718 strempty(o
->options
));
5723 bool exec_context_maintains_privileges(const ExecContext
*c
) {
5726 /* Returns true if the process forked off would run under
5727 * an unchanged UID or as root. */
5732 if (streq(c
->user
, "root") || streq(c
->user
, "0"))
5738 int exec_context_get_effective_ioprio(const ExecContext
*c
) {
5746 p
= ioprio_get(IOPRIO_WHO_PROCESS
, 0);
5748 return IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE
, 4);
5753 bool exec_context_get_effective_mount_apivfs(const ExecContext
*c
) {
5756 /* Explicit setting wins */
5757 if (c
->mount_apivfs_set
)
5758 return c
->mount_apivfs
;
5760 /* Default to "yes" if root directory or image are specified */
5761 if (exec_context_with_rootfs(c
))
5767 void exec_context_free_log_extra_fields(ExecContext
*c
) {
5770 for (size_t l
= 0; l
< c
->n_log_extra_fields
; l
++)
5771 free(c
->log_extra_fields
[l
].iov_base
);
5772 c
->log_extra_fields
= mfree(c
->log_extra_fields
);
5773 c
->n_log_extra_fields
= 0;
5776 void exec_context_revert_tty(ExecContext
*c
) {
5777 _cleanup_close_
int fd
= -1;
5784 /* First, reset the TTY (possibly kicking everybody else from the TTY) */
5785 exec_context_tty_reset(c
, NULL
);
5787 /* And then undo what chown_terminal() did earlier. Note that we only do this if we have a path
5788 * configured. If the TTY was passed to us as file descriptor we assume the TTY is opened and managed
5789 * by whoever passed it to us and thus knows better when and how to chmod()/chown() it back. */
5790 if (!exec_context_may_touch_tty(c
))
5793 path
= exec_context_tty_path(c
);
5797 fd
= open(path
, O_PATH
|O_CLOEXEC
);
5799 return (void) log_full_errno(errno
== ENOENT
? LOG_DEBUG
: LOG_WARNING
, errno
,
5800 "Failed to open TTY inode of '%s' to adjust ownership/access mode, ignoring: %m",
5803 if (fstat(fd
, &st
) < 0)
5804 return (void) log_warning_errno(errno
, "Failed to stat TTY '%s', ignoring: %m", path
);
5806 /* Let's add a superficial check that we only do this for stuff that looks like a TTY. We only check
5807 * if things are a character device, since a proper check either means we'd have to open the TTY and
5808 * use isatty(), but we'd rather not do that since opening TTYs comes with all kinds of side-effects
5809 * and is slow. Or we'd have to hardcode dev_t major information, which we'd rather avoid. Why bother
5810 * with this at all? → https://github.com/systemd/systemd/issues/19213 */
5811 if (!S_ISCHR(st
.st_mode
))
5812 return log_warning("Configured TTY '%s' is not actually a character device, ignoring.", path
);
5814 r
= fchmod_and_chown(fd
, TTY_MODE
, 0, TTY_GID
);
5816 log_warning_errno(r
, "Failed to reset TTY ownership/access mode of %s, ignoring: %m", path
);
5819 int exec_context_get_clean_directories(
5825 _cleanup_strv_free_
char **l
= NULL
;
5832 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
5835 if (!FLAGS_SET(mask
, 1U << t
))
5841 STRV_FOREACH(i
, c
->directories
[t
].paths
) {
5844 j
= path_join(prefix
[t
], *i
);
5848 r
= strv_consume(&l
, j
);
5852 /* Also remove private directories unconditionally. */
5853 if (t
!= EXEC_DIRECTORY_CONFIGURATION
) {
5854 j
= path_join(prefix
[t
], "private", *i
);
5858 r
= strv_consume(&l
, j
);
5869 int exec_context_get_clean_mask(ExecContext
*c
, ExecCleanMask
*ret
) {
5870 ExecCleanMask mask
= 0;
5875 for (ExecDirectoryType t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++)
5876 if (!strv_isempty(c
->directories
[t
].paths
))
5883 void exec_status_start(ExecStatus
*s
, pid_t pid
) {
5890 dual_timestamp_get(&s
->start_timestamp
);
5893 void exec_status_exit(ExecStatus
*s
, const ExecContext
*context
, pid_t pid
, int code
, int status
) {
5901 dual_timestamp_get(&s
->exit_timestamp
);
5906 if (context
&& context
->utmp_id
)
5907 (void) utmp_put_dead_process(context
->utmp_id
, pid
, code
, status
);
5910 void exec_status_reset(ExecStatus
*s
) {
5913 *s
= (ExecStatus
) {};
5916 void exec_status_dump(const ExecStatus
*s
, FILE *f
, const char *prefix
) {
5917 char buf
[FORMAT_TIMESTAMP_MAX
];
5925 prefix
= strempty(prefix
);
5928 "%sPID: "PID_FMT
"\n",
5931 if (dual_timestamp_is_set(&s
->start_timestamp
))
5933 "%sStart Timestamp: %s\n",
5934 prefix
, format_timestamp(buf
, sizeof(buf
), s
->start_timestamp
.realtime
));
5936 if (dual_timestamp_is_set(&s
->exit_timestamp
))
5938 "%sExit Timestamp: %s\n"
5940 "%sExit Status: %i\n",
5941 prefix
, format_timestamp(buf
, sizeof(buf
), s
->exit_timestamp
.realtime
),
5942 prefix
, sigchld_code_to_string(s
->code
),
5946 static char *exec_command_line(char **argv
) {
5954 STRV_FOREACH(a
, argv
)
5962 STRV_FOREACH(a
, argv
) {
5969 if (strpbrk(*a
, WHITESPACE
)) {
5980 /* FIXME: this doesn't really handle arguments that have
5981 * spaces and ticks in them */
5986 static void exec_command_dump(ExecCommand
*c
, FILE *f
, const char *prefix
) {
5987 _cleanup_free_
char *cmd
= NULL
;
5988 const char *prefix2
;
5993 prefix
= strempty(prefix
);
5994 prefix2
= strjoina(prefix
, "\t");
5996 cmd
= exec_command_line(c
->argv
);
5998 "%sCommand Line: %s\n",
5999 prefix
, cmd
? cmd
: strerror_safe(ENOMEM
));
6001 exec_status_dump(&c
->exec_status
, f
, prefix2
);
6004 void exec_command_dump_list(ExecCommand
*c
, FILE *f
, const char *prefix
) {
6007 prefix
= strempty(prefix
);
6009 LIST_FOREACH(command
, c
, c
)
6010 exec_command_dump(c
, f
, prefix
);
6013 void exec_command_append_list(ExecCommand
**l
, ExecCommand
*e
) {
6020 /* It's kind of important, that we keep the order here */
6021 LIST_FIND_TAIL(command
, *l
, end
);
6022 LIST_INSERT_AFTER(command
, *l
, end
, e
);
6027 int exec_command_set(ExecCommand
*c
, const char *path
, ...) {
6035 l
= strv_new_ap(path
, ap
);
6047 free_and_replace(c
->path
, p
);
6049 return strv_free_and_replace(c
->argv
, l
);
6052 int exec_command_append(ExecCommand
*c
, const char *path
, ...) {
6053 _cleanup_strv_free_
char **l
= NULL
;
6061 l
= strv_new_ap(path
, ap
);
6067 r
= strv_extend_strv(&c
->argv
, l
, false);
6074 static void *remove_tmpdir_thread(void *p
) {
6075 _cleanup_free_
char *path
= p
;
6077 (void) rm_rf(path
, REMOVE_ROOT
|REMOVE_PHYSICAL
);
6081 static ExecRuntime
* exec_runtime_free(ExecRuntime
*rt
, bool destroy
) {
6088 (void) hashmap_remove(rt
->manager
->exec_runtime_by_id
, rt
->id
);
6090 /* When destroy is true, then rm_rf tmp_dir and var_tmp_dir. */
6092 if (destroy
&& rt
->tmp_dir
&& !streq(rt
->tmp_dir
, RUN_SYSTEMD_EMPTY
)) {
6093 log_debug("Spawning thread to nuke %s", rt
->tmp_dir
);
6095 r
= asynchronous_job(remove_tmpdir_thread
, rt
->tmp_dir
);
6097 log_warning_errno(r
, "Failed to nuke %s: %m", rt
->tmp_dir
);
6102 if (destroy
&& rt
->var_tmp_dir
&& !streq(rt
->var_tmp_dir
, RUN_SYSTEMD_EMPTY
)) {
6103 log_debug("Spawning thread to nuke %s", rt
->var_tmp_dir
);
6105 r
= asynchronous_job(remove_tmpdir_thread
, rt
->var_tmp_dir
);
6107 log_warning_errno(r
, "Failed to nuke %s: %m", rt
->var_tmp_dir
);
6109 rt
->var_tmp_dir
= NULL
;
6112 rt
->id
= mfree(rt
->id
);
6113 rt
->tmp_dir
= mfree(rt
->tmp_dir
);
6114 rt
->var_tmp_dir
= mfree(rt
->var_tmp_dir
);
6115 safe_close_pair(rt
->netns_storage_socket
);
6116 safe_close_pair(rt
->ipcns_storage_socket
);
6120 static void exec_runtime_freep(ExecRuntime
**rt
) {
6121 (void) exec_runtime_free(*rt
, false);
6124 static int exec_runtime_allocate(ExecRuntime
**ret
, const char *id
) {
6125 _cleanup_free_
char *id_copy
= NULL
;
6130 id_copy
= strdup(id
);
6134 n
= new(ExecRuntime
, 1);
6138 *n
= (ExecRuntime
) {
6139 .id
= TAKE_PTR(id_copy
),
6140 .netns_storage_socket
= { -1, -1 },
6141 .ipcns_storage_socket
= { -1, -1 },
6148 static int exec_runtime_add(
6153 int netns_storage_socket
[2],
6154 int ipcns_storage_socket
[2],
6155 ExecRuntime
**ret
) {
6157 _cleanup_(exec_runtime_freep
) ExecRuntime
*rt
= NULL
;
6163 /* tmp_dir, var_tmp_dir, {net,ipc}ns_storage_socket fds are donated on success */
6165 r
= exec_runtime_allocate(&rt
, id
);
6169 r
= hashmap_ensure_put(&m
->exec_runtime_by_id
, &string_hash_ops
, rt
->id
, rt
);
6173 assert(!!rt
->tmp_dir
== !!rt
->var_tmp_dir
); /* We require both to be set together */
6174 rt
->tmp_dir
= TAKE_PTR(*tmp_dir
);
6175 rt
->var_tmp_dir
= TAKE_PTR(*var_tmp_dir
);
6177 if (netns_storage_socket
) {
6178 rt
->netns_storage_socket
[0] = TAKE_FD(netns_storage_socket
[0]);
6179 rt
->netns_storage_socket
[1] = TAKE_FD(netns_storage_socket
[1]);
6182 if (ipcns_storage_socket
) {
6183 rt
->ipcns_storage_socket
[0] = TAKE_FD(ipcns_storage_socket
[0]);
6184 rt
->ipcns_storage_socket
[1] = TAKE_FD(ipcns_storage_socket
[1]);
6191 /* do not remove created ExecRuntime object when the operation succeeds. */
6196 static int exec_runtime_make(
6198 const ExecContext
*c
,
6200 ExecRuntime
**ret
) {
6202 _cleanup_(namespace_cleanup_tmpdirp
) char *tmp_dir
= NULL
, *var_tmp_dir
= NULL
;
6203 _cleanup_close_pair_
int netns_storage_socket
[2] = { -1, -1 }, ipcns_storage_socket
[2] = { -1, -1 };
6210 /* It is not necessary to create ExecRuntime object. */
6211 if (!c
->private_network
&& !c
->private_ipc
&& !c
->private_tmp
&& !c
->network_namespace_path
) {
6216 if (c
->private_tmp
&&
6217 !(prefixed_path_strv_contains(c
->inaccessible_paths
, "/tmp") &&
6218 (prefixed_path_strv_contains(c
->inaccessible_paths
, "/var/tmp") ||
6219 prefixed_path_strv_contains(c
->inaccessible_paths
, "/var")))) {
6220 r
= setup_tmp_dirs(id
, &tmp_dir
, &var_tmp_dir
);
6225 if (c
->private_network
|| c
->network_namespace_path
) {
6226 if (socketpair(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
, 0, netns_storage_socket
) < 0)
6230 if (c
->private_ipc
|| c
->ipc_namespace_path
) {
6231 if (socketpair(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
, 0, ipcns_storage_socket
) < 0)
6235 r
= exec_runtime_add(m
, id
, &tmp_dir
, &var_tmp_dir
, netns_storage_socket
, ipcns_storage_socket
, ret
);
6242 int exec_runtime_acquire(Manager
*m
, const ExecContext
*c
, const char *id
, bool create
, ExecRuntime
**ret
) {
6250 rt
= hashmap_get(m
->exec_runtime_by_id
, id
);
6252 /* We already have a ExecRuntime object, let's increase the ref count and reuse it */
6260 /* If not found, then create a new object. */
6261 r
= exec_runtime_make(m
, c
, id
, &rt
);
6265 /* When r == 0, it is not necessary to create ExecRuntime object. */
6271 /* increment reference counter. */
6277 ExecRuntime
*exec_runtime_unref(ExecRuntime
*rt
, bool destroy
) {
6281 assert(rt
->n_ref
> 0);
6287 return exec_runtime_free(rt
, destroy
);
6290 int exec_runtime_serialize(const Manager
*m
, FILE *f
, FDSet
*fds
) {
6297 HASHMAP_FOREACH(rt
, m
->exec_runtime_by_id
) {
6298 fprintf(f
, "exec-runtime=%s", rt
->id
);
6301 fprintf(f
, " tmp-dir=%s", rt
->tmp_dir
);
6303 if (rt
->var_tmp_dir
)
6304 fprintf(f
, " var-tmp-dir=%s", rt
->var_tmp_dir
);
6306 if (rt
->netns_storage_socket
[0] >= 0) {
6309 copy
= fdset_put_dup(fds
, rt
->netns_storage_socket
[0]);
6313 fprintf(f
, " netns-socket-0=%i", copy
);
6316 if (rt
->netns_storage_socket
[1] >= 0) {
6319 copy
= fdset_put_dup(fds
, rt
->netns_storage_socket
[1]);
6323 fprintf(f
, " netns-socket-1=%i", copy
);
6326 if (rt
->ipcns_storage_socket
[0] >= 0) {
6329 copy
= fdset_put_dup(fds
, rt
->ipcns_storage_socket
[0]);
6333 fprintf(f
, " ipcns-socket-0=%i", copy
);
6336 if (rt
->ipcns_storage_socket
[1] >= 0) {
6339 copy
= fdset_put_dup(fds
, rt
->ipcns_storage_socket
[1]);
6343 fprintf(f
, " ipcns-socket-1=%i", copy
);
6352 int exec_runtime_deserialize_compat(Unit
*u
, const char *key
, const char *value
, FDSet
*fds
) {
6353 _cleanup_(exec_runtime_freep
) ExecRuntime
*rt_create
= NULL
;
6357 /* This is for the migration from old (v237 or earlier) deserialization text.
6358 * Due to the bug #7790, this may not work with the units that use JoinsNamespaceOf=.
6359 * Even if the ExecRuntime object originally created by the other unit, we cannot judge
6360 * so or not from the serialized text, then we always creates a new object owned by this. */
6366 /* Manager manages ExecRuntime objects by the unit id.
6367 * So, we omit the serialized text when the unit does not have id (yet?)... */
6368 if (isempty(u
->id
)) {
6369 log_unit_debug(u
, "Invocation ID not found. Dropping runtime parameter.");
6373 if (hashmap_ensure_allocated(&u
->manager
->exec_runtime_by_id
, &string_hash_ops
) < 0)
6376 rt
= hashmap_get(u
->manager
->exec_runtime_by_id
, u
->id
);
6378 if (exec_runtime_allocate(&rt_create
, u
->id
) < 0)
6384 if (streq(key
, "tmp-dir")) {
6385 if (free_and_strdup_warn(&rt
->tmp_dir
, value
) < 0)
6388 } else if (streq(key
, "var-tmp-dir")) {
6389 if (free_and_strdup_warn(&rt
->var_tmp_dir
, value
) < 0)
6392 } else if (streq(key
, "netns-socket-0")) {
6395 if (safe_atoi(value
, &fd
) < 0 || !fdset_contains(fds
, fd
)) {
6396 log_unit_debug(u
, "Failed to parse netns socket value: %s", value
);
6400 safe_close(rt
->netns_storage_socket
[0]);
6401 rt
->netns_storage_socket
[0] = fdset_remove(fds
, fd
);
6403 } else if (streq(key
, "netns-socket-1")) {
6406 if (safe_atoi(value
, &fd
) < 0 || !fdset_contains(fds
, fd
)) {
6407 log_unit_debug(u
, "Failed to parse netns socket value: %s", value
);
6411 safe_close(rt
->netns_storage_socket
[1]);
6412 rt
->netns_storage_socket
[1] = fdset_remove(fds
, fd
);
6417 /* If the object is newly created, then put it to the hashmap which manages ExecRuntime objects. */
6419 r
= hashmap_put(u
->manager
->exec_runtime_by_id
, rt_create
->id
, rt_create
);
6421 log_unit_debug_errno(u
, r
, "Failed to put runtime parameter to manager's storage: %m");
6425 rt_create
->manager
= u
->manager
;
6428 TAKE_PTR(rt_create
);
6434 int exec_runtime_deserialize_one(Manager
*m
, const char *value
, FDSet
*fds
) {
6435 _cleanup_free_
char *tmp_dir
= NULL
, *var_tmp_dir
= NULL
;
6437 int r
, netns_fdpair
[] = {-1, -1}, ipcns_fdpair
[] = {-1, -1};
6438 const char *p
, *v
= value
;
6445 n
= strcspn(v
, " ");
6446 id
= strndupa(v
, n
);
6451 v
= startswith(p
, "tmp-dir=");
6453 n
= strcspn(v
, " ");
6454 tmp_dir
= strndup(v
, n
);
6462 v
= startswith(p
, "var-tmp-dir=");
6464 n
= strcspn(v
, " ");
6465 var_tmp_dir
= strndup(v
, n
);
6473 v
= startswith(p
, "netns-socket-0=");
6477 n
= strcspn(v
, " ");
6478 buf
= strndupa(v
, n
);
6480 r
= safe_atoi(buf
, &netns_fdpair
[0]);
6482 return log_debug_errno(r
, "Unable to parse exec-runtime specification netns-socket-0=%s: %m", buf
);
6483 if (!fdset_contains(fds
, netns_fdpair
[0]))
6484 return log_debug_errno(SYNTHETIC_ERRNO(EBADF
),
6485 "exec-runtime specification netns-socket-0= refers to unknown fd %d: %m", netns_fdpair
[0]);
6486 netns_fdpair
[0] = fdset_remove(fds
, netns_fdpair
[0]);
6492 v
= startswith(p
, "netns-socket-1=");
6496 n
= strcspn(v
, " ");
6497 buf
= strndupa(v
, n
);
6499 r
= safe_atoi(buf
, &netns_fdpair
[1]);
6501 return log_debug_errno(r
, "Unable to parse exec-runtime specification netns-socket-1=%s: %m", buf
);
6502 if (!fdset_contains(fds
, netns_fdpair
[1]))
6503 return log_debug_errno(SYNTHETIC_ERRNO(EBADF
),
6504 "exec-runtime specification netns-socket-1= refers to unknown fd %d: %m", netns_fdpair
[1]);
6505 netns_fdpair
[1] = fdset_remove(fds
, netns_fdpair
[1]);
6511 v
= startswith(p
, "ipcns-socket-0=");
6515 n
= strcspn(v
, " ");
6516 buf
= strndupa(v
, n
);
6518 r
= safe_atoi(buf
, &ipcns_fdpair
[0]);
6520 return log_debug_errno(r
, "Unable to parse exec-runtime specification ipcns-socket-0=%s: %m", buf
);
6521 if (!fdset_contains(fds
, ipcns_fdpair
[0]))
6522 return log_debug_errno(SYNTHETIC_ERRNO(EBADF
),
6523 "exec-runtime specification ipcns-socket-0= refers to unknown fd %d: %m", ipcns_fdpair
[0]);
6524 ipcns_fdpair
[0] = fdset_remove(fds
, ipcns_fdpair
[0]);
6530 v
= startswith(p
, "ipcns-socket-1=");
6534 n
= strcspn(v
, " ");
6535 buf
= strndupa(v
, n
);
6537 r
= safe_atoi(buf
, &ipcns_fdpair
[1]);
6539 return log_debug_errno(r
, "Unable to parse exec-runtime specification ipcns-socket-1=%s: %m", buf
);
6540 if (!fdset_contains(fds
, ipcns_fdpair
[1]))
6541 return log_debug_errno(SYNTHETIC_ERRNO(EBADF
),
6542 "exec-runtime specification ipcns-socket-1= refers to unknown fd %d: %m", ipcns_fdpair
[1]);
6543 ipcns_fdpair
[1] = fdset_remove(fds
, ipcns_fdpair
[1]);
6547 r
= exec_runtime_add(m
, id
, &tmp_dir
, &var_tmp_dir
, netns_fdpair
, ipcns_fdpair
, NULL
);
6549 return log_debug_errno(r
, "Failed to add exec-runtime: %m");
6553 void exec_runtime_vacuum(Manager
*m
) {
6558 /* Free unreferenced ExecRuntime objects. This is used after manager deserialization process. */
6560 HASHMAP_FOREACH(rt
, m
->exec_runtime_by_id
) {
6564 (void) exec_runtime_free(rt
, false);
6568 void exec_params_clear(ExecParameters
*p
) {
6572 p
->environment
= strv_free(p
->environment
);
6573 p
->fd_names
= strv_free(p
->fd_names
);
6574 p
->fds
= mfree(p
->fds
);
6575 p
->exec_fd
= safe_close(p
->exec_fd
);
6578 ExecSetCredential
*exec_set_credential_free(ExecSetCredential
*sc
) {
6587 DEFINE_HASH_OPS_WITH_VALUE_DESTRUCTOR(exec_set_credential_hash_ops
, char, string_hash_func
, string_compare_func
, ExecSetCredential
, exec_set_credential_free
);
6589 static const char* const exec_input_table
[_EXEC_INPUT_MAX
] = {
6590 [EXEC_INPUT_NULL
] = "null",
6591 [EXEC_INPUT_TTY
] = "tty",
6592 [EXEC_INPUT_TTY_FORCE
] = "tty-force",
6593 [EXEC_INPUT_TTY_FAIL
] = "tty-fail",
6594 [EXEC_INPUT_SOCKET
] = "socket",
6595 [EXEC_INPUT_NAMED_FD
] = "fd",
6596 [EXEC_INPUT_DATA
] = "data",
6597 [EXEC_INPUT_FILE
] = "file",
6600 DEFINE_STRING_TABLE_LOOKUP(exec_input
, ExecInput
);
6602 static const char* const exec_output_table
[_EXEC_OUTPUT_MAX
] = {
6603 [EXEC_OUTPUT_INHERIT
] = "inherit",
6604 [EXEC_OUTPUT_NULL
] = "null",
6605 [EXEC_OUTPUT_TTY
] = "tty",
6606 [EXEC_OUTPUT_KMSG
] = "kmsg",
6607 [EXEC_OUTPUT_KMSG_AND_CONSOLE
] = "kmsg+console",
6608 [EXEC_OUTPUT_JOURNAL
] = "journal",
6609 [EXEC_OUTPUT_JOURNAL_AND_CONSOLE
] = "journal+console",
6610 [EXEC_OUTPUT_SOCKET
] = "socket",
6611 [EXEC_OUTPUT_NAMED_FD
] = "fd",
6612 [EXEC_OUTPUT_FILE
] = "file",
6613 [EXEC_OUTPUT_FILE_APPEND
] = "append",
6614 [EXEC_OUTPUT_FILE_TRUNCATE
] = "truncate",
6617 DEFINE_STRING_TABLE_LOOKUP(exec_output
, ExecOutput
);
6619 static const char* const exec_utmp_mode_table
[_EXEC_UTMP_MODE_MAX
] = {
6620 [EXEC_UTMP_INIT
] = "init",
6621 [EXEC_UTMP_LOGIN
] = "login",
6622 [EXEC_UTMP_USER
] = "user",
6625 DEFINE_STRING_TABLE_LOOKUP(exec_utmp_mode
, ExecUtmpMode
);
6627 static const char* const exec_preserve_mode_table
[_EXEC_PRESERVE_MODE_MAX
] = {
6628 [EXEC_PRESERVE_NO
] = "no",
6629 [EXEC_PRESERVE_YES
] = "yes",
6630 [EXEC_PRESERVE_RESTART
] = "restart",
6633 DEFINE_STRING_TABLE_LOOKUP_WITH_BOOLEAN(exec_preserve_mode
, ExecPreserveMode
, EXEC_PRESERVE_YES
);
6635 /* This table maps ExecDirectoryType to the setting it is configured with in the unit */
6636 static const char* const exec_directory_type_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
6637 [EXEC_DIRECTORY_RUNTIME
] = "RuntimeDirectory",
6638 [EXEC_DIRECTORY_STATE
] = "StateDirectory",
6639 [EXEC_DIRECTORY_CACHE
] = "CacheDirectory",
6640 [EXEC_DIRECTORY_LOGS
] = "LogsDirectory",
6641 [EXEC_DIRECTORY_CONFIGURATION
] = "ConfigurationDirectory",
6644 DEFINE_STRING_TABLE_LOOKUP(exec_directory_type
, ExecDirectoryType
);
6646 /* And this table maps ExecDirectoryType too, but to a generic term identifying the type of resource. This
6647 * one is supposed to be generic enough to be used for unit types that don't use ExecContext and per-unit
6648 * directories, specifically .timer units with their timestamp touch file. */
6649 static const char* const exec_resource_type_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
6650 [EXEC_DIRECTORY_RUNTIME
] = "runtime",
6651 [EXEC_DIRECTORY_STATE
] = "state",
6652 [EXEC_DIRECTORY_CACHE
] = "cache",
6653 [EXEC_DIRECTORY_LOGS
] = "logs",
6654 [EXEC_DIRECTORY_CONFIGURATION
] = "configuration",
6657 DEFINE_STRING_TABLE_LOOKUP(exec_resource_type
, ExecDirectoryType
);
6659 /* And this table also maps ExecDirectoryType, to the environment variable we pass the selected directory to
6660 * the service payload in. */
6661 static const char* const exec_directory_env_name_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
6662 [EXEC_DIRECTORY_RUNTIME
] = "RUNTIME_DIRECTORY",
6663 [EXEC_DIRECTORY_STATE
] = "STATE_DIRECTORY",
6664 [EXEC_DIRECTORY_CACHE
] = "CACHE_DIRECTORY",
6665 [EXEC_DIRECTORY_LOGS
] = "LOGS_DIRECTORY",
6666 [EXEC_DIRECTORY_CONFIGURATION
] = "CONFIGURATION_DIRECTORY",
6669 DEFINE_PRIVATE_STRING_TABLE_LOOKUP_TO_STRING(exec_directory_env_name
, ExecDirectoryType
);
6671 static const char* const exec_keyring_mode_table
[_EXEC_KEYRING_MODE_MAX
] = {
6672 [EXEC_KEYRING_INHERIT
] = "inherit",
6673 [EXEC_KEYRING_PRIVATE
] = "private",
6674 [EXEC_KEYRING_SHARED
] = "shared",
6677 DEFINE_STRING_TABLE_LOOKUP(exec_keyring_mode
, ExecKeyringMode
);