1 /* SPDX-License-Identifier: LGPL-2.1+ */
10 #include <sys/capability.h>
11 #include <sys/eventfd.h>
13 #include <sys/personality.h>
14 #include <sys/prctl.h>
16 #include <sys/socket.h>
18 #include <sys/types.h>
24 #include <security/pam_appl.h>
28 #include <selinux/selinux.h>
36 #include <sys/apparmor.h>
39 #include "sd-messages.h"
42 #include "alloc-util.h"
44 #include "apparmor-util.h"
49 #include "capability-util.h"
50 #include "chown-recursive.h"
51 #include "cpu-set-util.h"
55 #include "errno-list.h"
57 #include "exit-status.h"
59 #include "format-util.h"
61 #include "glob-util.h"
68 #include "memory-util.h"
71 #include "namespace.h"
72 #include "parse-util.h"
73 #include "path-util.h"
74 #include "process-util.h"
75 #include "rlimit-util.h"
78 #include "seccomp-util.h"
80 #include "securebits-util.h"
81 #include "selinux-util.h"
82 #include "signal-util.h"
83 #include "smack-util.h"
84 #include "socket-util.h"
86 #include "stat-util.h"
87 #include "string-table.h"
88 #include "string-util.h"
90 #include "syslog-util.h"
91 #include "terminal-util.h"
92 #include "umask-util.h"
94 #include "user-util.h"
95 #include "utmp-wtmp.h"
97 #define IDLE_TIMEOUT_USEC (5*USEC_PER_SEC)
98 #define IDLE_TIMEOUT2_USEC (1*USEC_PER_SEC)
100 #define SNDBUF_SIZE (8*1024*1024)
102 static int shift_fds(int fds
[], size_t n_fds
) {
103 int start
, restart_from
;
108 /* Modifies the fds array! (sorts it) */
118 for (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 (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_SYSLOG_AND_CONSOLE
,
228 EXEC_OUTPUT_KMSG_AND_CONSOLE
,
229 EXEC_OUTPUT_JOURNAL_AND_CONSOLE
);
232 static bool is_syslog_output(ExecOutput o
) {
235 EXEC_OUTPUT_SYSLOG_AND_CONSOLE
);
238 static bool is_kmsg_output(ExecOutput o
) {
241 EXEC_OUTPUT_KMSG_AND_CONSOLE
);
244 static bool exec_context_needs_term(const ExecContext
*c
) {
247 /* Return true if the execution context suggests we should set $TERM to something useful. */
249 if (is_terminal_input(c
->std_input
))
252 if (is_terminal_output(c
->std_output
))
255 if (is_terminal_output(c
->std_error
))
258 return !!c
->tty_path
;
261 static int open_null_as(int flags
, int nfd
) {
266 fd
= open("/dev/null", flags
|O_NOCTTY
);
270 return move_fd(fd
, nfd
, false);
273 static int connect_journal_socket(int fd
, uid_t uid
, gid_t gid
) {
274 static const union sockaddr_union sa
= {
275 .un
.sun_family
= AF_UNIX
,
276 .un
.sun_path
= "/run/systemd/journal/stdout",
278 uid_t olduid
= UID_INVALID
;
279 gid_t oldgid
= GID_INVALID
;
282 if (gid_is_valid(gid
)) {
285 if (setegid(gid
) < 0)
289 if (uid_is_valid(uid
)) {
292 if (seteuid(uid
) < 0) {
298 r
= connect(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
)) < 0 ? -errno
: 0;
300 /* If we fail to restore the uid or gid, things will likely
301 fail later on. This should only happen if an LSM interferes. */
303 if (uid_is_valid(uid
))
304 (void) seteuid(olduid
);
307 if (gid_is_valid(gid
))
308 (void) setegid(oldgid
);
313 static int connect_logger_as(
315 const ExecContext
*context
,
316 const ExecParameters
*params
,
323 _cleanup_close_
int fd
= -1;
328 assert(output
< _EXEC_OUTPUT_MAX
);
332 fd
= socket(AF_UNIX
, SOCK_STREAM
, 0);
336 r
= connect_journal_socket(fd
, uid
, gid
);
340 if (shutdown(fd
, SHUT_RD
) < 0)
343 (void) fd_inc_sndbuf(fd
, SNDBUF_SIZE
);
353 context
->syslog_identifier
?: ident
,
354 params
->flags
& EXEC_PASS_LOG_UNIT
? unit
->id
: "",
355 context
->syslog_priority
,
356 !!context
->syslog_level_prefix
,
357 is_syslog_output(output
),
358 is_kmsg_output(output
),
359 is_terminal_output(output
)) < 0)
362 return move_fd(TAKE_FD(fd
), nfd
, false);
365 static int open_terminal_as(const char *path
, int flags
, int nfd
) {
371 fd
= open_terminal(path
, flags
| O_NOCTTY
);
375 return move_fd(fd
, nfd
, false);
378 static int acquire_path(const char *path
, int flags
, mode_t mode
) {
379 union sockaddr_union sa
= {};
380 _cleanup_close_
int fd
= -1;
385 if (IN_SET(flags
& O_ACCMODE
, O_WRONLY
, O_RDWR
))
388 fd
= open(path
, flags
|O_NOCTTY
, mode
);
392 if (errno
!= ENXIO
) /* ENXIO is returned when we try to open() an AF_UNIX file system socket on Linux */
394 if (strlen(path
) >= sizeof(sa
.un
.sun_path
)) /* Too long, can't be a UNIX socket */
397 /* So, it appears the specified path could be an AF_UNIX socket. Let's see if we can connect to it. */
399 fd
= socket(AF_UNIX
, SOCK_STREAM
, 0);
403 salen
= sockaddr_un_set_path(&sa
.un
, path
);
407 if (connect(fd
, &sa
.sa
, salen
) < 0)
408 return errno
== EINVAL
? -ENXIO
: -errno
; /* Propagate initial error if we get EINVAL, i.e. we have
409 * indication that his wasn't an AF_UNIX socket after all */
411 if ((flags
& O_ACCMODE
) == O_RDONLY
)
412 r
= shutdown(fd
, SHUT_WR
);
413 else if ((flags
& O_ACCMODE
) == O_WRONLY
)
414 r
= shutdown(fd
, SHUT_RD
);
423 static int fixup_input(
424 const ExecContext
*context
,
426 bool apply_tty_stdin
) {
432 std_input
= context
->std_input
;
434 if (is_terminal_input(std_input
) && !apply_tty_stdin
)
435 return EXEC_INPUT_NULL
;
437 if (std_input
== EXEC_INPUT_SOCKET
&& socket_fd
< 0)
438 return EXEC_INPUT_NULL
;
440 if (std_input
== EXEC_INPUT_DATA
&& context
->stdin_data_size
== 0)
441 return EXEC_INPUT_NULL
;
446 static int fixup_output(ExecOutput std_output
, int socket_fd
) {
448 if (std_output
== EXEC_OUTPUT_SOCKET
&& socket_fd
< 0)
449 return EXEC_OUTPUT_INHERIT
;
454 static int setup_input(
455 const ExecContext
*context
,
456 const ExecParameters
*params
,
458 int named_iofds
[3]) {
465 if (params
->stdin_fd
>= 0) {
466 if (dup2(params
->stdin_fd
, STDIN_FILENO
) < 0)
469 /* Try to make this the controlling tty, if it is a tty, and reset it */
470 if (isatty(STDIN_FILENO
)) {
471 (void) ioctl(STDIN_FILENO
, TIOCSCTTY
, context
->std_input
== EXEC_INPUT_TTY_FORCE
);
472 (void) reset_terminal_fd(STDIN_FILENO
, true);
478 i
= fixup_input(context
, socket_fd
, params
->flags
& EXEC_APPLY_TTY_STDIN
);
482 case EXEC_INPUT_NULL
:
483 return open_null_as(O_RDONLY
, STDIN_FILENO
);
486 case EXEC_INPUT_TTY_FORCE
:
487 case EXEC_INPUT_TTY_FAIL
: {
490 fd
= acquire_terminal(exec_context_tty_path(context
),
491 i
== EXEC_INPUT_TTY_FAIL
? ACQUIRE_TERMINAL_TRY
:
492 i
== EXEC_INPUT_TTY_FORCE
? ACQUIRE_TERMINAL_FORCE
:
493 ACQUIRE_TERMINAL_WAIT
,
498 return move_fd(fd
, STDIN_FILENO
, false);
501 case EXEC_INPUT_SOCKET
:
502 assert(socket_fd
>= 0);
504 return dup2(socket_fd
, STDIN_FILENO
) < 0 ? -errno
: STDIN_FILENO
;
506 case EXEC_INPUT_NAMED_FD
:
507 assert(named_iofds
[STDIN_FILENO
] >= 0);
509 (void) fd_nonblock(named_iofds
[STDIN_FILENO
], false);
510 return dup2(named_iofds
[STDIN_FILENO
], STDIN_FILENO
) < 0 ? -errno
: STDIN_FILENO
;
512 case EXEC_INPUT_DATA
: {
515 fd
= acquire_data_fd(context
->stdin_data
, context
->stdin_data_size
, 0);
519 return move_fd(fd
, STDIN_FILENO
, false);
522 case EXEC_INPUT_FILE
: {
526 assert(context
->stdio_file
[STDIN_FILENO
]);
528 rw
= (context
->std_output
== EXEC_OUTPUT_FILE
&& streq_ptr(context
->stdio_file
[STDIN_FILENO
], context
->stdio_file
[STDOUT_FILENO
])) ||
529 (context
->std_error
== EXEC_OUTPUT_FILE
&& streq_ptr(context
->stdio_file
[STDIN_FILENO
], context
->stdio_file
[STDERR_FILENO
]));
531 fd
= acquire_path(context
->stdio_file
[STDIN_FILENO
], rw
? O_RDWR
: O_RDONLY
, 0666 & ~context
->umask
);
535 return move_fd(fd
, STDIN_FILENO
, false);
539 assert_not_reached("Unknown input type");
543 static bool can_inherit_stderr_from_stdout(
544 const ExecContext
*context
,
550 /* Returns true, if given the specified STDERR and STDOUT output we can directly dup() the stdout fd to the
553 if (e
== EXEC_OUTPUT_INHERIT
)
558 if (e
== EXEC_OUTPUT_NAMED_FD
)
559 return streq_ptr(context
->stdio_fdname
[STDOUT_FILENO
], context
->stdio_fdname
[STDERR_FILENO
]);
561 if (IN_SET(e
, EXEC_OUTPUT_FILE
, EXEC_OUTPUT_FILE_APPEND
))
562 return streq_ptr(context
->stdio_file
[STDOUT_FILENO
], context
->stdio_file
[STDERR_FILENO
]);
567 static int setup_output(
569 const ExecContext
*context
,
570 const ExecParameters
*params
,
577 dev_t
*journal_stream_dev
,
578 ino_t
*journal_stream_ino
) {
588 assert(journal_stream_dev
);
589 assert(journal_stream_ino
);
591 if (fileno
== STDOUT_FILENO
&& params
->stdout_fd
>= 0) {
593 if (dup2(params
->stdout_fd
, STDOUT_FILENO
) < 0)
596 return STDOUT_FILENO
;
599 if (fileno
== STDERR_FILENO
&& params
->stderr_fd
>= 0) {
600 if (dup2(params
->stderr_fd
, STDERR_FILENO
) < 0)
603 return STDERR_FILENO
;
606 i
= fixup_input(context
, socket_fd
, params
->flags
& EXEC_APPLY_TTY_STDIN
);
607 o
= fixup_output(context
->std_output
, socket_fd
);
609 if (fileno
== STDERR_FILENO
) {
611 e
= fixup_output(context
->std_error
, socket_fd
);
613 /* This expects the input and output are already set up */
615 /* Don't change the stderr file descriptor if we inherit all
616 * the way and are not on a tty */
617 if (e
== EXEC_OUTPUT_INHERIT
&&
618 o
== EXEC_OUTPUT_INHERIT
&&
619 i
== EXEC_INPUT_NULL
&&
620 !is_terminal_input(context
->std_input
) &&
624 /* Duplicate from stdout if possible */
625 if (can_inherit_stderr_from_stdout(context
, o
, e
))
626 return dup2(STDOUT_FILENO
, fileno
) < 0 ? -errno
: fileno
;
630 } else if (o
== EXEC_OUTPUT_INHERIT
) {
631 /* If input got downgraded, inherit the original value */
632 if (i
== EXEC_INPUT_NULL
&& is_terminal_input(context
->std_input
))
633 return open_terminal_as(exec_context_tty_path(context
), O_WRONLY
, fileno
);
635 /* If the input is connected to anything that's not a /dev/null or a data fd, inherit that... */
636 if (!IN_SET(i
, EXEC_INPUT_NULL
, EXEC_INPUT_DATA
))
637 return dup2(STDIN_FILENO
, fileno
) < 0 ? -errno
: fileno
;
639 /* If we are not started from PID 1 we just inherit STDOUT from our parent process. */
643 /* We need to open /dev/null here anew, to get the right access mode. */
644 return open_null_as(O_WRONLY
, fileno
);
649 case EXEC_OUTPUT_NULL
:
650 return open_null_as(O_WRONLY
, fileno
);
652 case EXEC_OUTPUT_TTY
:
653 if (is_terminal_input(i
))
654 return dup2(STDIN_FILENO
, fileno
) < 0 ? -errno
: fileno
;
656 /* We don't reset the terminal if this is just about output */
657 return open_terminal_as(exec_context_tty_path(context
), O_WRONLY
, fileno
);
659 case EXEC_OUTPUT_SYSLOG
:
660 case EXEC_OUTPUT_SYSLOG_AND_CONSOLE
:
661 case EXEC_OUTPUT_KMSG
:
662 case EXEC_OUTPUT_KMSG_AND_CONSOLE
:
663 case EXEC_OUTPUT_JOURNAL
:
664 case EXEC_OUTPUT_JOURNAL_AND_CONSOLE
:
665 r
= connect_logger_as(unit
, context
, params
, o
, ident
, fileno
, uid
, gid
);
667 log_unit_warning_errno(unit
, r
, "Failed to connect %s to the journal socket, ignoring: %m", fileno
== STDOUT_FILENO
? "stdout" : "stderr");
668 r
= open_null_as(O_WRONLY
, fileno
);
672 /* If we connected this fd to the journal via a stream, patch the device/inode into the passed
673 * parameters, but only then. This is useful so that we can set $JOURNAL_STREAM that permits
674 * services to detect whether they are connected to the journal or not.
676 * If both stdout and stderr are connected to a stream then let's make sure to store the data
677 * about STDERR as that's usually the best way to do logging. */
679 if (fstat(fileno
, &st
) >= 0 &&
680 (*journal_stream_ino
== 0 || fileno
== STDERR_FILENO
)) {
681 *journal_stream_dev
= st
.st_dev
;
682 *journal_stream_ino
= st
.st_ino
;
687 case EXEC_OUTPUT_SOCKET
:
688 assert(socket_fd
>= 0);
690 return dup2(socket_fd
, fileno
) < 0 ? -errno
: fileno
;
692 case EXEC_OUTPUT_NAMED_FD
:
693 assert(named_iofds
[fileno
] >= 0);
695 (void) fd_nonblock(named_iofds
[fileno
], false);
696 return dup2(named_iofds
[fileno
], fileno
) < 0 ? -errno
: fileno
;
698 case EXEC_OUTPUT_FILE
:
699 case EXEC_OUTPUT_FILE_APPEND
: {
703 assert(context
->stdio_file
[fileno
]);
705 rw
= context
->std_input
== EXEC_INPUT_FILE
&&
706 streq_ptr(context
->stdio_file
[fileno
], context
->stdio_file
[STDIN_FILENO
]);
709 return dup2(STDIN_FILENO
, fileno
) < 0 ? -errno
: fileno
;
712 if (o
== EXEC_OUTPUT_FILE_APPEND
)
715 fd
= acquire_path(context
->stdio_file
[fileno
], flags
, 0666 & ~context
->umask
);
719 return move_fd(fd
, fileno
, 0);
723 assert_not_reached("Unknown error type");
727 static int chown_terminal(int fd
, uid_t uid
) {
732 /* Before we chown/chmod the TTY, let's ensure this is actually a tty */
733 if (isatty(fd
) < 1) {
734 if (IN_SET(errno
, EINVAL
, ENOTTY
))
735 return 0; /* not a tty */
740 /* This might fail. What matters are the results. */
741 r
= fchmod_and_chown(fd
, TTY_MODE
, uid
, -1);
748 static int setup_confirm_stdio(const char *vc
, int *_saved_stdin
, int *_saved_stdout
) {
749 _cleanup_close_
int fd
= -1, saved_stdin
= -1, saved_stdout
= -1;
752 assert(_saved_stdin
);
753 assert(_saved_stdout
);
755 saved_stdin
= fcntl(STDIN_FILENO
, F_DUPFD
, 3);
759 saved_stdout
= fcntl(STDOUT_FILENO
, F_DUPFD
, 3);
760 if (saved_stdout
< 0)
763 fd
= acquire_terminal(vc
, ACQUIRE_TERMINAL_WAIT
, DEFAULT_CONFIRM_USEC
);
767 r
= chown_terminal(fd
, getuid());
771 r
= reset_terminal_fd(fd
, true);
775 r
= rearrange_stdio(fd
, fd
, STDERR_FILENO
);
780 *_saved_stdin
= saved_stdin
;
781 *_saved_stdout
= saved_stdout
;
783 saved_stdin
= saved_stdout
= -1;
788 static void write_confirm_error_fd(int err
, int fd
, const Unit
*u
) {
791 if (err
== -ETIMEDOUT
)
792 dprintf(fd
, "Confirmation question timed out for %s, assuming positive response.\n", u
->id
);
795 dprintf(fd
, "Couldn't ask confirmation for %s: %m, assuming positive response.\n", u
->id
);
799 static void write_confirm_error(int err
, const char *vc
, const Unit
*u
) {
800 _cleanup_close_
int fd
= -1;
804 fd
= open_terminal(vc
, O_WRONLY
|O_NOCTTY
|O_CLOEXEC
);
808 write_confirm_error_fd(err
, fd
, u
);
811 static int restore_confirm_stdio(int *saved_stdin
, int *saved_stdout
) {
815 assert(saved_stdout
);
819 if (*saved_stdin
>= 0)
820 if (dup2(*saved_stdin
, STDIN_FILENO
) < 0)
823 if (*saved_stdout
>= 0)
824 if (dup2(*saved_stdout
, STDOUT_FILENO
) < 0)
827 *saved_stdin
= safe_close(*saved_stdin
);
828 *saved_stdout
= safe_close(*saved_stdout
);
834 CONFIRM_PRETEND_FAILURE
= -1,
835 CONFIRM_PRETEND_SUCCESS
= 0,
839 static int ask_for_confirmation(const char *vc
, Unit
*u
, const char *cmdline
) {
840 int saved_stdout
= -1, saved_stdin
= -1, r
;
841 _cleanup_free_
char *e
= NULL
;
844 /* For any internal errors, assume a positive response. */
845 r
= setup_confirm_stdio(vc
, &saved_stdin
, &saved_stdout
);
847 write_confirm_error(r
, vc
, u
);
848 return CONFIRM_EXECUTE
;
851 /* confirm_spawn might have been disabled while we were sleeping. */
852 if (manager_is_confirm_spawn_disabled(u
->manager
)) {
857 e
= ellipsize(cmdline
, 60, 100);
865 r
= ask_char(&c
, "yfshiDjcn", "Execute %s? [y, f, s – h for help] ", e
);
867 write_confirm_error_fd(r
, STDOUT_FILENO
, u
);
874 printf("Resuming normal execution.\n");
875 manager_disable_confirm_spawn();
879 unit_dump(u
, stdout
, " ");
880 continue; /* ask again */
882 printf("Failing execution.\n");
883 r
= CONFIRM_PRETEND_FAILURE
;
886 printf(" c - continue, proceed without asking anymore\n"
887 " D - dump, show the state of the unit\n"
888 " f - fail, don't execute the command and pretend it failed\n"
890 " i - info, show a short summary of the unit\n"
891 " j - jobs, show jobs that are in progress\n"
892 " s - skip, don't execute the command and pretend it succeeded\n"
893 " y - yes, execute the command\n");
894 continue; /* ask again */
896 printf(" Description: %s\n"
899 u
->id
, u
->description
, cmdline
);
900 continue; /* ask again */
902 manager_dump_jobs(u
->manager
, stdout
, " ");
903 continue; /* ask again */
905 /* 'n' was removed in favor of 'f'. */
906 printf("Didn't understand 'n', did you mean 'f'?\n");
907 continue; /* ask again */
909 printf("Skipping execution.\n");
910 r
= CONFIRM_PRETEND_SUCCESS
;
916 assert_not_reached("Unhandled choice");
922 restore_confirm_stdio(&saved_stdin
, &saved_stdout
);
926 static int get_fixed_user(const ExecContext
*c
, const char **user
,
927 uid_t
*uid
, gid_t
*gid
,
928 const char **home
, const char **shell
) {
937 /* Note that we don't set $HOME or $SHELL if they are not particularly enlightening anyway
938 * (i.e. are "/" or "/bin/nologin"). */
941 r
= get_user_creds(&name
, uid
, gid
, home
, shell
, USER_CREDS_CLEAN
);
949 static int get_fixed_group(const ExecContext
*c
, const char **group
, gid_t
*gid
) {
959 r
= get_group_creds(&name
, gid
, 0);
967 static int get_supplementary_groups(const ExecContext
*c
, const char *user
,
968 const char *group
, gid_t gid
,
969 gid_t
**supplementary_gids
, int *ngids
) {
973 bool keep_groups
= false;
974 gid_t
*groups
= NULL
;
975 _cleanup_free_ gid_t
*l_gids
= NULL
;
980 * If user is given, then lookup GID and supplementary groups list.
981 * We avoid NSS lookups for gid=0. Also we have to initialize groups
982 * here and as early as possible so we keep the list of supplementary
983 * groups of the caller.
985 if (user
&& gid_is_valid(gid
) && gid
!= 0) {
986 /* First step, initialize groups from /etc/groups */
987 if (initgroups(user
, gid
) < 0)
993 if (strv_isempty(c
->supplementary_groups
))
997 * If SupplementaryGroups= was passed then NGROUPS_MAX has to
998 * be positive, otherwise fail.
1001 ngroups_max
= (int) sysconf(_SC_NGROUPS_MAX
);
1002 if (ngroups_max
<= 0) {
1006 return -EOPNOTSUPP
; /* For all other values */
1009 l_gids
= new(gid_t
, ngroups_max
);
1015 * Lookup the list of groups that the user belongs to, we
1016 * avoid NSS lookups here too for gid=0.
1019 if (getgrouplist(user
, gid
, l_gids
, &k
) < 0)
1024 STRV_FOREACH(i
, c
->supplementary_groups
) {
1027 if (k
>= ngroups_max
)
1031 r
= get_group_creds(&g
, l_gids
+k
, 0);
1039 * Sets ngids to zero to drop all supplementary groups, happens
1040 * when we are under root and SupplementaryGroups= is empty.
1047 /* Otherwise get the final list of supplementary groups */
1048 groups
= memdup(l_gids
, sizeof(gid_t
) * k
);
1052 *supplementary_gids
= groups
;
1060 static int enforce_groups(gid_t gid
, const gid_t
*supplementary_gids
, int ngids
) {
1063 /* Handle SupplementaryGroups= if it is not empty */
1065 r
= maybe_setgroups(ngids
, supplementary_gids
);
1070 if (gid_is_valid(gid
)) {
1071 /* Then set our gids */
1072 if (setresgid(gid
, gid
, gid
) < 0)
1079 static int enforce_user(const ExecContext
*context
, uid_t uid
) {
1082 if (!uid_is_valid(uid
))
1085 /* Sets (but doesn't look up) the uid and make sure we keep the
1086 * capabilities while doing so. */
1088 if (context
->capability_ambient_set
!= 0) {
1090 /* First step: If we need to keep capabilities but
1091 * drop privileges we need to make sure we keep our
1092 * caps, while we drop privileges. */
1094 int sb
= context
->secure_bits
| 1<<SECURE_KEEP_CAPS
;
1096 if (prctl(PR_GET_SECUREBITS
) != sb
)
1097 if (prctl(PR_SET_SECUREBITS
, sb
) < 0)
1102 /* Second step: actually set the uids */
1103 if (setresuid(uid
, uid
, uid
) < 0)
1106 /* At this point we should have all necessary capabilities but
1107 are otherwise a normal user. However, the caps might got
1108 corrupted due to the setresuid() so we need clean them up
1109 later. This is done outside of this call. */
1116 static int null_conv(
1118 const struct pam_message
**msg
,
1119 struct pam_response
**resp
,
1120 void *appdata_ptr
) {
1122 /* We don't support conversations */
1124 return PAM_CONV_ERR
;
1129 static int setup_pam(
1136 int fds
[], size_t n_fds
) {
1140 static const struct pam_conv conv
= {
1145 _cleanup_(barrier_destroy
) Barrier barrier
= BARRIER_NULL
;
1146 pam_handle_t
*handle
= NULL
;
1148 int pam_code
= PAM_SUCCESS
, r
;
1149 char **nv
, **e
= NULL
;
1150 bool close_session
= false;
1151 pid_t pam_pid
= 0, parent_pid
;
1158 /* We set up PAM in the parent process, then fork. The child
1159 * will then stay around until killed via PR_GET_PDEATHSIG or
1160 * systemd via the cgroup logic. It will then remove the PAM
1161 * session again. The parent process will exec() the actual
1162 * daemon. We do things this way to ensure that the main PID
1163 * of the daemon is the one we initially fork()ed. */
1165 r
= barrier_create(&barrier
);
1169 if (log_get_max_level() < LOG_DEBUG
)
1170 flags
|= PAM_SILENT
;
1172 pam_code
= pam_start(name
, user
, &conv
, &handle
);
1173 if (pam_code
!= PAM_SUCCESS
) {
1179 _cleanup_free_
char *q
= NULL
;
1181 /* Hmm, so no TTY was explicitly passed, but an fd passed to us directly might be a TTY. Let's figure
1182 * out if that's the case, and read the TTY off it. */
1184 if (getttyname_malloc(STDIN_FILENO
, &q
) >= 0)
1185 tty
= strjoina("/dev/", q
);
1189 pam_code
= pam_set_item(handle
, PAM_TTY
, tty
);
1190 if (pam_code
!= PAM_SUCCESS
)
1194 STRV_FOREACH(nv
, *env
) {
1195 pam_code
= pam_putenv(handle
, *nv
);
1196 if (pam_code
!= PAM_SUCCESS
)
1200 pam_code
= pam_acct_mgmt(handle
, flags
);
1201 if (pam_code
!= PAM_SUCCESS
)
1204 pam_code
= pam_open_session(handle
, flags
);
1205 if (pam_code
!= PAM_SUCCESS
)
1208 close_session
= true;
1210 e
= pam_getenvlist(handle
);
1212 pam_code
= PAM_BUF_ERR
;
1216 /* Block SIGTERM, so that we know that it won't get lost in
1219 assert_se(sigprocmask_many(SIG_BLOCK
, &old_ss
, SIGTERM
, -1) >= 0);
1221 parent_pid
= getpid_cached();
1223 r
= safe_fork("(sd-pam)", 0, &pam_pid
);
1227 int sig
, ret
= EXIT_PAM
;
1229 /* The child's job is to reset the PAM session on
1231 barrier_set_role(&barrier
, BARRIER_CHILD
);
1233 /* Make sure we don't keep open the passed fds in this child. We assume that otherwise only those fds
1234 * are open here that have been opened by PAM. */
1235 (void) close_many(fds
, n_fds
);
1237 /* Drop privileges - we don't need any to pam_close_session
1238 * and this will make PR_SET_PDEATHSIG work in most cases.
1239 * If this fails, ignore the error - but expect sd-pam threads
1240 * to fail to exit normally */
1242 r
= maybe_setgroups(0, NULL
);
1244 log_warning_errno(r
, "Failed to setgroups() in sd-pam: %m");
1245 if (setresgid(gid
, gid
, gid
) < 0)
1246 log_warning_errno(errno
, "Failed to setresgid() in sd-pam: %m");
1247 if (setresuid(uid
, uid
, uid
) < 0)
1248 log_warning_errno(errno
, "Failed to setresuid() in sd-pam: %m");
1250 (void) ignore_signals(SIGPIPE
, -1);
1252 /* Wait until our parent died. This will only work if
1253 * the above setresuid() succeeds, otherwise the kernel
1254 * will not allow unprivileged parents kill their privileged
1255 * children this way. We rely on the control groups kill logic
1256 * to do the rest for us. */
1257 if (prctl(PR_SET_PDEATHSIG
, SIGTERM
) < 0)
1260 /* Tell the parent that our setup is done. This is especially
1261 * important regarding dropping privileges. Otherwise, unit
1262 * setup might race against our setresuid(2) call.
1264 * If the parent aborted, we'll detect this below, hence ignore
1265 * return failure here. */
1266 (void) barrier_place(&barrier
);
1268 /* Check if our parent process might already have died? */
1269 if (getppid() == parent_pid
) {
1272 assert_se(sigemptyset(&ss
) >= 0);
1273 assert_se(sigaddset(&ss
, SIGTERM
) >= 0);
1276 if (sigwait(&ss
, &sig
) < 0) {
1283 assert(sig
== SIGTERM
);
1288 /* If our parent died we'll end the session */
1289 if (getppid() != parent_pid
) {
1290 pam_code
= pam_close_session(handle
, flags
);
1291 if (pam_code
!= PAM_SUCCESS
)
1298 pam_end(handle
, pam_code
| flags
);
1302 barrier_set_role(&barrier
, BARRIER_PARENT
);
1304 /* If the child was forked off successfully it will do all the
1305 * cleanups, so forget about the handle here. */
1308 /* Unblock SIGTERM again in the parent */
1309 assert_se(sigprocmask(SIG_SETMASK
, &old_ss
, NULL
) >= 0);
1311 /* We close the log explicitly here, since the PAM modules
1312 * might have opened it, but we don't want this fd around. */
1315 /* Synchronously wait for the child to initialize. We don't care for
1316 * errors as we cannot recover. However, warn loudly if it happens. */
1317 if (!barrier_place_and_sync(&barrier
))
1318 log_error("PAM initialization failed");
1320 return strv_free_and_replace(*env
, e
);
1323 if (pam_code
!= PAM_SUCCESS
) {
1324 log_error("PAM failed: %s", pam_strerror(handle
, pam_code
));
1325 r
= -EPERM
; /* PAM errors do not map to errno */
1327 log_error_errno(r
, "PAM failed: %m");
1331 pam_code
= pam_close_session(handle
, flags
);
1333 pam_end(handle
, pam_code
| flags
);
1345 static void rename_process_from_path(const char *path
) {
1346 char process_name
[11];
1350 /* This resulting string must fit in 10 chars (i.e. the length
1351 * of "/sbin/init") to look pretty in /bin/ps */
1355 rename_process("(...)");
1361 /* The end of the process name is usually more
1362 * interesting, since the first bit might just be
1368 process_name
[0] = '(';
1369 memcpy(process_name
+1, p
, l
);
1370 process_name
[1+l
] = ')';
1371 process_name
[1+l
+1] = 0;
1373 rename_process(process_name
);
1376 static bool context_has_address_families(const ExecContext
*c
) {
1379 return c
->address_families_whitelist
||
1380 !set_isempty(c
->address_families
);
1383 static bool context_has_syscall_filters(const ExecContext
*c
) {
1386 return c
->syscall_whitelist
||
1387 !hashmap_isempty(c
->syscall_filter
);
1390 static bool context_has_no_new_privileges(const ExecContext
*c
) {
1393 if (c
->no_new_privileges
)
1396 if (have_effective_cap(CAP_SYS_ADMIN
)) /* if we are privileged, we don't need NNP */
1399 /* We need NNP if we have any form of seccomp and are unprivileged */
1400 return context_has_address_families(c
) ||
1401 c
->memory_deny_write_execute
||
1402 c
->restrict_realtime
||
1403 c
->restrict_suid_sgid
||
1404 exec_context_restrict_namespaces_set(c
) ||
1405 c
->protect_kernel_tunables
||
1406 c
->protect_kernel_modules
||
1407 c
->private_devices
||
1408 context_has_syscall_filters(c
) ||
1409 !set_isempty(c
->syscall_archs
) ||
1410 c
->lock_personality
||
1411 c
->protect_hostname
;
1416 static bool skip_seccomp_unavailable(const Unit
* u
, const char* msg
) {
1418 if (is_seccomp_available())
1421 log_unit_debug(u
, "SECCOMP features not detected in the kernel, skipping %s", msg
);
1425 static int apply_syscall_filter(const Unit
* u
, const ExecContext
*c
, bool needs_ambient_hack
) {
1426 uint32_t negative_action
, default_action
, action
;
1432 if (!context_has_syscall_filters(c
))
1435 if (skip_seccomp_unavailable(u
, "SystemCallFilter="))
1438 negative_action
= c
->syscall_errno
== 0 ? scmp_act_kill_process() : SCMP_ACT_ERRNO(c
->syscall_errno
);
1440 if (c
->syscall_whitelist
) {
1441 default_action
= negative_action
;
1442 action
= SCMP_ACT_ALLOW
;
1444 default_action
= SCMP_ACT_ALLOW
;
1445 action
= negative_action
;
1448 if (needs_ambient_hack
) {
1449 r
= seccomp_filter_set_add(c
->syscall_filter
, c
->syscall_whitelist
, syscall_filter_sets
+ SYSCALL_FILTER_SET_SETUID
);
1454 return seccomp_load_syscall_filter_set_raw(default_action
, c
->syscall_filter
, action
, false);
1457 static int apply_syscall_archs(const Unit
*u
, const ExecContext
*c
) {
1461 if (set_isempty(c
->syscall_archs
))
1464 if (skip_seccomp_unavailable(u
, "SystemCallArchitectures="))
1467 return seccomp_restrict_archs(c
->syscall_archs
);
1470 static int apply_address_families(const Unit
* u
, const ExecContext
*c
) {
1474 if (!context_has_address_families(c
))
1477 if (skip_seccomp_unavailable(u
, "RestrictAddressFamilies="))
1480 return seccomp_restrict_address_families(c
->address_families
, c
->address_families_whitelist
);
1483 static int apply_memory_deny_write_execute(const Unit
* u
, const ExecContext
*c
) {
1487 if (!c
->memory_deny_write_execute
)
1490 if (skip_seccomp_unavailable(u
, "MemoryDenyWriteExecute="))
1493 return seccomp_memory_deny_write_execute();
1496 static int apply_restrict_realtime(const Unit
* u
, const ExecContext
*c
) {
1500 if (!c
->restrict_realtime
)
1503 if (skip_seccomp_unavailable(u
, "RestrictRealtime="))
1506 return seccomp_restrict_realtime();
1509 static int apply_restrict_suid_sgid(const Unit
* u
, const ExecContext
*c
) {
1513 if (!c
->restrict_suid_sgid
)
1516 if (skip_seccomp_unavailable(u
, "RestrictSUIDSGID="))
1519 return seccomp_restrict_suid_sgid();
1522 static int apply_protect_sysctl(const Unit
*u
, const ExecContext
*c
) {
1526 /* Turn off the legacy sysctl() system call. Many distributions turn this off while building the kernel, but
1527 * let's protect even those systems where this is left on in the kernel. */
1529 if (!c
->protect_kernel_tunables
)
1532 if (skip_seccomp_unavailable(u
, "ProtectKernelTunables="))
1535 return seccomp_protect_sysctl();
1538 static int apply_protect_kernel_modules(const Unit
*u
, const ExecContext
*c
) {
1542 /* Turn off module syscalls on ProtectKernelModules=yes */
1544 if (!c
->protect_kernel_modules
)
1547 if (skip_seccomp_unavailable(u
, "ProtectKernelModules="))
1550 return seccomp_load_syscall_filter_set(SCMP_ACT_ALLOW
, syscall_filter_sets
+ SYSCALL_FILTER_SET_MODULE
, SCMP_ACT_ERRNO(EPERM
), false);
1553 static int apply_private_devices(const Unit
*u
, const ExecContext
*c
) {
1557 /* If PrivateDevices= is set, also turn off iopl and all @raw-io syscalls. */
1559 if (!c
->private_devices
)
1562 if (skip_seccomp_unavailable(u
, "PrivateDevices="))
1565 return seccomp_load_syscall_filter_set(SCMP_ACT_ALLOW
, syscall_filter_sets
+ SYSCALL_FILTER_SET_RAW_IO
, SCMP_ACT_ERRNO(EPERM
), false);
1568 static int apply_restrict_namespaces(const Unit
*u
, const ExecContext
*c
) {
1572 if (!exec_context_restrict_namespaces_set(c
))
1575 if (skip_seccomp_unavailable(u
, "RestrictNamespaces="))
1578 return seccomp_restrict_namespaces(c
->restrict_namespaces
);
1581 static int apply_lock_personality(const Unit
* u
, const ExecContext
*c
) {
1582 unsigned long personality
;
1588 if (!c
->lock_personality
)
1591 if (skip_seccomp_unavailable(u
, "LockPersonality="))
1594 personality
= c
->personality
;
1596 /* If personality is not specified, use either PER_LINUX or PER_LINUX32 depending on what is currently set. */
1597 if (personality
== PERSONALITY_INVALID
) {
1599 r
= opinionated_personality(&personality
);
1604 return seccomp_lock_personality(personality
);
1609 static void do_idle_pipe_dance(int idle_pipe
[static 4]) {
1612 idle_pipe
[1] = safe_close(idle_pipe
[1]);
1613 idle_pipe
[2] = safe_close(idle_pipe
[2]);
1615 if (idle_pipe
[0] >= 0) {
1618 r
= fd_wait_for_event(idle_pipe
[0], POLLHUP
, IDLE_TIMEOUT_USEC
);
1620 if (idle_pipe
[3] >= 0 && r
== 0 /* timeout */) {
1623 /* Signal systemd that we are bored and want to continue. */
1624 n
= write(idle_pipe
[3], "x", 1);
1626 /* Wait for systemd to react to the signal above. */
1627 fd_wait_for_event(idle_pipe
[0], POLLHUP
, IDLE_TIMEOUT2_USEC
);
1630 idle_pipe
[0] = safe_close(idle_pipe
[0]);
1634 idle_pipe
[3] = safe_close(idle_pipe
[3]);
1637 static const char *exec_directory_env_name_to_string(ExecDirectoryType t
);
1639 static int build_environment(
1641 const ExecContext
*c
,
1642 const ExecParameters
*p
,
1645 const char *username
,
1647 dev_t journal_stream_dev
,
1648 ino_t journal_stream_ino
,
1651 _cleanup_strv_free_
char **our_env
= NULL
;
1652 ExecDirectoryType t
;
1661 our_env
= new0(char*, 14 + _EXEC_DIRECTORY_TYPE_MAX
);
1666 _cleanup_free_
char *joined
= NULL
;
1668 if (asprintf(&x
, "LISTEN_PID="PID_FMT
, getpid_cached()) < 0)
1670 our_env
[n_env
++] = x
;
1672 if (asprintf(&x
, "LISTEN_FDS=%zu", n_fds
) < 0)
1674 our_env
[n_env
++] = x
;
1676 joined
= strv_join(p
->fd_names
, ":");
1680 x
= strjoin("LISTEN_FDNAMES=", joined
);
1683 our_env
[n_env
++] = x
;
1686 if ((p
->flags
& EXEC_SET_WATCHDOG
) && p
->watchdog_usec
> 0) {
1687 if (asprintf(&x
, "WATCHDOG_PID="PID_FMT
, getpid_cached()) < 0)
1689 our_env
[n_env
++] = x
;
1691 if (asprintf(&x
, "WATCHDOG_USEC="USEC_FMT
, p
->watchdog_usec
) < 0)
1693 our_env
[n_env
++] = x
;
1696 /* If this is D-Bus, tell the nss-systemd module, since it relies on being able to use D-Bus look up dynamic
1697 * users via PID 1, possibly dead-locking the dbus daemon. This way it will not use D-Bus to resolve names, but
1698 * check the database directly. */
1699 if (p
->flags
& EXEC_NSS_BYPASS_BUS
) {
1700 x
= strdup("SYSTEMD_NSS_BYPASS_BUS=1");
1703 our_env
[n_env
++] = x
;
1707 x
= strappend("HOME=", home
);
1711 path_simplify(x
+ 5, true);
1712 our_env
[n_env
++] = x
;
1716 x
= strappend("LOGNAME=", username
);
1719 our_env
[n_env
++] = x
;
1721 x
= strappend("USER=", username
);
1724 our_env
[n_env
++] = x
;
1728 x
= strappend("SHELL=", shell
);
1732 path_simplify(x
+ 6, true);
1733 our_env
[n_env
++] = x
;
1736 if (!sd_id128_is_null(u
->invocation_id
)) {
1737 if (asprintf(&x
, "INVOCATION_ID=" SD_ID128_FORMAT_STR
, SD_ID128_FORMAT_VAL(u
->invocation_id
)) < 0)
1740 our_env
[n_env
++] = x
;
1743 if (exec_context_needs_term(c
)) {
1744 const char *tty_path
, *term
= NULL
;
1746 tty_path
= exec_context_tty_path(c
);
1748 /* If we are forked off PID 1 and we are supposed to operate on /dev/console, then let's try to inherit
1749 * the $TERM set for PID 1. This is useful for containers so that the $TERM the container manager
1750 * passes to PID 1 ends up all the way in the console login shown. */
1752 if (path_equal(tty_path
, "/dev/console") && getppid() == 1)
1753 term
= getenv("TERM");
1755 term
= default_term_for_tty(tty_path
);
1757 x
= strappend("TERM=", term
);
1760 our_env
[n_env
++] = x
;
1763 if (journal_stream_dev
!= 0 && journal_stream_ino
!= 0) {
1764 if (asprintf(&x
, "JOURNAL_STREAM=" DEV_FMT
":" INO_FMT
, journal_stream_dev
, journal_stream_ino
) < 0)
1767 our_env
[n_env
++] = x
;
1770 for (t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
1771 _cleanup_free_
char *pre
= NULL
, *joined
= NULL
;
1777 if (strv_isempty(c
->directories
[t
].paths
))
1780 n
= exec_directory_env_name_to_string(t
);
1784 pre
= strjoin(p
->prefix
[t
], "/");
1788 joined
= strv_join_prefix(c
->directories
[t
].paths
, ":", pre
);
1792 x
= strjoin(n
, "=", joined
);
1796 our_env
[n_env
++] = x
;
1799 our_env
[n_env
++] = NULL
;
1800 assert(n_env
<= 14 + _EXEC_DIRECTORY_TYPE_MAX
);
1802 *ret
= TAKE_PTR(our_env
);
1807 static int build_pass_environment(const ExecContext
*c
, char ***ret
) {
1808 _cleanup_strv_free_
char **pass_env
= NULL
;
1809 size_t n_env
= 0, n_bufsize
= 0;
1812 STRV_FOREACH(i
, c
->pass_environment
) {
1813 _cleanup_free_
char *x
= NULL
;
1819 x
= strjoin(*i
, "=", v
);
1823 if (!GREEDY_REALLOC(pass_env
, n_bufsize
, n_env
+ 2))
1826 pass_env
[n_env
++] = TAKE_PTR(x
);
1827 pass_env
[n_env
] = NULL
;
1830 *ret
= TAKE_PTR(pass_env
);
1835 static bool exec_needs_mount_namespace(
1836 const ExecContext
*context
,
1837 const ExecParameters
*params
,
1838 const ExecRuntime
*runtime
) {
1843 if (context
->root_image
)
1846 if (!strv_isempty(context
->read_write_paths
) ||
1847 !strv_isempty(context
->read_only_paths
) ||
1848 !strv_isempty(context
->inaccessible_paths
))
1851 if (context
->n_bind_mounts
> 0)
1854 if (context
->n_temporary_filesystems
> 0)
1857 if (!IN_SET(context
->mount_flags
, 0, MS_SHARED
))
1860 if (context
->private_tmp
&& runtime
&& (runtime
->tmp_dir
|| runtime
->var_tmp_dir
))
1863 if (context
->private_devices
||
1864 context
->private_mounts
||
1865 context
->protect_system
!= PROTECT_SYSTEM_NO
||
1866 context
->protect_home
!= PROTECT_HOME_NO
||
1867 context
->protect_kernel_tunables
||
1868 context
->protect_kernel_modules
||
1869 context
->protect_control_groups
)
1872 if (context
->root_directory
) {
1873 ExecDirectoryType t
;
1875 if (context
->mount_apivfs
)
1878 for (t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
1879 if (!params
->prefix
[t
])
1882 if (!strv_isempty(context
->directories
[t
].paths
))
1887 if (context
->dynamic_user
&&
1888 (!strv_isempty(context
->directories
[EXEC_DIRECTORY_STATE
].paths
) ||
1889 !strv_isempty(context
->directories
[EXEC_DIRECTORY_CACHE
].paths
) ||
1890 !strv_isempty(context
->directories
[EXEC_DIRECTORY_LOGS
].paths
)))
1896 static int setup_private_users(uid_t uid
, gid_t gid
) {
1897 _cleanup_free_
char *uid_map
= NULL
, *gid_map
= NULL
;
1898 _cleanup_close_pair_
int errno_pipe
[2] = { -1, -1 };
1899 _cleanup_close_
int unshare_ready_fd
= -1;
1900 _cleanup_(sigkill_waitp
) pid_t pid
= 0;
1905 /* Set up a user namespace and map root to root, the selected UID/GID to itself, and everything else to
1906 * nobody. In order to be able to write this mapping we need CAP_SETUID in the original user namespace, which
1907 * we however lack after opening the user namespace. To work around this we fork() a temporary child process,
1908 * which waits for the parent to create the new user namespace while staying in the original namespace. The
1909 * child then writes the UID mapping, under full privileges. The parent waits for the child to finish and
1910 * continues execution normally. */
1912 if (uid
!= 0 && uid_is_valid(uid
)) {
1913 r
= asprintf(&uid_map
,
1914 "0 0 1\n" /* Map root → root */
1915 UID_FMT
" " UID_FMT
" 1\n", /* Map $UID → $UID */
1920 uid_map
= strdup("0 0 1\n"); /* The case where the above is the same */
1925 if (gid
!= 0 && gid_is_valid(gid
)) {
1926 r
= asprintf(&gid_map
,
1927 "0 0 1\n" /* Map root → root */
1928 GID_FMT
" " GID_FMT
" 1\n", /* Map $GID → $GID */
1933 gid_map
= strdup("0 0 1\n"); /* The case where the above is the same */
1938 /* Create a communication channel so that the parent can tell the child when it finished creating the user
1940 unshare_ready_fd
= eventfd(0, EFD_CLOEXEC
);
1941 if (unshare_ready_fd
< 0)
1944 /* Create a communication channel so that the child can tell the parent a proper error code in case it
1946 if (pipe2(errno_pipe
, O_CLOEXEC
) < 0)
1949 r
= safe_fork("(sd-userns)", FORK_RESET_SIGNALS
|FORK_DEATHSIG
, &pid
);
1953 _cleanup_close_
int fd
= -1;
1957 /* Child process, running in the original user namespace. Let's update the parent's UID/GID map from
1958 * here, after the parent opened its own user namespace. */
1961 errno_pipe
[0] = safe_close(errno_pipe
[0]);
1963 /* Wait until the parent unshared the user namespace */
1964 if (read(unshare_ready_fd
, &c
, sizeof(c
)) < 0) {
1969 /* Disable the setgroups() system call in the child user namespace, for good. */
1970 a
= procfs_file_alloca(ppid
, "setgroups");
1971 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
1973 if (errno
!= ENOENT
) {
1978 /* If the file is missing the kernel is too old, let's continue anyway. */
1980 if (write(fd
, "deny\n", 5) < 0) {
1985 fd
= safe_close(fd
);
1988 /* First write the GID map */
1989 a
= procfs_file_alloca(ppid
, "gid_map");
1990 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
1995 if (write(fd
, gid_map
, strlen(gid_map
)) < 0) {
1999 fd
= safe_close(fd
);
2001 /* The write the UID map */
2002 a
= procfs_file_alloca(ppid
, "uid_map");
2003 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
2008 if (write(fd
, uid_map
, strlen(uid_map
)) < 0) {
2013 _exit(EXIT_SUCCESS
);
2016 (void) write(errno_pipe
[1], &r
, sizeof(r
));
2017 _exit(EXIT_FAILURE
);
2020 errno_pipe
[1] = safe_close(errno_pipe
[1]);
2022 if (unshare(CLONE_NEWUSER
) < 0)
2025 /* Let the child know that the namespace is ready now */
2026 if (write(unshare_ready_fd
, &c
, sizeof(c
)) < 0)
2029 /* Try to read an error code from the child */
2030 n
= read(errno_pipe
[0], &r
, sizeof(r
));
2033 if (n
== sizeof(r
)) { /* an error code was sent to us */
2038 if (n
!= 0) /* on success we should have read 0 bytes */
2041 r
= wait_for_terminate_and_check("(sd-userns)", pid
, 0);
2045 if (r
!= EXIT_SUCCESS
) /* If something strange happened with the child, let's consider this fatal, too */
2051 static int setup_exec_directory(
2052 const ExecContext
*context
,
2053 const ExecParameters
*params
,
2056 ExecDirectoryType type
,
2059 static const int exit_status_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
2060 [EXEC_DIRECTORY_RUNTIME
] = EXIT_RUNTIME_DIRECTORY
,
2061 [EXEC_DIRECTORY_STATE
] = EXIT_STATE_DIRECTORY
,
2062 [EXEC_DIRECTORY_CACHE
] = EXIT_CACHE_DIRECTORY
,
2063 [EXEC_DIRECTORY_LOGS
] = EXIT_LOGS_DIRECTORY
,
2064 [EXEC_DIRECTORY_CONFIGURATION
] = EXIT_CONFIGURATION_DIRECTORY
,
2071 assert(type
>= 0 && type
< _EXEC_DIRECTORY_TYPE_MAX
);
2072 assert(exit_status
);
2074 if (!params
->prefix
[type
])
2077 if (params
->flags
& EXEC_CHOWN_DIRECTORIES
) {
2078 if (!uid_is_valid(uid
))
2080 if (!gid_is_valid(gid
))
2084 STRV_FOREACH(rt
, context
->directories
[type
].paths
) {
2085 _cleanup_free_
char *p
= NULL
, *pp
= NULL
;
2087 p
= path_join(params
->prefix
[type
], *rt
);
2093 r
= mkdir_parents_label(p
, 0755);
2097 if (context
->dynamic_user
&&
2098 (!IN_SET(type
, EXEC_DIRECTORY_RUNTIME
, EXEC_DIRECTORY_CONFIGURATION
) ||
2099 (type
== EXEC_DIRECTORY_RUNTIME
&& context
->runtime_directory_preserve_mode
!= EXEC_PRESERVE_NO
))) {
2100 _cleanup_free_
char *private_root
= NULL
;
2102 /* So, here's one extra complication when dealing with DynamicUser=1 units. In that case we
2103 * want to avoid leaving a directory around fully accessible that is owned by a dynamic user
2104 * whose UID is later on reused. To lock this down we use the same trick used by container
2105 * managers to prohibit host users to get access to files of the same UID in containers: we
2106 * place everything inside a directory that has an access mode of 0700 and is owned root:root,
2107 * so that it acts as security boundary for unprivileged host code. We then use fs namespacing
2108 * to make this directory permeable for the service itself.
2110 * Specifically: for a service which wants a special directory "foo/" we first create a
2111 * directory "private/" with access mode 0700 owned by root:root. Then we place "foo" inside of
2112 * that directory (i.e. "private/foo/"), and make "foo" a symlink to "private/foo". This way,
2113 * privileged host users can access "foo/" as usual, but unprivileged host users can't look
2114 * into it. Inside of the namespaceof the container "private/" is replaced by a more liberally
2115 * accessible tmpfs, into which the host's "private/foo/" is mounted under the same name, thus
2116 * disabling the access boundary for the service and making sure it only gets access to the
2117 * dirs it needs but no others. Tricky? Yes, absolutely, but it works!
2119 * Note that we don't do this for EXEC_DIRECTORY_CONFIGURATION as that's assumed not to be
2120 * owned by the service itself.
2121 * Also, note that we don't do this for EXEC_DIRECTORY_RUNTIME as that's often used for sharing
2122 * files or sockets with other services. */
2124 private_root
= path_join(params
->prefix
[type
], "private");
2125 if (!private_root
) {
2130 /* First set up private root if it doesn't exist yet, with access mode 0700 and owned by root:root */
2131 r
= mkdir_safe_label(private_root
, 0700, 0, 0, MKDIR_WARN_MODE
);
2135 pp
= path_join(private_root
, *rt
);
2141 /* Create all directories between the configured directory and this private root, and mark them 0755 */
2142 r
= mkdir_parents_label(pp
, 0755);
2146 if (is_dir(p
, false) > 0 &&
2147 (laccess(pp
, F_OK
) < 0 && errno
== ENOENT
)) {
2149 /* Hmm, the private directory doesn't exist yet, but the normal one exists? If so, move
2150 * it over. Most likely the service has been upgraded from one that didn't use
2151 * DynamicUser=1, to one that does. */
2153 if (rename(p
, pp
) < 0) {
2158 /* Otherwise, create the actual directory for the service */
2160 r
= mkdir_label(pp
, context
->directories
[type
].mode
);
2161 if (r
< 0 && r
!= -EEXIST
)
2165 /* And link it up from the original place */
2166 r
= symlink_idempotent(pp
, p
, true);
2171 r
= mkdir_label(p
, context
->directories
[type
].mode
);
2176 if (type
== EXEC_DIRECTORY_CONFIGURATION
) {
2179 /* Don't change the owner/access mode of the configuration directory,
2180 * as in the common case it is not written to by a service, and shall
2181 * not be writable. */
2183 if (stat(p
, &st
) < 0) {
2188 /* Still complain if the access mode doesn't match */
2189 if (((st
.st_mode
^ context
->directories
[type
].mode
) & 07777) != 0)
2190 log_warning("%s \'%s\' already exists but the mode is different. "
2191 "(File system: %o %sMode: %o)",
2192 exec_directory_type_to_string(type
), *rt
,
2193 st
.st_mode
& 07777, exec_directory_type_to_string(type
), context
->directories
[type
].mode
& 07777);
2200 /* Lock down the access mode (we use chmod_and_chown() to make this idempotent. We don't
2201 * specify UID/GID here, so that path_chown_recursive() can optimize things depending on the
2202 * current UID/GID ownership.) */
2203 r
= chmod_and_chown(pp
?: p
, context
->directories
[type
].mode
, UID_INVALID
, GID_INVALID
);
2207 /* Then, change the ownership of the whole tree, if necessary. When dynamic users are used we
2208 * drop the suid/sgid bits, since we really don't want SUID/SGID files for dynamic UID/GID
2209 * assignments to exist.*/
2210 r
= path_chown_recursive(pp
?: p
, uid
, gid
, context
->dynamic_user
? 01777 : 07777);
2218 *exit_status
= exit_status_table
[type
];
2223 static int setup_smack(
2224 const ExecContext
*context
,
2225 const ExecCommand
*command
) {
2232 if (context
->smack_process_label
) {
2233 r
= mac_smack_apply_pid(0, context
->smack_process_label
);
2237 #ifdef SMACK_DEFAULT_PROCESS_LABEL
2239 _cleanup_free_
char *exec_label
= NULL
;
2241 r
= mac_smack_read(command
->path
, SMACK_ATTR_EXEC
, &exec_label
);
2242 if (r
< 0 && !IN_SET(r
, -ENODATA
, -EOPNOTSUPP
))
2245 r
= mac_smack_apply_pid(0, exec_label
? : SMACK_DEFAULT_PROCESS_LABEL
);
2255 static int compile_bind_mounts(
2256 const ExecContext
*context
,
2257 const ExecParameters
*params
,
2258 BindMount
**ret_bind_mounts
,
2259 size_t *ret_n_bind_mounts
,
2260 char ***ret_empty_directories
) {
2262 _cleanup_strv_free_
char **empty_directories
= NULL
;
2263 BindMount
*bind_mounts
;
2265 ExecDirectoryType t
;
2270 assert(ret_bind_mounts
);
2271 assert(ret_n_bind_mounts
);
2272 assert(ret_empty_directories
);
2274 n
= context
->n_bind_mounts
;
2275 for (t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
2276 if (!params
->prefix
[t
])
2279 n
+= strv_length(context
->directories
[t
].paths
);
2283 *ret_bind_mounts
= NULL
;
2284 *ret_n_bind_mounts
= 0;
2285 *ret_empty_directories
= NULL
;
2289 bind_mounts
= new(BindMount
, n
);
2293 for (i
= 0; i
< context
->n_bind_mounts
; i
++) {
2294 BindMount
*item
= context
->bind_mounts
+ i
;
2297 s
= strdup(item
->source
);
2303 d
= strdup(item
->destination
);
2310 bind_mounts
[h
++] = (BindMount
) {
2313 .read_only
= item
->read_only
,
2314 .recursive
= item
->recursive
,
2315 .ignore_enoent
= item
->ignore_enoent
,
2319 for (t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
2322 if (!params
->prefix
[t
])
2325 if (strv_isempty(context
->directories
[t
].paths
))
2328 if (context
->dynamic_user
&&
2329 !IN_SET(t
, EXEC_DIRECTORY_RUNTIME
, EXEC_DIRECTORY_CONFIGURATION
) &&
2330 !(context
->root_directory
|| context
->root_image
)) {
2333 /* So this is for a dynamic user, and we need to make sure the process can access its own
2334 * directory. For that we overmount the usually inaccessible "private" subdirectory with a
2335 * tmpfs that makes it accessible and is empty except for the submounts we do this for. */
2337 private_root
= strjoin(params
->prefix
[t
], "/private");
2338 if (!private_root
) {
2343 r
= strv_consume(&empty_directories
, private_root
);
2348 STRV_FOREACH(suffix
, context
->directories
[t
].paths
) {
2351 if (context
->dynamic_user
&&
2352 !IN_SET(t
, EXEC_DIRECTORY_RUNTIME
, EXEC_DIRECTORY_CONFIGURATION
))
2353 s
= strjoin(params
->prefix
[t
], "/private/", *suffix
);
2355 s
= strjoin(params
->prefix
[t
], "/", *suffix
);
2361 if (context
->dynamic_user
&&
2362 !IN_SET(t
, EXEC_DIRECTORY_RUNTIME
, EXEC_DIRECTORY_CONFIGURATION
) &&
2363 (context
->root_directory
|| context
->root_image
))
2364 /* When RootDirectory= or RootImage= are set, then the symbolic link to the private
2365 * directory is not created on the root directory. So, let's bind-mount the directory
2366 * on the 'non-private' place. */
2367 d
= strjoin(params
->prefix
[t
], "/", *suffix
);
2376 bind_mounts
[h
++] = (BindMount
) {
2380 .nosuid
= context
->dynamic_user
, /* don't allow suid/sgid when DynamicUser= is on */
2382 .ignore_enoent
= false,
2389 *ret_bind_mounts
= bind_mounts
;
2390 *ret_n_bind_mounts
= n
;
2391 *ret_empty_directories
= TAKE_PTR(empty_directories
);
2396 bind_mount_free_many(bind_mounts
, h
);
2400 static int apply_mount_namespace(
2402 const ExecCommand
*command
,
2403 const ExecContext
*context
,
2404 const ExecParameters
*params
,
2405 const ExecRuntime
*runtime
,
2406 char **error_path
) {
2408 _cleanup_strv_free_
char **empty_directories
= NULL
;
2409 char *tmp
= NULL
, *var
= NULL
;
2410 const char *root_dir
= NULL
, *root_image
= NULL
;
2411 NamespaceInfo ns_info
;
2412 bool needs_sandboxing
;
2413 BindMount
*bind_mounts
= NULL
;
2414 size_t n_bind_mounts
= 0;
2419 /* The runtime struct only contains the parent of the private /tmp,
2420 * which is non-accessible to world users. Inside of it there's a /tmp
2421 * that is sticky, and that's the one we want to use here. */
2423 if (context
->private_tmp
&& runtime
) {
2424 if (runtime
->tmp_dir
)
2425 tmp
= strjoina(runtime
->tmp_dir
, "/tmp");
2426 if (runtime
->var_tmp_dir
)
2427 var
= strjoina(runtime
->var_tmp_dir
, "/tmp");
2430 if (params
->flags
& EXEC_APPLY_CHROOT
) {
2431 root_image
= context
->root_image
;
2434 root_dir
= context
->root_directory
;
2437 r
= compile_bind_mounts(context
, params
, &bind_mounts
, &n_bind_mounts
, &empty_directories
);
2441 needs_sandboxing
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && !(command
->flags
& EXEC_COMMAND_FULLY_PRIVILEGED
);
2442 if (needs_sandboxing
)
2443 ns_info
= (NamespaceInfo
) {
2444 .ignore_protect_paths
= false,
2445 .private_dev
= context
->private_devices
,
2446 .protect_control_groups
= context
->protect_control_groups
,
2447 .protect_kernel_tunables
= context
->protect_kernel_tunables
,
2448 .protect_kernel_modules
= context
->protect_kernel_modules
,
2449 .protect_hostname
= context
->protect_hostname
,
2450 .mount_apivfs
= context
->mount_apivfs
,
2451 .private_mounts
= context
->private_mounts
,
2453 else if (!context
->dynamic_user
&& root_dir
)
2455 * If DynamicUser=no and RootDirectory= is set then lets pass a relaxed
2456 * sandbox info, otherwise enforce it, don't ignore protected paths and
2457 * fail if we are enable to apply the sandbox inside the mount namespace.
2459 ns_info
= (NamespaceInfo
) {
2460 .ignore_protect_paths
= true,
2463 ns_info
= (NamespaceInfo
) {};
2465 if (context
->mount_flags
== MS_SHARED
)
2466 log_unit_debug(u
, "shared mount propagation hidden by other fs namespacing unit settings: ignoring");
2468 r
= setup_namespace(root_dir
, root_image
,
2469 &ns_info
, context
->read_write_paths
,
2470 needs_sandboxing
? context
->read_only_paths
: NULL
,
2471 needs_sandboxing
? context
->inaccessible_paths
: NULL
,
2475 context
->temporary_filesystems
,
2476 context
->n_temporary_filesystems
,
2479 needs_sandboxing
? context
->protect_home
: PROTECT_HOME_NO
,
2480 needs_sandboxing
? context
->protect_system
: PROTECT_SYSTEM_NO
,
2481 context
->mount_flags
,
2482 DISSECT_IMAGE_DISCARD_ON_LOOP
,
2485 bind_mount_free_many(bind_mounts
, n_bind_mounts
);
2487 /* If we couldn't set up the namespace this is probably due to a missing capability. setup_namespace() reports
2488 * that with a special, recognizable error ENOANO. In this case, silently proceed, but only if exclusively
2489 * sandboxing options were used, i.e. nothing such as RootDirectory= or BindMount= that would result in a
2490 * completely different execution environment. */
2492 if (n_bind_mounts
== 0 &&
2493 context
->n_temporary_filesystems
== 0 &&
2494 !root_dir
&& !root_image
&&
2495 !context
->dynamic_user
) {
2496 log_unit_debug(u
, "Failed to set up namespace, assuming containerized execution and ignoring.");
2500 log_unit_debug(u
, "Failed to set up namespace, and refusing to continue since the selected namespacing options alter mount environment non-trivially.\n"
2501 "Bind mounts: %zu, temporary filesystems: %zu, root directory: %s, root image: %s, dynamic user: %s",
2502 n_bind_mounts
, context
->n_temporary_filesystems
, yes_no(root_dir
), yes_no(root_image
), yes_no(context
->dynamic_user
));
2510 static int apply_working_directory(
2511 const ExecContext
*context
,
2512 const ExecParameters
*params
,
2514 const bool needs_mount_ns
,
2520 assert(exit_status
);
2522 if (context
->working_directory_home
) {
2525 *exit_status
= EXIT_CHDIR
;
2531 } else if (context
->working_directory
)
2532 wd
= context
->working_directory
;
2536 if (params
->flags
& EXEC_APPLY_CHROOT
) {
2537 if (!needs_mount_ns
&& context
->root_directory
)
2538 if (chroot(context
->root_directory
) < 0) {
2539 *exit_status
= EXIT_CHROOT
;
2545 d
= prefix_roota(context
->root_directory
, wd
);
2547 if (chdir(d
) < 0 && !context
->working_directory_missing_ok
) {
2548 *exit_status
= EXIT_CHDIR
;
2555 static int setup_keyring(
2557 const ExecContext
*context
,
2558 const ExecParameters
*p
,
2559 uid_t uid
, gid_t gid
) {
2561 key_serial_t keyring
;
2570 /* Let's set up a new per-service "session" kernel keyring for each system service. This has the benefit that
2571 * each service runs with its own keyring shared among all processes of the service, but with no hook-up beyond
2572 * that scope, and in particular no link to the per-UID keyring. If we don't do this the keyring will be
2573 * automatically created on-demand and then linked to the per-UID keyring, by the kernel. The kernel's built-in
2574 * on-demand behaviour is very appropriate for login users, but probably not so much for system services, where
2575 * UIDs are not necessarily specific to a service but reused (at least in the case of UID 0). */
2577 if (context
->keyring_mode
== EXEC_KEYRING_INHERIT
)
2580 /* Acquiring a reference to the user keyring is nasty. We briefly change identity in order to get things set up
2581 * properly by the kernel. If we don't do that then we can't create it atomically, and that sucks for parallel
2582 * execution. This mimics what pam_keyinit does, too. Setting up session keyring, to be owned by the right user
2583 * & group is just as nasty as acquiring a reference to the user keyring. */
2585 saved_uid
= getuid();
2586 saved_gid
= getgid();
2588 if (gid_is_valid(gid
) && gid
!= saved_gid
) {
2589 if (setregid(gid
, -1) < 0)
2590 return log_unit_error_errno(u
, errno
, "Failed to change GID for user keyring: %m");
2593 if (uid_is_valid(uid
) && uid
!= saved_uid
) {
2594 if (setreuid(uid
, -1) < 0) {
2595 r
= log_unit_error_errno(u
, errno
, "Failed to change UID for user keyring: %m");
2600 keyring
= keyctl(KEYCTL_JOIN_SESSION_KEYRING
, 0, 0, 0, 0);
2601 if (keyring
== -1) {
2602 if (errno
== ENOSYS
)
2603 log_unit_debug_errno(u
, errno
, "Kernel keyring not supported, ignoring.");
2604 else if (IN_SET(errno
, EACCES
, EPERM
))
2605 log_unit_debug_errno(u
, errno
, "Kernel keyring access prohibited, ignoring.");
2606 else if (errno
== EDQUOT
)
2607 log_unit_debug_errno(u
, errno
, "Out of kernel keyrings to allocate, ignoring.");
2609 r
= log_unit_error_errno(u
, errno
, "Setting up kernel keyring failed: %m");
2614 /* When requested link the user keyring into the session keyring. */
2615 if (context
->keyring_mode
== EXEC_KEYRING_SHARED
) {
2617 if (keyctl(KEYCTL_LINK
,
2618 KEY_SPEC_USER_KEYRING
,
2619 KEY_SPEC_SESSION_KEYRING
, 0, 0) < 0) {
2620 r
= log_unit_error_errno(u
, errno
, "Failed to link user keyring into session keyring: %m");
2625 /* Restore uid/gid back */
2626 if (uid_is_valid(uid
) && uid
!= saved_uid
) {
2627 if (setreuid(saved_uid
, -1) < 0) {
2628 r
= log_unit_error_errno(u
, errno
, "Failed to change UID back for user keyring: %m");
2633 if (gid_is_valid(gid
) && gid
!= saved_gid
) {
2634 if (setregid(saved_gid
, -1) < 0)
2635 return log_unit_error_errno(u
, errno
, "Failed to change GID back for user keyring: %m");
2638 /* Populate they keyring with the invocation ID by default, as original saved_uid. */
2639 if (!sd_id128_is_null(u
->invocation_id
)) {
2642 key
= add_key("user", "invocation_id", &u
->invocation_id
, sizeof(u
->invocation_id
), KEY_SPEC_SESSION_KEYRING
);
2644 log_unit_debug_errno(u
, errno
, "Failed to add invocation ID to keyring, ignoring: %m");
2646 if (keyctl(KEYCTL_SETPERM
, key
,
2647 KEY_POS_VIEW
|KEY_POS_READ
|KEY_POS_SEARCH
|
2648 KEY_USR_VIEW
|KEY_USR_READ
|KEY_USR_SEARCH
, 0, 0) < 0)
2649 r
= log_unit_error_errno(u
, errno
, "Failed to restrict invocation ID permission: %m");
2654 /* Revert back uid & gid for the the last time, and exit */
2655 /* no extra logging, as only the first already reported error matters */
2656 if (getuid() != saved_uid
)
2657 (void) setreuid(saved_uid
, -1);
2659 if (getgid() != saved_gid
)
2660 (void) setregid(saved_gid
, -1);
2665 static void append_socket_pair(int *array
, size_t *n
, const int pair
[static 2]) {
2673 array
[(*n
)++] = pair
[0];
2675 array
[(*n
)++] = pair
[1];
2678 static int close_remaining_fds(
2679 const ExecParameters
*params
,
2680 const ExecRuntime
*runtime
,
2681 const DynamicCreds
*dcreds
,
2685 int *fds
, size_t n_fds
) {
2687 size_t n_dont_close
= 0;
2688 int dont_close
[n_fds
+ 12];
2692 if (params
->stdin_fd
>= 0)
2693 dont_close
[n_dont_close
++] = params
->stdin_fd
;
2694 if (params
->stdout_fd
>= 0)
2695 dont_close
[n_dont_close
++] = params
->stdout_fd
;
2696 if (params
->stderr_fd
>= 0)
2697 dont_close
[n_dont_close
++] = params
->stderr_fd
;
2700 dont_close
[n_dont_close
++] = socket_fd
;
2702 dont_close
[n_dont_close
++] = exec_fd
;
2704 memcpy(dont_close
+ n_dont_close
, fds
, sizeof(int) * n_fds
);
2705 n_dont_close
+= n_fds
;
2709 append_socket_pair(dont_close
, &n_dont_close
, runtime
->netns_storage_socket
);
2713 append_socket_pair(dont_close
, &n_dont_close
, dcreds
->user
->storage_socket
);
2715 append_socket_pair(dont_close
, &n_dont_close
, dcreds
->group
->storage_socket
);
2718 if (user_lookup_fd
>= 0)
2719 dont_close
[n_dont_close
++] = user_lookup_fd
;
2721 return close_all_fds(dont_close
, n_dont_close
);
2724 static int send_user_lookup(
2732 /* Send the resolved UID/GID to PID 1 after we learnt it. We send a single datagram, containing the UID/GID
2733 * data as well as the unit name. Note that we suppress sending this if no user/group to resolve was
2736 if (user_lookup_fd
< 0)
2739 if (!uid_is_valid(uid
) && !gid_is_valid(gid
))
2742 if (writev(user_lookup_fd
,
2744 IOVEC_INIT(&uid
, sizeof(uid
)),
2745 IOVEC_INIT(&gid
, sizeof(gid
)),
2746 IOVEC_INIT_STRING(unit
->id
) }, 3) < 0)
2752 static int acquire_home(const ExecContext
*c
, uid_t uid
, const char** home
, char **buf
) {
2759 /* If WorkingDirectory=~ is set, try to acquire a usable home directory. */
2764 if (!c
->working_directory_home
)
2767 r
= get_home_dir(buf
);
2775 static int compile_suggested_paths(const ExecContext
*c
, const ExecParameters
*p
, char ***ret
) {
2776 _cleanup_strv_free_
char ** list
= NULL
;
2777 ExecDirectoryType t
;
2784 assert(c
->dynamic_user
);
2786 /* Compile a list of paths that it might make sense to read the owning UID from to use as initial candidate for
2787 * dynamic UID allocation, in order to save us from doing costly recursive chown()s of the special
2790 for (t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
2793 if (t
== EXEC_DIRECTORY_CONFIGURATION
)
2799 STRV_FOREACH(i
, c
->directories
[t
].paths
) {
2802 if (t
== EXEC_DIRECTORY_RUNTIME
)
2803 e
= strjoin(p
->prefix
[t
], "/", *i
);
2805 e
= strjoin(p
->prefix
[t
], "/private/", *i
);
2809 r
= strv_consume(&list
, e
);
2815 *ret
= TAKE_PTR(list
);
2820 static char *exec_command_line(char **argv
);
2822 static int exec_parameters_get_cgroup_path(const ExecParameters
*params
, char **ret
) {
2823 bool using_subcgroup
;
2829 if (!params
->cgroup_path
)
2832 /* If we are called for a unit where cgroup delegation is on, and the payload created its own populated
2833 * subcgroup (which we expect it to do, after all it asked for delegation), then we cannot place the control
2834 * processes started after the main unit's process in the unit's main cgroup because it is now an inner one,
2835 * and inner cgroups may not contain processes. Hence, if delegation is on, and this is a control process,
2836 * let's use ".control" as subcgroup instead. Note that we do so only for ExecStartPost=, ExecReload=,
2837 * ExecStop=, ExecStopPost=, i.e. for the commands where the main process is already forked. For ExecStartPre=
2838 * this is not necessary, the cgroup is still empty. We distinguish these cases with the EXEC_CONTROL_CGROUP
2839 * flag, which is only passed for the former statements, not for the latter. */
2841 using_subcgroup
= FLAGS_SET(params
->flags
, EXEC_CONTROL_CGROUP
|EXEC_CGROUP_DELEGATE
|EXEC_IS_CONTROL
);
2842 if (using_subcgroup
)
2843 p
= strjoin(params
->cgroup_path
, "/.control");
2845 p
= strdup(params
->cgroup_path
);
2850 return using_subcgroup
;
2853 static int exec_child(
2855 const ExecCommand
*command
,
2856 const ExecContext
*context
,
2857 const ExecParameters
*params
,
2858 ExecRuntime
*runtime
,
2859 DynamicCreds
*dcreds
,
2863 size_t n_socket_fds
,
2864 size_t n_storage_fds
,
2869 _cleanup_strv_free_
char **our_env
= NULL
, **pass_env
= NULL
, **accum_env
= NULL
, **replaced_argv
= NULL
;
2870 int *fds_with_exec_fd
, n_fds_with_exec_fd
, r
, ngids
= 0, exec_fd
= -1;
2871 _cleanup_free_ gid_t
*supplementary_gids
= NULL
;
2872 const char *username
= NULL
, *groupname
= NULL
;
2873 _cleanup_free_
char *home_buffer
= NULL
;
2874 const char *home
= NULL
, *shell
= NULL
;
2875 char **final_argv
= NULL
;
2876 dev_t journal_stream_dev
= 0;
2877 ino_t journal_stream_ino
= 0;
2878 bool needs_sandboxing
, /* Do we need to set up full sandboxing? (i.e. all namespacing, all MAC stuff, caps, yadda yadda */
2879 needs_setuid
, /* Do we need to do the actual setresuid()/setresgid() calls? */
2880 needs_mount_namespace
, /* Do we need to set up a mount namespace for this kernel? */
2881 needs_ambient_hack
; /* Do we need to apply the ambient capabilities hack? */
2883 _cleanup_free_
char *mac_selinux_context_net
= NULL
;
2884 bool use_selinux
= false;
2887 bool use_smack
= false;
2890 bool use_apparmor
= false;
2892 uid_t uid
= UID_INVALID
;
2893 gid_t gid
= GID_INVALID
;
2895 ExecDirectoryType dt
;
2902 assert(exit_status
);
2904 rename_process_from_path(command
->path
);
2906 /* We reset exactly these signals, since they are the
2907 * only ones we set to SIG_IGN in the main daemon. All
2908 * others we leave untouched because we set them to
2909 * SIG_DFL or a valid handler initially, both of which
2910 * will be demoted to SIG_DFL. */
2911 (void) default_signals(SIGNALS_CRASH_HANDLER
,
2912 SIGNALS_IGNORE
, -1);
2914 if (context
->ignore_sigpipe
)
2915 (void) ignore_signals(SIGPIPE
, -1);
2917 r
= reset_signal_mask();
2919 *exit_status
= EXIT_SIGNAL_MASK
;
2920 return log_unit_error_errno(unit
, r
, "Failed to set process signal mask: %m");
2923 if (params
->idle_pipe
)
2924 do_idle_pipe_dance(params
->idle_pipe
);
2926 /* Close fds we don't need very early to make sure we don't block init reexecution because it cannot bind its
2927 * sockets. Among the fds we close are the logging fds, and we want to keep them closed, so that we don't have
2928 * any fds open we don't really want open during the transition. In order to make logging work, we switch the
2929 * log subsystem into open_when_needed mode, so that it reopens the logs on every single log call. */
2932 log_set_open_when_needed(true);
2934 /* In case anything used libc syslog(), close this here, too */
2937 n_fds
= n_socket_fds
+ n_storage_fds
;
2938 r
= close_remaining_fds(params
, runtime
, dcreds
, user_lookup_fd
, socket_fd
, params
->exec_fd
, fds
, n_fds
);
2940 *exit_status
= EXIT_FDS
;
2941 return log_unit_error_errno(unit
, r
, "Failed to close unwanted file descriptors: %m");
2944 if (!context
->same_pgrp
)
2946 *exit_status
= EXIT_SETSID
;
2947 return log_unit_error_errno(unit
, errno
, "Failed to create new process session: %m");
2950 exec_context_tty_reset(context
, params
);
2952 if (unit_shall_confirm_spawn(unit
)) {
2953 const char *vc
= params
->confirm_spawn
;
2954 _cleanup_free_
char *cmdline
= NULL
;
2956 cmdline
= exec_command_line(command
->argv
);
2958 *exit_status
= EXIT_MEMORY
;
2962 r
= ask_for_confirmation(vc
, unit
, cmdline
);
2963 if (r
!= CONFIRM_EXECUTE
) {
2964 if (r
== CONFIRM_PRETEND_SUCCESS
) {
2965 *exit_status
= EXIT_SUCCESS
;
2968 *exit_status
= EXIT_CONFIRM
;
2969 log_unit_error(unit
, "Execution cancelled by the user");
2974 /* We are about to invoke NSS and PAM modules. Let's tell them what we are doing here, maybe they care. This is
2975 * used by nss-resolve to disable itself when we are about to start systemd-resolved, to avoid deadlocks. Note
2976 * that these env vars do not survive the execve(), which means they really only apply to the PAM and NSS
2977 * invocations themselves. Also note that while we'll only invoke NSS modules involved in user management they
2978 * might internally call into other NSS modules that are involved in hostname resolution, we never know. */
2979 if (setenv("SYSTEMD_ACTIVATION_UNIT", unit
->id
, true) != 0 ||
2980 setenv("SYSTEMD_ACTIVATION_SCOPE", MANAGER_IS_SYSTEM(unit
->manager
) ? "system" : "user", true) != 0) {
2981 *exit_status
= EXIT_MEMORY
;
2982 return log_unit_error_errno(unit
, errno
, "Failed to update environment: %m");
2985 if (context
->dynamic_user
&& dcreds
) {
2986 _cleanup_strv_free_
char **suggested_paths
= NULL
;
2988 /* On top of that, make sure we bypass our own NSS module nss-systemd comprehensively for any NSS
2989 * checks, if DynamicUser=1 is used, as we shouldn't create a feedback loop with ourselves here.*/
2990 if (putenv((char*) "SYSTEMD_NSS_DYNAMIC_BYPASS=1") != 0) {
2991 *exit_status
= EXIT_USER
;
2992 return log_unit_error_errno(unit
, errno
, "Failed to update environment: %m");
2995 r
= compile_suggested_paths(context
, params
, &suggested_paths
);
2997 *exit_status
= EXIT_MEMORY
;
3001 r
= dynamic_creds_realize(dcreds
, suggested_paths
, &uid
, &gid
);
3003 *exit_status
= EXIT_USER
;
3005 log_unit_error(unit
, "Failed to update dynamic user credentials: User or group with specified name already exists.");
3008 return log_unit_error_errno(unit
, r
, "Failed to update dynamic user credentials: %m");
3011 if (!uid_is_valid(uid
)) {
3012 *exit_status
= EXIT_USER
;
3013 log_unit_error(unit
, "UID validation failed for \""UID_FMT
"\"", uid
);
3017 if (!gid_is_valid(gid
)) {
3018 *exit_status
= EXIT_USER
;
3019 log_unit_error(unit
, "GID validation failed for \""GID_FMT
"\"", gid
);
3024 username
= dcreds
->user
->name
;
3027 r
= get_fixed_user(context
, &username
, &uid
, &gid
, &home
, &shell
);
3029 *exit_status
= EXIT_USER
;
3030 return log_unit_error_errno(unit
, r
, "Failed to determine user credentials: %m");
3033 r
= get_fixed_group(context
, &groupname
, &gid
);
3035 *exit_status
= EXIT_GROUP
;
3036 return log_unit_error_errno(unit
, r
, "Failed to determine group credentials: %m");
3040 /* Initialize user supplementary groups and get SupplementaryGroups= ones */
3041 r
= get_supplementary_groups(context
, username
, groupname
, gid
,
3042 &supplementary_gids
, &ngids
);
3044 *exit_status
= EXIT_GROUP
;
3045 return log_unit_error_errno(unit
, r
, "Failed to determine supplementary groups: %m");
3048 r
= send_user_lookup(unit
, user_lookup_fd
, uid
, gid
);
3050 *exit_status
= EXIT_USER
;
3051 return log_unit_error_errno(unit
, r
, "Failed to send user credentials to PID1: %m");
3054 user_lookup_fd
= safe_close(user_lookup_fd
);
3056 r
= acquire_home(context
, uid
, &home
, &home_buffer
);
3058 *exit_status
= EXIT_CHDIR
;
3059 return log_unit_error_errno(unit
, r
, "Failed to determine $HOME for user: %m");
3062 /* If a socket is connected to STDIN/STDOUT/STDERR, we
3063 * must sure to drop O_NONBLOCK */
3065 (void) fd_nonblock(socket_fd
, false);
3067 /* Journald will try to look-up our cgroup in order to populate _SYSTEMD_CGROUP and _SYSTEMD_UNIT fields.
3068 * Hence we need to migrate to the target cgroup from init.scope before connecting to journald */
3069 if (params
->cgroup_path
) {
3070 _cleanup_free_
char *p
= NULL
;
3072 r
= exec_parameters_get_cgroup_path(params
, &p
);
3074 *exit_status
= EXIT_CGROUP
;
3075 return log_unit_error_errno(unit
, r
, "Failed to acquire cgroup path: %m");
3078 r
= cg_attach_everywhere(params
->cgroup_supported
, p
, 0, NULL
, NULL
);
3080 *exit_status
= EXIT_CGROUP
;
3081 return log_unit_error_errno(unit
, r
, "Failed to attach to cgroup %s: %m", p
);
3085 if (context
->network_namespace_path
&& runtime
&& runtime
->netns_storage_socket
[0] >= 0) {
3086 r
= open_netns_path(runtime
->netns_storage_socket
, context
->network_namespace_path
);
3088 *exit_status
= EXIT_NETWORK
;
3089 return log_unit_error_errno(unit
, r
, "Failed to open network namespace path %s: %m", context
->network_namespace_path
);
3093 r
= setup_input(context
, params
, socket_fd
, named_iofds
);
3095 *exit_status
= EXIT_STDIN
;
3096 return log_unit_error_errno(unit
, r
, "Failed to set up standard input: %m");
3099 r
= setup_output(unit
, context
, params
, STDOUT_FILENO
, socket_fd
, named_iofds
, basename(command
->path
), uid
, gid
, &journal_stream_dev
, &journal_stream_ino
);
3101 *exit_status
= EXIT_STDOUT
;
3102 return log_unit_error_errno(unit
, r
, "Failed to set up standard output: %m");
3105 r
= setup_output(unit
, context
, params
, STDERR_FILENO
, socket_fd
, named_iofds
, basename(command
->path
), uid
, gid
, &journal_stream_dev
, &journal_stream_ino
);
3107 *exit_status
= EXIT_STDERR
;
3108 return log_unit_error_errno(unit
, r
, "Failed to set up standard error output: %m");
3111 if (context
->oom_score_adjust_set
) {
3112 /* When we can't make this change due to EPERM, then let's silently skip over it. User namespaces
3113 * prohibit write access to this file, and we shouldn't trip up over that. */
3114 r
= set_oom_score_adjust(context
->oom_score_adjust
);
3115 if (IN_SET(r
, -EPERM
, -EACCES
))
3116 log_unit_debug_errno(unit
, r
, "Failed to adjust OOM setting, assuming containerized execution, ignoring: %m");
3118 *exit_status
= EXIT_OOM_ADJUST
;
3119 return log_unit_error_errno(unit
, r
, "Failed to adjust OOM setting: %m");
3123 if (context
->nice_set
)
3124 if (setpriority(PRIO_PROCESS
, 0, context
->nice
) < 0) {
3125 *exit_status
= EXIT_NICE
;
3126 return log_unit_error_errno(unit
, errno
, "Failed to set up process scheduling priority (nice level): %m");
3129 if (context
->cpu_sched_set
) {
3130 struct sched_param param
= {
3131 .sched_priority
= context
->cpu_sched_priority
,
3134 r
= sched_setscheduler(0,
3135 context
->cpu_sched_policy
|
3136 (context
->cpu_sched_reset_on_fork
?
3137 SCHED_RESET_ON_FORK
: 0),
3140 *exit_status
= EXIT_SETSCHEDULER
;
3141 return log_unit_error_errno(unit
, errno
, "Failed to set up CPU scheduling: %m");
3145 if (context
->cpu_set
.set
)
3146 if (sched_setaffinity(0, context
->cpu_set
.allocated
, context
->cpu_set
.set
) < 0) {
3147 *exit_status
= EXIT_CPUAFFINITY
;
3148 return log_unit_error_errno(unit
, errno
, "Failed to set up CPU affinity: %m");
3151 if (context
->ioprio_set
)
3152 if (ioprio_set(IOPRIO_WHO_PROCESS
, 0, context
->ioprio
) < 0) {
3153 *exit_status
= EXIT_IOPRIO
;
3154 return log_unit_error_errno(unit
, errno
, "Failed to set up IO scheduling priority: %m");
3157 if (context
->timer_slack_nsec
!= NSEC_INFINITY
)
3158 if (prctl(PR_SET_TIMERSLACK
, context
->timer_slack_nsec
) < 0) {
3159 *exit_status
= EXIT_TIMERSLACK
;
3160 return log_unit_error_errno(unit
, errno
, "Failed to set up timer slack: %m");
3163 if (context
->personality
!= PERSONALITY_INVALID
) {
3164 r
= safe_personality(context
->personality
);
3166 *exit_status
= EXIT_PERSONALITY
;
3167 return log_unit_error_errno(unit
, r
, "Failed to set up execution domain (personality): %m");
3171 if (context
->utmp_id
)
3172 utmp_put_init_process(context
->utmp_id
, getpid_cached(), getsid(0),
3174 context
->utmp_mode
== EXEC_UTMP_INIT
? INIT_PROCESS
:
3175 context
->utmp_mode
== EXEC_UTMP_LOGIN
? LOGIN_PROCESS
:
3179 if (uid_is_valid(uid
)) {
3180 r
= chown_terminal(STDIN_FILENO
, uid
);
3182 *exit_status
= EXIT_STDIN
;
3183 return log_unit_error_errno(unit
, r
, "Failed to change ownership of terminal: %m");
3187 /* If delegation is enabled we'll pass ownership of the cgroup to the user of the new process. On cgroup v1
3188 * this is only about systemd's own hierarchy, i.e. not the controller hierarchies, simply because that's not
3189 * safe. On cgroup v2 there's only one hierarchy anyway, and delegation is safe there, hence in that case only
3190 * touch a single hierarchy too. */
3191 if (params
->cgroup_path
&& context
->user
&& (params
->flags
& EXEC_CGROUP_DELEGATE
)) {
3192 r
= cg_set_access(SYSTEMD_CGROUP_CONTROLLER
, params
->cgroup_path
, uid
, gid
);
3194 *exit_status
= EXIT_CGROUP
;
3195 return log_unit_error_errno(unit
, r
, "Failed to adjust control group access: %m");
3199 for (dt
= 0; dt
< _EXEC_DIRECTORY_TYPE_MAX
; dt
++) {
3200 r
= setup_exec_directory(context
, params
, uid
, gid
, dt
, exit_status
);
3202 return log_unit_error_errno(unit
, r
, "Failed to set up special execution directory in %s: %m", params
->prefix
[dt
]);
3205 r
= build_environment(
3217 *exit_status
= EXIT_MEMORY
;
3221 r
= build_pass_environment(context
, &pass_env
);
3223 *exit_status
= EXIT_MEMORY
;
3227 accum_env
= strv_env_merge(5,
3228 params
->environment
,
3231 context
->environment
,
3235 *exit_status
= EXIT_MEMORY
;
3238 accum_env
= strv_env_clean(accum_env
);
3240 (void) umask(context
->umask
);
3242 r
= setup_keyring(unit
, context
, params
, uid
, gid
);
3244 *exit_status
= EXIT_KEYRING
;
3245 return log_unit_error_errno(unit
, r
, "Failed to set up kernel keyring: %m");
3248 /* We need sandboxing if the caller asked us to apply it and the command isn't explicitly excepted from it */
3249 needs_sandboxing
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && !(command
->flags
& EXEC_COMMAND_FULLY_PRIVILEGED
);
3251 /* 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 */
3252 needs_ambient_hack
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && (command
->flags
& EXEC_COMMAND_AMBIENT_MAGIC
) && !ambient_capabilities_supported();
3254 /* 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 */
3255 if (needs_ambient_hack
)
3256 needs_setuid
= false;
3258 needs_setuid
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && !(command
->flags
& (EXEC_COMMAND_FULLY_PRIVILEGED
|EXEC_COMMAND_NO_SETUID
));
3260 if (needs_sandboxing
) {
3261 /* MAC enablement checks need to be done before a new mount ns is created, as they rely on /sys being
3262 * present. The actual MAC context application will happen later, as late as possible, to avoid
3263 * impacting our own code paths. */
3266 use_selinux
= mac_selinux_use();
3269 use_smack
= mac_smack_use();
3272 use_apparmor
= mac_apparmor_use();
3276 if (needs_sandboxing
) {
3279 /* Let's set the resource limits before we call into PAM, so that pam_limits wins over what
3280 * is set here. (See below.) */
3282 r
= setrlimit_closest_all((const struct rlimit
* const *) context
->rlimit
, &which_failed
);
3284 *exit_status
= EXIT_LIMITS
;
3285 return log_unit_error_errno(unit
, r
, "Failed to adjust resource limit RLIMIT_%s: %m", rlimit_to_string(which_failed
));
3291 /* Let's call into PAM after we set up our own idea of resource limits to that pam_limits
3292 * wins here. (See above.) */
3294 if (context
->pam_name
&& username
) {
3295 r
= setup_pam(context
->pam_name
, username
, uid
, gid
, context
->tty_path
, &accum_env
, fds
, n_fds
);
3297 *exit_status
= EXIT_PAM
;
3298 return log_unit_error_errno(unit
, r
, "Failed to set up PAM session: %m");
3303 if ((context
->private_network
|| context
->network_namespace_path
) && runtime
&& runtime
->netns_storage_socket
[0] >= 0) {
3305 if (ns_type_supported(NAMESPACE_NET
)) {
3306 r
= setup_netns(runtime
->netns_storage_socket
);
3308 *exit_status
= EXIT_NETWORK
;
3309 return log_unit_error_errno(unit
, r
, "Failed to set up network namespacing: %m");
3311 } else if (context
->network_namespace_path
) {
3312 *exit_status
= EXIT_NETWORK
;
3313 return log_unit_error_errno(unit
, SYNTHETIC_ERRNO(EOPNOTSUPP
), "NetworkNamespacePath= is not supported, refusing.");
3315 log_unit_warning(unit
, "PrivateNetwork=yes is configured, but the kernel does not support network namespaces, ignoring.");
3318 needs_mount_namespace
= exec_needs_mount_namespace(context
, params
, runtime
);
3319 if (needs_mount_namespace
) {
3320 _cleanup_free_
char *error_path
= NULL
;
3322 r
= apply_mount_namespace(unit
, command
, context
, params
, runtime
, &error_path
);
3324 *exit_status
= EXIT_NAMESPACE
;
3325 return log_unit_error_errno(unit
, r
, "Failed to set up mount namespacing%s%s: %m",
3326 error_path
? ": " : "", strempty(error_path
));
3330 if (context
->protect_hostname
) {
3331 if (ns_type_supported(NAMESPACE_UTS
)) {
3332 if (unshare(CLONE_NEWUTS
) < 0) {
3333 *exit_status
= EXIT_NAMESPACE
;
3334 return log_unit_error_errno(unit
, errno
, "Failed to set up UTS namespacing: %m");
3337 log_unit_warning(unit
, "ProtectHostname=yes is configured, but the kernel does not support UTS namespaces, ignoring namespace setup.");
3339 r
= seccomp_protect_hostname();
3341 *exit_status
= EXIT_SECCOMP
;
3342 return log_unit_error_errno(unit
, r
, "Failed to apply hostname restrictions: %m");
3347 /* Drop groups as early as possbile */
3349 r
= enforce_groups(gid
, supplementary_gids
, ngids
);
3351 *exit_status
= EXIT_GROUP
;
3352 return log_unit_error_errno(unit
, r
, "Changing group credentials failed: %m");
3356 if (needs_sandboxing
) {
3358 if (use_selinux
&& params
->selinux_context_net
&& socket_fd
>= 0) {
3359 r
= mac_selinux_get_child_mls_label(socket_fd
, command
->path
, context
->selinux_context
, &mac_selinux_context_net
);
3361 *exit_status
= EXIT_SELINUX_CONTEXT
;
3362 return log_unit_error_errno(unit
, r
, "Failed to determine SELinux context: %m");
3367 if (context
->private_users
) {
3368 r
= setup_private_users(uid
, gid
);
3370 *exit_status
= EXIT_USER
;
3371 return log_unit_error_errno(unit
, r
, "Failed to set up user namespacing: %m");
3376 /* We repeat the fd closing here, to make sure that nothing is leaked from the PAM modules. Note that we are
3377 * more aggressive this time since socket_fd and the netns fds we don't need anymore. We do keep the exec_fd
3378 * however if we have it as we want to keep it open until the final execve(). */
3380 if (params
->exec_fd
>= 0) {
3381 exec_fd
= params
->exec_fd
;
3383 if (exec_fd
< 3 + (int) n_fds
) {
3386 /* Let's move the exec fd far up, so that it's outside of the fd range we want to pass to the
3387 * process we are about to execute. */
3389 moved_fd
= fcntl(exec_fd
, F_DUPFD_CLOEXEC
, 3 + (int) n_fds
);
3391 *exit_status
= EXIT_FDS
;
3392 return log_unit_error_errno(unit
, errno
, "Couldn't move exec fd up: %m");
3395 safe_close(exec_fd
);
3398 /* This fd should be FD_CLOEXEC already, but let's make sure. */
3399 r
= fd_cloexec(exec_fd
, true);
3401 *exit_status
= EXIT_FDS
;
3402 return log_unit_error_errno(unit
, r
, "Failed to make exec fd FD_CLOEXEC: %m");
3406 fds_with_exec_fd
= newa(int, n_fds
+ 1);
3407 memcpy_safe(fds_with_exec_fd
, fds
, n_fds
* sizeof(int));
3408 fds_with_exec_fd
[n_fds
] = exec_fd
;
3409 n_fds_with_exec_fd
= n_fds
+ 1;
3411 fds_with_exec_fd
= fds
;
3412 n_fds_with_exec_fd
= n_fds
;
3415 r
= close_all_fds(fds_with_exec_fd
, n_fds_with_exec_fd
);
3417 r
= shift_fds(fds
, n_fds
);
3419 r
= flags_fds(fds
, n_socket_fds
, n_storage_fds
, context
->non_blocking
);
3421 *exit_status
= EXIT_FDS
;
3422 return log_unit_error_errno(unit
, r
, "Failed to adjust passed file descriptors: %m");
3425 /* At this point, the fds we want to pass to the program are all ready and set up, with O_CLOEXEC turned off
3426 * and at the right fd numbers. The are no other fds open, with one exception: the exec_fd if it is defined,
3427 * and it has O_CLOEXEC set, after all we want it to be closed by the execve(), so that our parent knows we
3430 secure_bits
= context
->secure_bits
;
3432 if (needs_sandboxing
) {
3435 /* Set the RTPRIO resource limit to 0, but only if nothing else was explicitly
3436 * requested. (Note this is placed after the general resource limit initialization, see
3437 * above, in order to take precedence.) */
3438 if (context
->restrict_realtime
&& !context
->rlimit
[RLIMIT_RTPRIO
]) {
3439 if (setrlimit(RLIMIT_RTPRIO
, &RLIMIT_MAKE_CONST(0)) < 0) {
3440 *exit_status
= EXIT_LIMITS
;
3441 return log_unit_error_errno(unit
, errno
, "Failed to adjust RLIMIT_RTPRIO resource limit: %m");
3446 /* LSM Smack needs the capability CAP_MAC_ADMIN to change the current execution security context of the
3447 * process. This is the latest place before dropping capabilities. Other MAC context are set later. */
3449 r
= setup_smack(context
, command
);
3451 *exit_status
= EXIT_SMACK_PROCESS_LABEL
;
3452 return log_unit_error_errno(unit
, r
, "Failed to set SMACK process label: %m");
3457 bset
= context
->capability_bounding_set
;
3458 /* If the ambient caps hack is enabled (which means the kernel can't do them, and the user asked for
3459 * our magic fallback), then let's add some extra caps, so that the service can drop privs of its own,
3460 * instead of us doing that */
3461 if (needs_ambient_hack
)
3462 bset
|= (UINT64_C(1) << CAP_SETPCAP
) |
3463 (UINT64_C(1) << CAP_SETUID
) |
3464 (UINT64_C(1) << CAP_SETGID
);
3466 if (!cap_test_all(bset
)) {
3467 r
= capability_bounding_set_drop(bset
, false);
3469 *exit_status
= EXIT_CAPABILITIES
;
3470 return log_unit_error_errno(unit
, r
, "Failed to drop capabilities: %m");
3474 /* This is done before enforce_user, but ambient set
3475 * does not survive over setresuid() if keep_caps is not set. */
3476 if (!needs_ambient_hack
&&
3477 context
->capability_ambient_set
!= 0) {
3478 r
= capability_ambient_set_apply(context
->capability_ambient_set
, true);
3480 *exit_status
= EXIT_CAPABILITIES
;
3481 return log_unit_error_errno(unit
, r
, "Failed to apply ambient capabilities (before UID change): %m");
3487 if (uid_is_valid(uid
)) {
3488 r
= enforce_user(context
, uid
);
3490 *exit_status
= EXIT_USER
;
3491 return log_unit_error_errno(unit
, r
, "Failed to change UID to " UID_FMT
": %m", uid
);
3494 if (!needs_ambient_hack
&&
3495 context
->capability_ambient_set
!= 0) {
3497 /* Fix the ambient capabilities after user change. */
3498 r
= capability_ambient_set_apply(context
->capability_ambient_set
, false);
3500 *exit_status
= EXIT_CAPABILITIES
;
3501 return log_unit_error_errno(unit
, r
, "Failed to apply ambient capabilities (after UID change): %m");
3504 /* If we were asked to change user and ambient capabilities
3505 * were requested, we had to add keep-caps to the securebits
3506 * so that we would maintain the inherited capability set
3507 * through the setresuid(). Make sure that the bit is added
3508 * also to the context secure_bits so that we don't try to
3509 * drop the bit away next. */
3511 secure_bits
|= 1<<SECURE_KEEP_CAPS
;
3516 /* Apply working directory here, because the working directory might be on NFS and only the user running
3517 * this service might have the correct privilege to change to the working directory */
3518 r
= apply_working_directory(context
, params
, home
, needs_mount_namespace
, exit_status
);
3520 return log_unit_error_errno(unit
, r
, "Changing to the requested working directory failed: %m");
3522 if (needs_sandboxing
) {
3523 /* Apply other MAC contexts late, but before seccomp syscall filtering, as those should really be last to
3524 * influence our own codepaths as little as possible. Moreover, applying MAC contexts usually requires
3525 * syscalls that are subject to seccomp filtering, hence should probably be applied before the syscalls
3526 * are restricted. */
3530 char *exec_context
= mac_selinux_context_net
?: context
->selinux_context
;
3533 r
= setexeccon(exec_context
);
3535 *exit_status
= EXIT_SELINUX_CONTEXT
;
3536 return log_unit_error_errno(unit
, r
, "Failed to change SELinux context to %s: %m", exec_context
);
3543 if (use_apparmor
&& context
->apparmor_profile
) {
3544 r
= aa_change_onexec(context
->apparmor_profile
);
3545 if (r
< 0 && !context
->apparmor_profile_ignore
) {
3546 *exit_status
= EXIT_APPARMOR_PROFILE
;
3547 return log_unit_error_errno(unit
, errno
, "Failed to prepare AppArmor profile change to %s: %m", context
->apparmor_profile
);
3552 /* PR_GET_SECUREBITS is not privileged, while PR_SET_SECUREBITS is. So to suppress potential EPERMs
3553 * we'll try not to call PR_SET_SECUREBITS unless necessary. */
3554 if (prctl(PR_GET_SECUREBITS
) != secure_bits
)
3555 if (prctl(PR_SET_SECUREBITS
, secure_bits
) < 0) {
3556 *exit_status
= EXIT_SECUREBITS
;
3557 return log_unit_error_errno(unit
, errno
, "Failed to set process secure bits: %m");
3560 if (context_has_no_new_privileges(context
))
3561 if (prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0) < 0) {
3562 *exit_status
= EXIT_NO_NEW_PRIVILEGES
;
3563 return log_unit_error_errno(unit
, errno
, "Failed to disable new privileges: %m");
3567 r
= apply_address_families(unit
, context
);
3569 *exit_status
= EXIT_ADDRESS_FAMILIES
;
3570 return log_unit_error_errno(unit
, r
, "Failed to restrict address families: %m");
3573 r
= apply_memory_deny_write_execute(unit
, context
);
3575 *exit_status
= EXIT_SECCOMP
;
3576 return log_unit_error_errno(unit
, r
, "Failed to disable writing to executable memory: %m");
3579 r
= apply_restrict_realtime(unit
, context
);
3581 *exit_status
= EXIT_SECCOMP
;
3582 return log_unit_error_errno(unit
, r
, "Failed to apply realtime restrictions: %m");
3585 r
= apply_restrict_suid_sgid(unit
, context
);
3587 *exit_status
= EXIT_SECCOMP
;
3588 return log_unit_error_errno(unit
, r
, "Failed to apply SUID/SGID restrictions: %m");
3591 r
= apply_restrict_namespaces(unit
, context
);
3593 *exit_status
= EXIT_SECCOMP
;
3594 return log_unit_error_errno(unit
, r
, "Failed to apply namespace restrictions: %m");
3597 r
= apply_protect_sysctl(unit
, context
);
3599 *exit_status
= EXIT_SECCOMP
;
3600 return log_unit_error_errno(unit
, r
, "Failed to apply sysctl restrictions: %m");
3603 r
= apply_protect_kernel_modules(unit
, context
);
3605 *exit_status
= EXIT_SECCOMP
;
3606 return log_unit_error_errno(unit
, r
, "Failed to apply module loading restrictions: %m");
3609 r
= apply_private_devices(unit
, context
);
3611 *exit_status
= EXIT_SECCOMP
;
3612 return log_unit_error_errno(unit
, r
, "Failed to set up private devices: %m");
3615 r
= apply_syscall_archs(unit
, context
);
3617 *exit_status
= EXIT_SECCOMP
;
3618 return log_unit_error_errno(unit
, r
, "Failed to apply syscall architecture restrictions: %m");
3621 r
= apply_lock_personality(unit
, context
);
3623 *exit_status
= EXIT_SECCOMP
;
3624 return log_unit_error_errno(unit
, r
, "Failed to lock personalities: %m");
3627 /* This really should remain the last step before the execve(), to make sure our own code is unaffected
3628 * by the filter as little as possible. */
3629 r
= apply_syscall_filter(unit
, context
, needs_ambient_hack
);
3631 *exit_status
= EXIT_SECCOMP
;
3632 return log_unit_error_errno(unit
, r
, "Failed to apply system call filters: %m");
3637 if (!strv_isempty(context
->unset_environment
)) {
3640 ee
= strv_env_delete(accum_env
, 1, context
->unset_environment
);
3642 *exit_status
= EXIT_MEMORY
;
3646 strv_free_and_replace(accum_env
, ee
);
3649 if (!FLAGS_SET(command
->flags
, EXEC_COMMAND_NO_ENV_EXPAND
)) {
3650 replaced_argv
= replace_env_argv(command
->argv
, accum_env
);
3651 if (!replaced_argv
) {
3652 *exit_status
= EXIT_MEMORY
;
3655 final_argv
= replaced_argv
;
3657 final_argv
= command
->argv
;
3659 if (DEBUG_LOGGING
) {
3660 _cleanup_free_
char *line
;
3662 line
= exec_command_line(final_argv
);
3664 log_struct(LOG_DEBUG
,
3665 "EXECUTABLE=%s", command
->path
,
3666 LOG_UNIT_MESSAGE(unit
, "Executing: %s", line
),
3668 LOG_UNIT_INVOCATION_ID(unit
));
3674 /* We have finished with all our initializations. Let's now let the manager know that. From this point
3675 * on, if the manager sees POLLHUP on the exec_fd, then execve() was successful. */
3677 if (write(exec_fd
, &hot
, sizeof(hot
)) < 0) {
3678 *exit_status
= EXIT_EXEC
;
3679 return log_unit_error_errno(unit
, errno
, "Failed to enable exec_fd: %m");
3683 execve(command
->path
, final_argv
, accum_env
);
3689 /* The execve() failed. This means the exec_fd is still open. Which means we need to tell the manager
3690 * that POLLHUP on it no longer means execve() succeeded. */
3692 if (write(exec_fd
, &hot
, sizeof(hot
)) < 0) {
3693 *exit_status
= EXIT_EXEC
;
3694 return log_unit_error_errno(unit
, errno
, "Failed to disable exec_fd: %m");
3698 if (r
== -ENOENT
&& (command
->flags
& EXEC_COMMAND_IGNORE_FAILURE
)) {
3699 log_struct_errno(LOG_INFO
, r
,
3700 "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR
,
3702 LOG_UNIT_INVOCATION_ID(unit
),
3703 LOG_UNIT_MESSAGE(unit
, "Executable %s missing, skipping: %m",
3705 "EXECUTABLE=%s", command
->path
);
3709 *exit_status
= EXIT_EXEC
;
3710 return log_unit_error_errno(unit
, r
, "Failed to execute command: %m");
3713 static int exec_context_load_environment(const Unit
*unit
, const ExecContext
*c
, char ***l
);
3714 static int exec_context_named_iofds(const ExecContext
*c
, const ExecParameters
*p
, int named_iofds
[3]);
3716 int exec_spawn(Unit
*unit
,
3717 ExecCommand
*command
,
3718 const ExecContext
*context
,
3719 const ExecParameters
*params
,
3720 ExecRuntime
*runtime
,
3721 DynamicCreds
*dcreds
,
3724 int socket_fd
, r
, named_iofds
[3] = { -1, -1, -1 }, *fds
= NULL
;
3725 _cleanup_free_
char *subcgroup_path
= NULL
;
3726 _cleanup_strv_free_
char **files_env
= NULL
;
3727 size_t n_storage_fds
= 0, n_socket_fds
= 0;
3728 _cleanup_free_
char *line
= NULL
;
3736 assert(params
->fds
|| (params
->n_socket_fds
+ params
->n_storage_fds
<= 0));
3738 if (context
->std_input
== EXEC_INPUT_SOCKET
||
3739 context
->std_output
== EXEC_OUTPUT_SOCKET
||
3740 context
->std_error
== EXEC_OUTPUT_SOCKET
) {
3742 if (params
->n_socket_fds
> 1) {
3743 log_unit_error(unit
, "Got more than one socket.");
3747 if (params
->n_socket_fds
== 0) {
3748 log_unit_error(unit
, "Got no socket.");
3752 socket_fd
= params
->fds
[0];
3756 n_socket_fds
= params
->n_socket_fds
;
3757 n_storage_fds
= params
->n_storage_fds
;
3760 r
= exec_context_named_iofds(context
, params
, named_iofds
);
3762 return log_unit_error_errno(unit
, r
, "Failed to load a named file descriptor: %m");
3764 r
= exec_context_load_environment(unit
, context
, &files_env
);
3766 return log_unit_error_errno(unit
, r
, "Failed to load environment files: %m");
3768 line
= exec_command_line(command
->argv
);
3772 log_struct(LOG_DEBUG
,
3773 LOG_UNIT_MESSAGE(unit
, "About to execute: %s", line
),
3774 "EXECUTABLE=%s", command
->path
,
3776 LOG_UNIT_INVOCATION_ID(unit
));
3778 if (params
->cgroup_path
) {
3779 r
= exec_parameters_get_cgroup_path(params
, &subcgroup_path
);
3781 return log_unit_error_errno(unit
, r
, "Failed to acquire subcgroup path: %m");
3782 if (r
> 0) { /* We are using a child cgroup */
3783 r
= cg_create(SYSTEMD_CGROUP_CONTROLLER
, subcgroup_path
);
3785 return log_unit_error_errno(unit
, r
, "Failed to create control group '%s': %m", subcgroup_path
);
3791 return log_unit_error_errno(unit
, errno
, "Failed to fork: %m");
3794 int exit_status
= EXIT_SUCCESS
;
3796 r
= exec_child(unit
,
3808 unit
->manager
->user_lookup_fds
[1],
3812 log_struct_errno(LOG_ERR
, r
,
3813 "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR
,
3815 LOG_UNIT_INVOCATION_ID(unit
),
3816 LOG_UNIT_MESSAGE(unit
, "Failed at step %s spawning %s: %m",
3817 exit_status_to_string(exit_status
, EXIT_STATUS_SYSTEMD
),
3819 "EXECUTABLE=%s", command
->path
);
3824 log_unit_debug(unit
, "Forked %s as "PID_FMT
, command
->path
, pid
);
3826 /* We add the new process to the cgroup both in the child (so that we can be sure that no user code is ever
3827 * executed outside of the cgroup) and in the parent (so that we can be sure that when we kill the cgroup the
3828 * process will be killed too). */
3830 (void) cg_attach(SYSTEMD_CGROUP_CONTROLLER
, subcgroup_path
, pid
);
3832 exec_status_start(&command
->exec_status
, pid
);
3838 void exec_context_init(ExecContext
*c
) {
3839 ExecDirectoryType i
;
3844 c
->ioprio
= IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE
, 0);
3845 c
->cpu_sched_policy
= SCHED_OTHER
;
3846 c
->syslog_priority
= LOG_DAEMON
|LOG_INFO
;
3847 c
->syslog_level_prefix
= true;
3848 c
->ignore_sigpipe
= true;
3849 c
->timer_slack_nsec
= NSEC_INFINITY
;
3850 c
->personality
= PERSONALITY_INVALID
;
3851 for (i
= 0; i
< _EXEC_DIRECTORY_TYPE_MAX
; i
++)
3852 c
->directories
[i
].mode
= 0755;
3853 c
->capability_bounding_set
= CAP_ALL
;
3854 assert_cc(NAMESPACE_FLAGS_INITIAL
!= NAMESPACE_FLAGS_ALL
);
3855 c
->restrict_namespaces
= NAMESPACE_FLAGS_INITIAL
;
3856 c
->log_level_max
= -1;
3859 void exec_context_done(ExecContext
*c
) {
3860 ExecDirectoryType i
;
3865 c
->environment
= strv_free(c
->environment
);
3866 c
->environment_files
= strv_free(c
->environment_files
);
3867 c
->pass_environment
= strv_free(c
->pass_environment
);
3868 c
->unset_environment
= strv_free(c
->unset_environment
);
3870 rlimit_free_all(c
->rlimit
);
3872 for (l
= 0; l
< 3; l
++) {
3873 c
->stdio_fdname
[l
] = mfree(c
->stdio_fdname
[l
]);
3874 c
->stdio_file
[l
] = mfree(c
->stdio_file
[l
]);
3877 c
->working_directory
= mfree(c
->working_directory
);
3878 c
->root_directory
= mfree(c
->root_directory
);
3879 c
->root_image
= mfree(c
->root_image
);
3880 c
->tty_path
= mfree(c
->tty_path
);
3881 c
->syslog_identifier
= mfree(c
->syslog_identifier
);
3882 c
->user
= mfree(c
->user
);
3883 c
->group
= mfree(c
->group
);
3885 c
->supplementary_groups
= strv_free(c
->supplementary_groups
);
3887 c
->pam_name
= mfree(c
->pam_name
);
3889 c
->read_only_paths
= strv_free(c
->read_only_paths
);
3890 c
->read_write_paths
= strv_free(c
->read_write_paths
);
3891 c
->inaccessible_paths
= strv_free(c
->inaccessible_paths
);
3893 bind_mount_free_many(c
->bind_mounts
, c
->n_bind_mounts
);
3894 c
->bind_mounts
= NULL
;
3895 c
->n_bind_mounts
= 0;
3896 temporary_filesystem_free_many(c
->temporary_filesystems
, c
->n_temporary_filesystems
);
3897 c
->temporary_filesystems
= NULL
;
3898 c
->n_temporary_filesystems
= 0;
3900 cpu_set_reset(&c
->cpu_set
);
3902 c
->utmp_id
= mfree(c
->utmp_id
);
3903 c
->selinux_context
= mfree(c
->selinux_context
);
3904 c
->apparmor_profile
= mfree(c
->apparmor_profile
);
3905 c
->smack_process_label
= mfree(c
->smack_process_label
);
3907 c
->syscall_filter
= hashmap_free(c
->syscall_filter
);
3908 c
->syscall_archs
= set_free(c
->syscall_archs
);
3909 c
->address_families
= set_free(c
->address_families
);
3911 for (i
= 0; i
< _EXEC_DIRECTORY_TYPE_MAX
; i
++)
3912 c
->directories
[i
].paths
= strv_free(c
->directories
[i
].paths
);
3914 c
->log_level_max
= -1;
3916 exec_context_free_log_extra_fields(c
);
3918 c
->log_rate_limit_interval_usec
= 0;
3919 c
->log_rate_limit_burst
= 0;
3921 c
->stdin_data
= mfree(c
->stdin_data
);
3922 c
->stdin_data_size
= 0;
3924 c
->network_namespace_path
= mfree(c
->network_namespace_path
);
3927 int exec_context_destroy_runtime_directory(const ExecContext
*c
, const char *runtime_prefix
) {
3932 if (!runtime_prefix
)
3935 STRV_FOREACH(i
, c
->directories
[EXEC_DIRECTORY_RUNTIME
].paths
) {
3936 _cleanup_free_
char *p
;
3938 p
= path_join(runtime_prefix
, *i
);
3942 /* We execute this synchronously, since we need to be sure this is gone when we start the
3944 (void) rm_rf(p
, REMOVE_ROOT
);
3950 static void exec_command_done(ExecCommand
*c
) {
3953 c
->path
= mfree(c
->path
);
3954 c
->argv
= strv_free(c
->argv
);
3957 void exec_command_done_array(ExecCommand
*c
, size_t n
) {
3960 for (i
= 0; i
< n
; i
++)
3961 exec_command_done(c
+i
);
3964 ExecCommand
* exec_command_free_list(ExecCommand
*c
) {
3968 LIST_REMOVE(command
, c
, i
);
3969 exec_command_done(i
);
3976 void exec_command_free_array(ExecCommand
**c
, size_t n
) {
3979 for (i
= 0; i
< n
; i
++)
3980 c
[i
] = exec_command_free_list(c
[i
]);
3983 void exec_command_reset_status_array(ExecCommand
*c
, size_t n
) {
3986 for (i
= 0; i
< n
; i
++)
3987 exec_status_reset(&c
[i
].exec_status
);
3990 void exec_command_reset_status_list_array(ExecCommand
**c
, size_t n
) {
3993 for (i
= 0; i
< n
; i
++) {
3996 LIST_FOREACH(command
, z
, c
[i
])
3997 exec_status_reset(&z
->exec_status
);
4001 typedef struct InvalidEnvInfo
{
4006 static void invalid_env(const char *p
, void *userdata
) {
4007 InvalidEnvInfo
*info
= userdata
;
4009 log_unit_error(info
->unit
, "Ignoring invalid environment assignment '%s': %s", p
, info
->path
);
4012 const char* exec_context_fdname(const ExecContext
*c
, int fd_index
) {
4018 if (c
->std_input
!= EXEC_INPUT_NAMED_FD
)
4021 return c
->stdio_fdname
[STDIN_FILENO
] ?: "stdin";
4024 if (c
->std_output
!= EXEC_OUTPUT_NAMED_FD
)
4027 return c
->stdio_fdname
[STDOUT_FILENO
] ?: "stdout";
4030 if (c
->std_error
!= EXEC_OUTPUT_NAMED_FD
)
4033 return c
->stdio_fdname
[STDERR_FILENO
] ?: "stderr";
4040 static int exec_context_named_iofds(const ExecContext
*c
, const ExecParameters
*p
, int named_iofds
[static 3]) {
4042 const char* stdio_fdname
[3];
4048 targets
= (c
->std_input
== EXEC_INPUT_NAMED_FD
) +
4049 (c
->std_output
== EXEC_OUTPUT_NAMED_FD
) +
4050 (c
->std_error
== EXEC_OUTPUT_NAMED_FD
);
4052 for (i
= 0; i
< 3; i
++)
4053 stdio_fdname
[i
] = exec_context_fdname(c
, i
);
4055 n_fds
= p
->n_storage_fds
+ p
->n_socket_fds
;
4057 for (i
= 0; i
< n_fds
&& targets
> 0; i
++)
4058 if (named_iofds
[STDIN_FILENO
] < 0 &&
4059 c
->std_input
== EXEC_INPUT_NAMED_FD
&&
4060 stdio_fdname
[STDIN_FILENO
] &&
4061 streq(p
->fd_names
[i
], stdio_fdname
[STDIN_FILENO
])) {
4063 named_iofds
[STDIN_FILENO
] = p
->fds
[i
];
4066 } else if (named_iofds
[STDOUT_FILENO
] < 0 &&
4067 c
->std_output
== EXEC_OUTPUT_NAMED_FD
&&
4068 stdio_fdname
[STDOUT_FILENO
] &&
4069 streq(p
->fd_names
[i
], stdio_fdname
[STDOUT_FILENO
])) {
4071 named_iofds
[STDOUT_FILENO
] = p
->fds
[i
];
4074 } else if (named_iofds
[STDERR_FILENO
] < 0 &&
4075 c
->std_error
== EXEC_OUTPUT_NAMED_FD
&&
4076 stdio_fdname
[STDERR_FILENO
] &&
4077 streq(p
->fd_names
[i
], stdio_fdname
[STDERR_FILENO
])) {
4079 named_iofds
[STDERR_FILENO
] = p
->fds
[i
];
4083 return targets
== 0 ? 0 : -ENOENT
;
4086 static int exec_context_load_environment(const Unit
*unit
, const ExecContext
*c
, char ***l
) {
4087 char **i
, **r
= NULL
;
4092 STRV_FOREACH(i
, c
->environment_files
) {
4096 bool ignore
= false;
4098 _cleanup_globfree_ glob_t pglob
= {};
4107 if (!path_is_absolute(fn
)) {
4115 /* Filename supports globbing, take all matching files */
4116 k
= safe_glob(fn
, 0, &pglob
);
4125 /* When we don't match anything, -ENOENT should be returned */
4126 assert(pglob
.gl_pathc
> 0);
4128 for (n
= 0; n
< pglob
.gl_pathc
; n
++) {
4129 k
= load_env_file(NULL
, pglob
.gl_pathv
[n
], &p
);
4137 /* Log invalid environment variables with filename */
4139 InvalidEnvInfo info
= {
4141 .path
= pglob
.gl_pathv
[n
]
4144 p
= strv_env_clean_with_callback(p
, invalid_env
, &info
);
4152 m
= strv_env_merge(2, r
, p
);
4168 static bool tty_may_match_dev_console(const char *tty
) {
4169 _cleanup_free_
char *resolved
= NULL
;
4174 tty
= skip_dev_prefix(tty
);
4176 /* trivial identity? */
4177 if (streq(tty
, "console"))
4180 if (resolve_dev_console(&resolved
) < 0)
4181 return true; /* if we could not resolve, assume it may */
4183 /* "tty0" means the active VC, so it may be the same sometimes */
4184 return path_equal(resolved
, tty
) || (streq(resolved
, "tty0") && tty_is_vc(tty
));
4187 static bool exec_context_may_touch_tty(const ExecContext
*ec
) {
4190 return ec
->tty_reset
||
4192 ec
->tty_vt_disallocate
||
4193 is_terminal_input(ec
->std_input
) ||
4194 is_terminal_output(ec
->std_output
) ||
4195 is_terminal_output(ec
->std_error
);
4198 bool exec_context_may_touch_console(const ExecContext
*ec
) {
4200 return exec_context_may_touch_tty(ec
) &&
4201 tty_may_match_dev_console(exec_context_tty_path(ec
));
4204 static void strv_fprintf(FILE *f
, char **l
) {
4210 fprintf(f
, " %s", *g
);
4213 void exec_context_dump(const ExecContext
*c
, FILE* f
, const char *prefix
) {
4214 ExecDirectoryType dt
;
4222 prefix
= strempty(prefix
);
4226 "%sWorkingDirectory: %s\n"
4227 "%sRootDirectory: %s\n"
4228 "%sNonBlocking: %s\n"
4229 "%sPrivateTmp: %s\n"
4230 "%sPrivateDevices: %s\n"
4231 "%sProtectKernelTunables: %s\n"
4232 "%sProtectKernelModules: %s\n"
4233 "%sProtectControlGroups: %s\n"
4234 "%sPrivateNetwork: %s\n"
4235 "%sPrivateUsers: %s\n"
4236 "%sProtectHome: %s\n"
4237 "%sProtectSystem: %s\n"
4238 "%sMountAPIVFS: %s\n"
4239 "%sIgnoreSIGPIPE: %s\n"
4240 "%sMemoryDenyWriteExecute: %s\n"
4241 "%sRestrictRealtime: %s\n"
4242 "%sRestrictSUIDSGID: %s\n"
4243 "%sKeyringMode: %s\n"
4244 "%sProtectHostname: %s\n",
4246 prefix
, c
->working_directory
? c
->working_directory
: "/",
4247 prefix
, c
->root_directory
? c
->root_directory
: "/",
4248 prefix
, yes_no(c
->non_blocking
),
4249 prefix
, yes_no(c
->private_tmp
),
4250 prefix
, yes_no(c
->private_devices
),
4251 prefix
, yes_no(c
->protect_kernel_tunables
),
4252 prefix
, yes_no(c
->protect_kernel_modules
),
4253 prefix
, yes_no(c
->protect_control_groups
),
4254 prefix
, yes_no(c
->private_network
),
4255 prefix
, yes_no(c
->private_users
),
4256 prefix
, protect_home_to_string(c
->protect_home
),
4257 prefix
, protect_system_to_string(c
->protect_system
),
4258 prefix
, yes_no(c
->mount_apivfs
),
4259 prefix
, yes_no(c
->ignore_sigpipe
),
4260 prefix
, yes_no(c
->memory_deny_write_execute
),
4261 prefix
, yes_no(c
->restrict_realtime
),
4262 prefix
, yes_no(c
->restrict_suid_sgid
),
4263 prefix
, exec_keyring_mode_to_string(c
->keyring_mode
),
4264 prefix
, yes_no(c
->protect_hostname
));
4267 fprintf(f
, "%sRootImage: %s\n", prefix
, c
->root_image
);
4269 STRV_FOREACH(e
, c
->environment
)
4270 fprintf(f
, "%sEnvironment: %s\n", prefix
, *e
);
4272 STRV_FOREACH(e
, c
->environment_files
)
4273 fprintf(f
, "%sEnvironmentFile: %s\n", prefix
, *e
);
4275 STRV_FOREACH(e
, c
->pass_environment
)
4276 fprintf(f
, "%sPassEnvironment: %s\n", prefix
, *e
);
4278 STRV_FOREACH(e
, c
->unset_environment
)
4279 fprintf(f
, "%sUnsetEnvironment: %s\n", prefix
, *e
);
4281 fprintf(f
, "%sRuntimeDirectoryPreserve: %s\n", prefix
, exec_preserve_mode_to_string(c
->runtime_directory_preserve_mode
));
4283 for (dt
= 0; dt
< _EXEC_DIRECTORY_TYPE_MAX
; dt
++) {
4284 fprintf(f
, "%s%sMode: %04o\n", prefix
, exec_directory_type_to_string(dt
), c
->directories
[dt
].mode
);
4286 STRV_FOREACH(d
, c
->directories
[dt
].paths
)
4287 fprintf(f
, "%s%s: %s\n", prefix
, exec_directory_type_to_string(dt
), *d
);
4295 if (c
->oom_score_adjust_set
)
4297 "%sOOMScoreAdjust: %i\n",
4298 prefix
, c
->oom_score_adjust
);
4300 for (i
= 0; i
< RLIM_NLIMITS
; i
++)
4302 fprintf(f
, "%sLimit%s: " RLIM_FMT
"\n",
4303 prefix
, rlimit_to_string(i
), c
->rlimit
[i
]->rlim_max
);
4304 fprintf(f
, "%sLimit%sSoft: " RLIM_FMT
"\n",
4305 prefix
, rlimit_to_string(i
), c
->rlimit
[i
]->rlim_cur
);
4308 if (c
->ioprio_set
) {
4309 _cleanup_free_
char *class_str
= NULL
;
4311 r
= ioprio_class_to_string_alloc(IOPRIO_PRIO_CLASS(c
->ioprio
), &class_str
);
4313 fprintf(f
, "%sIOSchedulingClass: %s\n", prefix
, class_str
);
4315 fprintf(f
, "%sIOPriority: %lu\n", prefix
, IOPRIO_PRIO_DATA(c
->ioprio
));
4318 if (c
->cpu_sched_set
) {
4319 _cleanup_free_
char *policy_str
= NULL
;
4321 r
= sched_policy_to_string_alloc(c
->cpu_sched_policy
, &policy_str
);
4323 fprintf(f
, "%sCPUSchedulingPolicy: %s\n", prefix
, policy_str
);
4326 "%sCPUSchedulingPriority: %i\n"
4327 "%sCPUSchedulingResetOnFork: %s\n",
4328 prefix
, c
->cpu_sched_priority
,
4329 prefix
, yes_no(c
->cpu_sched_reset_on_fork
));
4332 if (c
->cpu_set
.set
) {
4333 _cleanup_free_
char *affinity
= NULL
;
4335 affinity
= cpu_set_to_range_string(&c
->cpu_set
);
4336 fprintf(f
, "%sCPUAffinity: %s\n", prefix
, affinity
);
4339 if (c
->timer_slack_nsec
!= NSEC_INFINITY
)
4340 fprintf(f
, "%sTimerSlackNSec: "NSEC_FMT
"\n", prefix
, c
->timer_slack_nsec
);
4343 "%sStandardInput: %s\n"
4344 "%sStandardOutput: %s\n"
4345 "%sStandardError: %s\n",
4346 prefix
, exec_input_to_string(c
->std_input
),
4347 prefix
, exec_output_to_string(c
->std_output
),
4348 prefix
, exec_output_to_string(c
->std_error
));
4350 if (c
->std_input
== EXEC_INPUT_NAMED_FD
)
4351 fprintf(f
, "%sStandardInputFileDescriptorName: %s\n", prefix
, c
->stdio_fdname
[STDIN_FILENO
]);
4352 if (c
->std_output
== EXEC_OUTPUT_NAMED_FD
)
4353 fprintf(f
, "%sStandardOutputFileDescriptorName: %s\n", prefix
, c
->stdio_fdname
[STDOUT_FILENO
]);
4354 if (c
->std_error
== EXEC_OUTPUT_NAMED_FD
)
4355 fprintf(f
, "%sStandardErrorFileDescriptorName: %s\n", prefix
, c
->stdio_fdname
[STDERR_FILENO
]);
4357 if (c
->std_input
== EXEC_INPUT_FILE
)
4358 fprintf(f
, "%sStandardInputFile: %s\n", prefix
, c
->stdio_file
[STDIN_FILENO
]);
4359 if (c
->std_output
== EXEC_OUTPUT_FILE
)
4360 fprintf(f
, "%sStandardOutputFile: %s\n", prefix
, c
->stdio_file
[STDOUT_FILENO
]);
4361 if (c
->std_output
== EXEC_OUTPUT_FILE_APPEND
)
4362 fprintf(f
, "%sStandardOutputFileToAppend: %s\n", prefix
, c
->stdio_file
[STDOUT_FILENO
]);
4363 if (c
->std_error
== EXEC_OUTPUT_FILE
)
4364 fprintf(f
, "%sStandardErrorFile: %s\n", prefix
, c
->stdio_file
[STDERR_FILENO
]);
4365 if (c
->std_error
== EXEC_OUTPUT_FILE_APPEND
)
4366 fprintf(f
, "%sStandardErrorFileToAppend: %s\n", prefix
, c
->stdio_file
[STDERR_FILENO
]);
4372 "%sTTYVHangup: %s\n"
4373 "%sTTYVTDisallocate: %s\n",
4374 prefix
, c
->tty_path
,
4375 prefix
, yes_no(c
->tty_reset
),
4376 prefix
, yes_no(c
->tty_vhangup
),
4377 prefix
, yes_no(c
->tty_vt_disallocate
));
4379 if (IN_SET(c
->std_output
,
4382 EXEC_OUTPUT_JOURNAL
,
4383 EXEC_OUTPUT_SYSLOG_AND_CONSOLE
,
4384 EXEC_OUTPUT_KMSG_AND_CONSOLE
,
4385 EXEC_OUTPUT_JOURNAL_AND_CONSOLE
) ||
4386 IN_SET(c
->std_error
,
4389 EXEC_OUTPUT_JOURNAL
,
4390 EXEC_OUTPUT_SYSLOG_AND_CONSOLE
,
4391 EXEC_OUTPUT_KMSG_AND_CONSOLE
,
4392 EXEC_OUTPUT_JOURNAL_AND_CONSOLE
)) {
4394 _cleanup_free_
char *fac_str
= NULL
, *lvl_str
= NULL
;
4396 r
= log_facility_unshifted_to_string_alloc(c
->syslog_priority
>> 3, &fac_str
);
4398 fprintf(f
, "%sSyslogFacility: %s\n", prefix
, fac_str
);
4400 r
= log_level_to_string_alloc(LOG_PRI(c
->syslog_priority
), &lvl_str
);
4402 fprintf(f
, "%sSyslogLevel: %s\n", prefix
, lvl_str
);
4405 if (c
->log_level_max
>= 0) {
4406 _cleanup_free_
char *t
= NULL
;
4408 (void) log_level_to_string_alloc(c
->log_level_max
, &t
);
4410 fprintf(f
, "%sLogLevelMax: %s\n", prefix
, strna(t
));
4413 if (c
->log_rate_limit_interval_usec
> 0) {
4414 char buf_timespan
[FORMAT_TIMESPAN_MAX
];
4417 "%sLogRateLimitIntervalSec: %s\n",
4418 prefix
, format_timespan(buf_timespan
, sizeof(buf_timespan
), c
->log_rate_limit_interval_usec
, USEC_PER_SEC
));
4421 if (c
->log_rate_limit_burst
> 0)
4422 fprintf(f
, "%sLogRateLimitBurst: %u\n", prefix
, c
->log_rate_limit_burst
);
4424 if (c
->n_log_extra_fields
> 0) {
4427 for (j
= 0; j
< c
->n_log_extra_fields
; j
++) {
4428 fprintf(f
, "%sLogExtraFields: ", prefix
);
4429 fwrite(c
->log_extra_fields
[j
].iov_base
,
4430 1, c
->log_extra_fields
[j
].iov_len
,
4436 if (c
->secure_bits
) {
4437 _cleanup_free_
char *str
= NULL
;
4439 r
= secure_bits_to_string_alloc(c
->secure_bits
, &str
);
4441 fprintf(f
, "%sSecure Bits: %s\n", prefix
, str
);
4444 if (c
->capability_bounding_set
!= CAP_ALL
) {
4445 _cleanup_free_
char *str
= NULL
;
4447 r
= capability_set_to_string_alloc(c
->capability_bounding_set
, &str
);
4449 fprintf(f
, "%sCapabilityBoundingSet: %s\n", prefix
, str
);
4452 if (c
->capability_ambient_set
!= 0) {
4453 _cleanup_free_
char *str
= NULL
;
4455 r
= capability_set_to_string_alloc(c
->capability_ambient_set
, &str
);
4457 fprintf(f
, "%sAmbientCapabilities: %s\n", prefix
, str
);
4461 fprintf(f
, "%sUser: %s\n", prefix
, c
->user
);
4463 fprintf(f
, "%sGroup: %s\n", prefix
, c
->group
);
4465 fprintf(f
, "%sDynamicUser: %s\n", prefix
, yes_no(c
->dynamic_user
));
4467 if (!strv_isempty(c
->supplementary_groups
)) {
4468 fprintf(f
, "%sSupplementaryGroups:", prefix
);
4469 strv_fprintf(f
, c
->supplementary_groups
);
4474 fprintf(f
, "%sPAMName: %s\n", prefix
, c
->pam_name
);
4476 if (!strv_isempty(c
->read_write_paths
)) {
4477 fprintf(f
, "%sReadWritePaths:", prefix
);
4478 strv_fprintf(f
, c
->read_write_paths
);
4482 if (!strv_isempty(c
->read_only_paths
)) {
4483 fprintf(f
, "%sReadOnlyPaths:", prefix
);
4484 strv_fprintf(f
, c
->read_only_paths
);
4488 if (!strv_isempty(c
->inaccessible_paths
)) {
4489 fprintf(f
, "%sInaccessiblePaths:", prefix
);
4490 strv_fprintf(f
, c
->inaccessible_paths
);
4494 if (c
->n_bind_mounts
> 0)
4495 for (i
= 0; i
< c
->n_bind_mounts
; i
++)
4496 fprintf(f
, "%s%s: %s%s:%s:%s\n", prefix
,
4497 c
->bind_mounts
[i
].read_only
? "BindReadOnlyPaths" : "BindPaths",
4498 c
->bind_mounts
[i
].ignore_enoent
? "-": "",
4499 c
->bind_mounts
[i
].source
,
4500 c
->bind_mounts
[i
].destination
,
4501 c
->bind_mounts
[i
].recursive
? "rbind" : "norbind");
4503 if (c
->n_temporary_filesystems
> 0)
4504 for (i
= 0; i
< c
->n_temporary_filesystems
; i
++) {
4505 TemporaryFileSystem
*t
= c
->temporary_filesystems
+ i
;
4507 fprintf(f
, "%sTemporaryFileSystem: %s%s%s\n", prefix
,
4509 isempty(t
->options
) ? "" : ":",
4510 strempty(t
->options
));
4515 "%sUtmpIdentifier: %s\n",
4516 prefix
, c
->utmp_id
);
4518 if (c
->selinux_context
)
4520 "%sSELinuxContext: %s%s\n",
4521 prefix
, c
->selinux_context_ignore
? "-" : "", c
->selinux_context
);
4523 if (c
->apparmor_profile
)
4525 "%sAppArmorProfile: %s%s\n",
4526 prefix
, c
->apparmor_profile_ignore
? "-" : "", c
->apparmor_profile
);
4528 if (c
->smack_process_label
)
4530 "%sSmackProcessLabel: %s%s\n",
4531 prefix
, c
->smack_process_label_ignore
? "-" : "", c
->smack_process_label
);
4533 if (c
->personality
!= PERSONALITY_INVALID
)
4535 "%sPersonality: %s\n",
4536 prefix
, strna(personality_to_string(c
->personality
)));
4539 "%sLockPersonality: %s\n",
4540 prefix
, yes_no(c
->lock_personality
));
4542 if (c
->syscall_filter
) {
4550 "%sSystemCallFilter: ",
4553 if (!c
->syscall_whitelist
)
4557 HASHMAP_FOREACH_KEY(val
, id
, c
->syscall_filter
, j
) {
4558 _cleanup_free_
char *name
= NULL
;
4559 const char *errno_name
= NULL
;
4560 int num
= PTR_TO_INT(val
);
4567 name
= seccomp_syscall_resolve_num_arch(SCMP_ARCH_NATIVE
, PTR_TO_INT(id
) - 1);
4568 fputs(strna(name
), f
);
4571 errno_name
= errno_to_name(num
);
4573 fprintf(f
, ":%s", errno_name
);
4575 fprintf(f
, ":%d", num
);
4583 if (c
->syscall_archs
) {
4590 "%sSystemCallArchitectures:",
4594 SET_FOREACH(id
, c
->syscall_archs
, j
)
4595 fprintf(f
, " %s", strna(seccomp_arch_to_string(PTR_TO_UINT32(id
) - 1)));
4600 if (exec_context_restrict_namespaces_set(c
)) {
4601 _cleanup_free_
char *s
= NULL
;
4603 r
= namespace_flags_to_string(c
->restrict_namespaces
, &s
);
4605 fprintf(f
, "%sRestrictNamespaces: %s\n",
4609 if (c
->network_namespace_path
)
4611 "%sNetworkNamespacePath: %s\n",
4612 prefix
, c
->network_namespace_path
);
4614 if (c
->syscall_errno
> 0) {
4615 const char *errno_name
;
4617 fprintf(f
, "%sSystemCallErrorNumber: ", prefix
);
4619 errno_name
= errno_to_name(c
->syscall_errno
);
4621 fprintf(f
, "%s\n", errno_name
);
4623 fprintf(f
, "%d\n", c
->syscall_errno
);
4627 bool exec_context_maintains_privileges(const ExecContext
*c
) {
4630 /* Returns true if the process forked off would run under
4631 * an unchanged UID or as root. */
4636 if (streq(c
->user
, "root") || streq(c
->user
, "0"))
4642 int exec_context_get_effective_ioprio(const ExecContext
*c
) {
4650 p
= ioprio_get(IOPRIO_WHO_PROCESS
, 0);
4652 return IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE
, 4);
4657 void exec_context_free_log_extra_fields(ExecContext
*c
) {
4662 for (l
= 0; l
< c
->n_log_extra_fields
; l
++)
4663 free(c
->log_extra_fields
[l
].iov_base
);
4664 c
->log_extra_fields
= mfree(c
->log_extra_fields
);
4665 c
->n_log_extra_fields
= 0;
4668 void exec_context_revert_tty(ExecContext
*c
) {
4673 /* First, reset the TTY (possibly kicking everybody else from the TTY) */
4674 exec_context_tty_reset(c
, NULL
);
4676 /* And then undo what chown_terminal() did earlier. Note that we only do this if we have a path
4677 * configured. If the TTY was passed to us as file descriptor we assume the TTY is opened and managed
4678 * by whoever passed it to us and thus knows better when and how to chmod()/chown() it back. */
4680 if (exec_context_may_touch_tty(c
)) {
4683 path
= exec_context_tty_path(c
);
4685 r
= chmod_and_chown(path
, TTY_MODE
, 0, TTY_GID
);
4686 if (r
< 0 && r
!= -ENOENT
)
4687 log_warning_errno(r
, "Failed to reset TTY ownership/access mode of %s, ignoring: %m", path
);
4692 void exec_status_start(ExecStatus
*s
, pid_t pid
) {
4699 dual_timestamp_get(&s
->start_timestamp
);
4702 void exec_status_exit(ExecStatus
*s
, const ExecContext
*context
, pid_t pid
, int code
, int status
) {
4705 if (s
->pid
!= pid
) {
4711 dual_timestamp_get(&s
->exit_timestamp
);
4716 if (context
&& context
->utmp_id
)
4717 (void) utmp_put_dead_process(context
->utmp_id
, pid
, code
, status
);
4720 void exec_status_reset(ExecStatus
*s
) {
4723 *s
= (ExecStatus
) {};
4726 void exec_status_dump(const ExecStatus
*s
, FILE *f
, const char *prefix
) {
4727 char buf
[FORMAT_TIMESTAMP_MAX
];
4735 prefix
= strempty(prefix
);
4738 "%sPID: "PID_FMT
"\n",
4741 if (dual_timestamp_is_set(&s
->start_timestamp
))
4743 "%sStart Timestamp: %s\n",
4744 prefix
, format_timestamp(buf
, sizeof(buf
), s
->start_timestamp
.realtime
));
4746 if (dual_timestamp_is_set(&s
->exit_timestamp
))
4748 "%sExit Timestamp: %s\n"
4750 "%sExit Status: %i\n",
4751 prefix
, format_timestamp(buf
, sizeof(buf
), s
->exit_timestamp
.realtime
),
4752 prefix
, sigchld_code_to_string(s
->code
),
4756 static char *exec_command_line(char **argv
) {
4764 STRV_FOREACH(a
, argv
)
4772 STRV_FOREACH(a
, argv
) {
4779 if (strpbrk(*a
, WHITESPACE
)) {
4790 /* FIXME: this doesn't really handle arguments that have
4791 * spaces and ticks in them */
4796 static void exec_command_dump(ExecCommand
*c
, FILE *f
, const char *prefix
) {
4797 _cleanup_free_
char *cmd
= NULL
;
4798 const char *prefix2
;
4803 prefix
= strempty(prefix
);
4804 prefix2
= strjoina(prefix
, "\t");
4806 cmd
= exec_command_line(c
->argv
);
4808 "%sCommand Line: %s\n",
4809 prefix
, cmd
? cmd
: strerror(ENOMEM
));
4811 exec_status_dump(&c
->exec_status
, f
, prefix2
);
4814 void exec_command_dump_list(ExecCommand
*c
, FILE *f
, const char *prefix
) {
4817 prefix
= strempty(prefix
);
4819 LIST_FOREACH(command
, c
, c
)
4820 exec_command_dump(c
, f
, prefix
);
4823 void exec_command_append_list(ExecCommand
**l
, ExecCommand
*e
) {
4830 /* It's kind of important, that we keep the order here */
4831 LIST_FIND_TAIL(command
, *l
, end
);
4832 LIST_INSERT_AFTER(command
, *l
, end
, e
);
4837 int exec_command_set(ExecCommand
*c
, const char *path
, ...) {
4845 l
= strv_new_ap(path
, ap
);
4857 free_and_replace(c
->path
, p
);
4859 return strv_free_and_replace(c
->argv
, l
);
4862 int exec_command_append(ExecCommand
*c
, const char *path
, ...) {
4863 _cleanup_strv_free_
char **l
= NULL
;
4871 l
= strv_new_ap(path
, ap
);
4877 r
= strv_extend_strv(&c
->argv
, l
, false);
4884 static void *remove_tmpdir_thread(void *p
) {
4885 _cleanup_free_
char *path
= p
;
4887 (void) rm_rf(path
, REMOVE_ROOT
|REMOVE_PHYSICAL
);
4891 static ExecRuntime
* exec_runtime_free(ExecRuntime
*rt
, bool destroy
) {
4898 (void) hashmap_remove(rt
->manager
->exec_runtime_by_id
, rt
->id
);
4900 /* When destroy is true, then rm_rf tmp_dir and var_tmp_dir. */
4901 if (destroy
&& rt
->tmp_dir
) {
4902 log_debug("Spawning thread to nuke %s", rt
->tmp_dir
);
4904 r
= asynchronous_job(remove_tmpdir_thread
, rt
->tmp_dir
);
4906 log_warning_errno(r
, "Failed to nuke %s: %m", rt
->tmp_dir
);
4913 if (destroy
&& rt
->var_tmp_dir
) {
4914 log_debug("Spawning thread to nuke %s", rt
->var_tmp_dir
);
4916 r
= asynchronous_job(remove_tmpdir_thread
, rt
->var_tmp_dir
);
4918 log_warning_errno(r
, "Failed to nuke %s: %m", rt
->var_tmp_dir
);
4919 free(rt
->var_tmp_dir
);
4922 rt
->var_tmp_dir
= NULL
;
4925 rt
->id
= mfree(rt
->id
);
4926 rt
->tmp_dir
= mfree(rt
->tmp_dir
);
4927 rt
->var_tmp_dir
= mfree(rt
->var_tmp_dir
);
4928 safe_close_pair(rt
->netns_storage_socket
);
4932 static void exec_runtime_freep(ExecRuntime
**rt
) {
4933 (void) exec_runtime_free(*rt
, false);
4936 static int exec_runtime_allocate(ExecRuntime
**ret
) {
4941 n
= new(ExecRuntime
, 1);
4945 *n
= (ExecRuntime
) {
4946 .netns_storage_socket
= { -1, -1 },
4953 static int exec_runtime_add(
4956 const char *tmp_dir
,
4957 const char *var_tmp_dir
,
4958 const int netns_storage_socket
[2],
4959 ExecRuntime
**ret
) {
4961 _cleanup_(exec_runtime_freep
) ExecRuntime
*rt
= NULL
;
4967 r
= hashmap_ensure_allocated(&m
->exec_runtime_by_id
, &string_hash_ops
);
4971 r
= exec_runtime_allocate(&rt
);
4975 rt
->id
= strdup(id
);
4980 rt
->tmp_dir
= strdup(tmp_dir
);
4984 /* When tmp_dir is set, then we require var_tmp_dir is also set. */
4985 assert(var_tmp_dir
);
4986 rt
->var_tmp_dir
= strdup(var_tmp_dir
);
4987 if (!rt
->var_tmp_dir
)
4991 if (netns_storage_socket
) {
4992 rt
->netns_storage_socket
[0] = netns_storage_socket
[0];
4993 rt
->netns_storage_socket
[1] = netns_storage_socket
[1];
4996 r
= hashmap_put(m
->exec_runtime_by_id
, rt
->id
, rt
);
5005 /* do not remove created ExecRuntime object when the operation succeeds. */
5010 static int exec_runtime_make(Manager
*m
, const ExecContext
*c
, const char *id
, ExecRuntime
**ret
) {
5011 _cleanup_free_
char *tmp_dir
= NULL
, *var_tmp_dir
= NULL
;
5012 _cleanup_close_pair_
int netns_storage_socket
[2] = { -1, -1 };
5019 /* It is not necessary to create ExecRuntime object. */
5020 if (!c
->private_network
&& !c
->private_tmp
&& !c
->network_namespace_path
)
5023 if (c
->private_tmp
) {
5024 r
= setup_tmp_dirs(id
, &tmp_dir
, &var_tmp_dir
);
5029 if (c
->private_network
|| c
->network_namespace_path
) {
5030 if (socketpair(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
, 0, netns_storage_socket
) < 0)
5034 r
= exec_runtime_add(m
, id
, tmp_dir
, var_tmp_dir
, netns_storage_socket
, ret
);
5039 netns_storage_socket
[0] = netns_storage_socket
[1] = -1;
5043 int exec_runtime_acquire(Manager
*m
, const ExecContext
*c
, const char *id
, bool create
, ExecRuntime
**ret
) {
5051 rt
= hashmap_get(m
->exec_runtime_by_id
, id
);
5053 /* We already have a ExecRuntime object, let's increase the ref count and reuse it */
5059 /* If not found, then create a new object. */
5060 r
= exec_runtime_make(m
, c
, id
, &rt
);
5062 /* When r == 0, it is not necessary to create ExecRuntime object. */
5066 /* increment reference counter. */
5072 ExecRuntime
*exec_runtime_unref(ExecRuntime
*rt
, bool destroy
) {
5076 assert(rt
->n_ref
> 0);
5082 return exec_runtime_free(rt
, destroy
);
5085 int exec_runtime_serialize(const Manager
*m
, FILE *f
, FDSet
*fds
) {
5093 HASHMAP_FOREACH(rt
, m
->exec_runtime_by_id
, i
) {
5094 fprintf(f
, "exec-runtime=%s", rt
->id
);
5097 fprintf(f
, " tmp-dir=%s", rt
->tmp_dir
);
5099 if (rt
->var_tmp_dir
)
5100 fprintf(f
, " var-tmp-dir=%s", rt
->var_tmp_dir
);
5102 if (rt
->netns_storage_socket
[0] >= 0) {
5105 copy
= fdset_put_dup(fds
, rt
->netns_storage_socket
[0]);
5109 fprintf(f
, " netns-socket-0=%i", copy
);
5112 if (rt
->netns_storage_socket
[1] >= 0) {
5115 copy
= fdset_put_dup(fds
, rt
->netns_storage_socket
[1]);
5119 fprintf(f
, " netns-socket-1=%i", copy
);
5128 int exec_runtime_deserialize_compat(Unit
*u
, const char *key
, const char *value
, FDSet
*fds
) {
5129 _cleanup_(exec_runtime_freep
) ExecRuntime
*rt_create
= NULL
;
5133 /* This is for the migration from old (v237 or earlier) deserialization text.
5134 * Due to the bug #7790, this may not work with the units that use JoinsNamespaceOf=.
5135 * Even if the ExecRuntime object originally created by the other unit, we cannot judge
5136 * so or not from the serialized text, then we always creates a new object owned by this. */
5142 /* Manager manages ExecRuntime objects by the unit id.
5143 * So, we omit the serialized text when the unit does not have id (yet?)... */
5144 if (isempty(u
->id
)) {
5145 log_unit_debug(u
, "Invocation ID not found. Dropping runtime parameter.");
5149 r
= hashmap_ensure_allocated(&u
->manager
->exec_runtime_by_id
, &string_hash_ops
);
5151 log_unit_debug_errno(u
, r
, "Failed to allocate storage for runtime parameter: %m");
5155 rt
= hashmap_get(u
->manager
->exec_runtime_by_id
, u
->id
);
5157 r
= exec_runtime_allocate(&rt_create
);
5161 rt_create
->id
= strdup(u
->id
);
5168 if (streq(key
, "tmp-dir")) {
5171 copy
= strdup(value
);
5175 free_and_replace(rt
->tmp_dir
, copy
);
5177 } else if (streq(key
, "var-tmp-dir")) {
5180 copy
= strdup(value
);
5184 free_and_replace(rt
->var_tmp_dir
, copy
);
5186 } else if (streq(key
, "netns-socket-0")) {
5189 if (safe_atoi(value
, &fd
) < 0 || !fdset_contains(fds
, fd
)) {
5190 log_unit_debug(u
, "Failed to parse netns socket value: %s", value
);
5194 safe_close(rt
->netns_storage_socket
[0]);
5195 rt
->netns_storage_socket
[0] = fdset_remove(fds
, fd
);
5197 } else if (streq(key
, "netns-socket-1")) {
5200 if (safe_atoi(value
, &fd
) < 0 || !fdset_contains(fds
, fd
)) {
5201 log_unit_debug(u
, "Failed to parse netns socket value: %s", value
);
5205 safe_close(rt
->netns_storage_socket
[1]);
5206 rt
->netns_storage_socket
[1] = fdset_remove(fds
, fd
);
5210 /* If the object is newly created, then put it to the hashmap which manages ExecRuntime objects. */
5212 r
= hashmap_put(u
->manager
->exec_runtime_by_id
, rt_create
->id
, rt_create
);
5214 log_unit_debug_errno(u
, r
, "Failed to put runtime parameter to manager's storage: %m");
5218 rt_create
->manager
= u
->manager
;
5227 void exec_runtime_deserialize_one(Manager
*m
, const char *value
, FDSet
*fds
) {
5228 char *id
= NULL
, *tmp_dir
= NULL
, *var_tmp_dir
= NULL
;
5229 int r
, fd0
= -1, fd1
= -1;
5230 const char *p
, *v
= value
;
5237 n
= strcspn(v
, " ");
5238 id
= strndupa(v
, n
);
5243 v
= startswith(p
, "tmp-dir=");
5245 n
= strcspn(v
, " ");
5246 tmp_dir
= strndupa(v
, n
);
5252 v
= startswith(p
, "var-tmp-dir=");
5254 n
= strcspn(v
, " ");
5255 var_tmp_dir
= strndupa(v
, n
);
5261 v
= startswith(p
, "netns-socket-0=");
5265 n
= strcspn(v
, " ");
5266 buf
= strndupa(v
, n
);
5267 if (safe_atoi(buf
, &fd0
) < 0 || !fdset_contains(fds
, fd0
)) {
5268 log_debug("Unable to process exec-runtime netns fd specification.");
5271 fd0
= fdset_remove(fds
, fd0
);
5277 v
= startswith(p
, "netns-socket-1=");
5281 n
= strcspn(v
, " ");
5282 buf
= strndupa(v
, n
);
5283 if (safe_atoi(buf
, &fd1
) < 0 || !fdset_contains(fds
, fd1
)) {
5284 log_debug("Unable to process exec-runtime netns fd specification.");
5287 fd1
= fdset_remove(fds
, fd1
);
5292 r
= exec_runtime_add(m
, id
, tmp_dir
, var_tmp_dir
, (int[]) { fd0
, fd1
}, NULL
);
5294 log_debug_errno(r
, "Failed to add exec-runtime: %m");
5297 void exec_runtime_vacuum(Manager
*m
) {
5303 /* Free unreferenced ExecRuntime objects. This is used after manager deserialization process. */
5305 HASHMAP_FOREACH(rt
, m
->exec_runtime_by_id
, i
) {
5309 (void) exec_runtime_free(rt
, false);
5313 void exec_params_clear(ExecParameters
*p
) {
5317 strv_free(p
->environment
);
5320 static const char* const exec_input_table
[_EXEC_INPUT_MAX
] = {
5321 [EXEC_INPUT_NULL
] = "null",
5322 [EXEC_INPUT_TTY
] = "tty",
5323 [EXEC_INPUT_TTY_FORCE
] = "tty-force",
5324 [EXEC_INPUT_TTY_FAIL
] = "tty-fail",
5325 [EXEC_INPUT_SOCKET
] = "socket",
5326 [EXEC_INPUT_NAMED_FD
] = "fd",
5327 [EXEC_INPUT_DATA
] = "data",
5328 [EXEC_INPUT_FILE
] = "file",
5331 DEFINE_STRING_TABLE_LOOKUP(exec_input
, ExecInput
);
5333 static const char* const exec_output_table
[_EXEC_OUTPUT_MAX
] = {
5334 [EXEC_OUTPUT_INHERIT
] = "inherit",
5335 [EXEC_OUTPUT_NULL
] = "null",
5336 [EXEC_OUTPUT_TTY
] = "tty",
5337 [EXEC_OUTPUT_SYSLOG
] = "syslog",
5338 [EXEC_OUTPUT_SYSLOG_AND_CONSOLE
] = "syslog+console",
5339 [EXEC_OUTPUT_KMSG
] = "kmsg",
5340 [EXEC_OUTPUT_KMSG_AND_CONSOLE
] = "kmsg+console",
5341 [EXEC_OUTPUT_JOURNAL
] = "journal",
5342 [EXEC_OUTPUT_JOURNAL_AND_CONSOLE
] = "journal+console",
5343 [EXEC_OUTPUT_SOCKET
] = "socket",
5344 [EXEC_OUTPUT_NAMED_FD
] = "fd",
5345 [EXEC_OUTPUT_FILE
] = "file",
5346 [EXEC_OUTPUT_FILE_APPEND
] = "append",
5349 DEFINE_STRING_TABLE_LOOKUP(exec_output
, ExecOutput
);
5351 static const char* const exec_utmp_mode_table
[_EXEC_UTMP_MODE_MAX
] = {
5352 [EXEC_UTMP_INIT
] = "init",
5353 [EXEC_UTMP_LOGIN
] = "login",
5354 [EXEC_UTMP_USER
] = "user",
5357 DEFINE_STRING_TABLE_LOOKUP(exec_utmp_mode
, ExecUtmpMode
);
5359 static const char* const exec_preserve_mode_table
[_EXEC_PRESERVE_MODE_MAX
] = {
5360 [EXEC_PRESERVE_NO
] = "no",
5361 [EXEC_PRESERVE_YES
] = "yes",
5362 [EXEC_PRESERVE_RESTART
] = "restart",
5365 DEFINE_STRING_TABLE_LOOKUP_WITH_BOOLEAN(exec_preserve_mode
, ExecPreserveMode
, EXEC_PRESERVE_YES
);
5367 static const char* const exec_directory_type_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
5368 [EXEC_DIRECTORY_RUNTIME
] = "RuntimeDirectory",
5369 [EXEC_DIRECTORY_STATE
] = "StateDirectory",
5370 [EXEC_DIRECTORY_CACHE
] = "CacheDirectory",
5371 [EXEC_DIRECTORY_LOGS
] = "LogsDirectory",
5372 [EXEC_DIRECTORY_CONFIGURATION
] = "ConfigurationDirectory",
5375 DEFINE_STRING_TABLE_LOOKUP(exec_directory_type
, ExecDirectoryType
);
5377 static const char* const exec_directory_env_name_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
5378 [EXEC_DIRECTORY_RUNTIME
] = "RUNTIME_DIRECTORY",
5379 [EXEC_DIRECTORY_STATE
] = "STATE_DIRECTORY",
5380 [EXEC_DIRECTORY_CACHE
] = "CACHE_DIRECTORY",
5381 [EXEC_DIRECTORY_LOGS
] = "LOGS_DIRECTORY",
5382 [EXEC_DIRECTORY_CONFIGURATION
] = "CONFIGURATION_DIRECTORY",
5385 DEFINE_PRIVATE_STRING_TABLE_LOOKUP_TO_STRING(exec_directory_env_name
, ExecDirectoryType
);
5387 static const char* const exec_keyring_mode_table
[_EXEC_KEYRING_MODE_MAX
] = {
5388 [EXEC_KEYRING_INHERIT
] = "inherit",
5389 [EXEC_KEYRING_PRIVATE
] = "private",
5390 [EXEC_KEYRING_SHARED
] = "shared",
5393 DEFINE_STRING_TABLE_LOOKUP(exec_keyring_mode
, ExecKeyringMode
);