1 /*-*- Mode: C; c-basic-offset: 8 -*-*/
4 This file is part of systemd.
6 Copyright 2010 Lennart Poettering
8 systemd is free software; you can redistribute it and/or modify it
9 under the terms of the GNU General Public License as published by
10 the Free Software Foundation; either version 2 of the License, or
11 (at your option) any later version.
13 systemd is distributed in the hope that it will be useful, but
14 WITHOUT ANY WARRANTY; without even the implied warranty of
15 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 General Public License for more details.
18 You should have received a copy of the GNU General Public License
19 along with systemd; If not, see <http://www.gnu.org/licenses/>.
29 #include <sys/socket.h>
31 #include <sys/prctl.h>
32 #include <linux/sched.h>
33 #include <sys/types.h>
44 #include "securebits.h"
46 static int close_fds(int except
[], unsigned n_except
) {
51 /* Modifies the fds array! (sorts it) */
53 if (!(d
= opendir("/proc/self/fd")))
56 while ((de
= readdir(d
))) {
59 if (de
->d_name
[0] == '.')
62 if ((r
= safe_atoi(de
->d_name
, &fd
)) < 0)
76 for (i
= 0; i
< n_except
; i
++)
77 if (except
[i
] == fd
) {
86 if ((r
= close_nointr(fd
)) < 0)
95 static int shift_fds(int fds
[], unsigned n_fds
) {
96 int start
, restart_from
;
109 for (i
= start
; i
< (int) n_fds
; i
++) {
112 /* Already at right index? */
116 if ((nfd
= fcntl(fds
[i
], F_DUPFD
, i
+3)) < 0)
119 assert_se(close_nointr(fds
[i
]) == 0);
122 /* Hmm, the fd we wanted isn't free? Then
123 * let's remember that and try again from here*/
124 if (nfd
!= i
+3 && restart_from
< 0)
128 if (restart_from
< 0)
131 start
= restart_from
;
137 static int flags_fds(int fds
[], unsigned n_fds
, bool nonblock
) {
145 /* Drops/Sets O_NONBLOCK and FD_CLOEXEC from the file flags */
147 for (i
= 0; i
< n_fds
; i
++) {
150 if ((flags
= fcntl(fds
[i
], F_GETFL
, 0)) < 0)
156 flags
&= ~O_NONBLOCK
;
158 if (fcntl(fds
[i
], F_SETFL
, flags
) < 0)
161 /* We unconditionally drop FD_CLOEXEC from the fds,
162 * since after all we want to pass these fds to our
164 if ((flags
= fcntl(fds
[i
], F_GETFD
, 0)) < 0)
167 if (fcntl(fds
[i
], F_SETFD
, flags
&~FD_CLOEXEC
) < 0)
174 static int replace_null_fd(int fd
, int flags
) {
180 if ((nfd
= open("/dev/null", flags
|O_NOCTTY
)) < 0)
184 close_nointr_nofail(nfd
);
191 static int setup_output(const ExecContext
*context
, const char *ident
) {
196 switch (context
->output
) {
198 case EXEC_OUTPUT_CONSOLE
:
201 case EXEC_OUTPUT_NULL
:
203 if ((r
= replace_null_fd(STDOUT_FILENO
, O_WRONLY
)) < 0 ||
204 (r
= replace_null_fd(STDERR_FILENO
, O_WRONLY
)) < 0)
209 case EXEC_OUTPUT_KERNEL
:
210 case EXEC_OUTPUT_SYSLOG
: {
215 struct sockaddr_un un
;
218 close_nointr(STDOUT_FILENO
);
219 close_nointr(STDERR_FILENO
);
221 if ((fd
= socket(AF_UNIX
, SOCK_STREAM
, 0)) < 0)
224 if (fd
!= STDOUT_FILENO
) {
225 close_nointr_nofail(fd
);
230 sa
.sa
.sa_family
= AF_UNIX
;
231 strncpy(sa
.un
.sun_path
+1, LOGGER_SOCKET
, sizeof(sa
.un
.sun_path
)-1);
233 if (connect(fd
, &sa
.sa
, sizeof(sa
)) < 0) {
234 close_nointr_nofail(fd
);
238 if (shutdown(fd
, SHUT_RD
) < 0) {
239 close_nointr_nofail(fd
);
243 if ((fd
= dup(fd
)) < 0) {
244 close_nointr_nofail(fd
);
248 if (fd
!= STDERR_FILENO
) {
249 close_nointr_nofail(fd
);
253 /* We speak a very simple protocol between log server
254 * and client: one line for the log destination (kmsg
255 * or syslog), followed by the priority field,
256 * followed by the process name. Since we replaced
257 * stdin/stderr we simple use stdio to write to
258 * it. Note that we use stderr, to minimize buffer
259 * flushing issues. */
265 context
->output
== EXEC_OUTPUT_KERNEL
? "kmsg" : "syslog",
266 context
->syslog_priority
,
267 context
->syslog_identifier
? context
->syslog_identifier
: ident
);
273 assert_not_reached("Unknown output type");
277 static int setup_input(const ExecContext
*context
) {
282 switch (context
->input
) {
284 case EXEC_INPUT_CONSOLE
:
287 case EXEC_INPUT_NULL
:
288 if ((r
= replace_null_fd(STDIN_FILENO
, O_RDONLY
)) < 0)
294 assert_not_reached("Unknown input type");
298 static int get_group_creds(const char *groupname
, gid_t
*gid
) {
305 /* We enforce some special rules for gid=0: in order to avoid
306 * NSS lookups for root we hardcode its data. */
308 if (streq(groupname
, "root") || streq(groupname
, "0")) {
313 if (safe_atolu(groupname
, &lu
) >= 0) {
315 g
= getgrgid((gid_t
) lu
);
318 g
= getgrnam(groupname
);
322 return errno
!= 0 ? -errno
: -ESRCH
;
328 static int get_user_creds(const char **username
, uid_t
*uid
, gid_t
*gid
, const char **home
) {
338 /* We enforce some special rules for uid=0: in order to avoid
339 * NSS lookups for root we hardcode its data. */
341 if (streq(*username
, "root") || streq(*username
, "0")) {
349 if (safe_atolu(*username
, &lu
) >= 0) {
351 p
= getpwuid((uid_t
) lu
);
353 /* If there are multiple users with the same id, make
354 * sure to leave $USER to the configured value instead
355 * of the first occurence in the database. However if
356 * the uid was configured by a numeric uid, then let's
357 * pick the real username from /etc/passwd. */
359 *username
= p
->pw_name
;
362 p
= getpwnam(*username
);
366 return errno
!= 0 ? -errno
: -ESRCH
;
374 static int enforce_groups(const ExecContext
*context
, const char *username
, gid_t gid
) {
375 bool keep_groups
= false;
380 /* Lookup and ser GID and supplementary group list. Here too
381 * we avoid NSS lookups for gid=0. */
383 if (context
->group
|| username
) {
386 if ((r
= get_group_creds(context
->group
, &gid
)) < 0)
389 /* First step, initialize groups from /etc/groups */
390 if (username
&& gid
!= 0) {
391 if (initgroups(username
, gid
) < 0)
397 /* Second step, set our gids */
398 if (setresgid(gid
, gid
, gid
) < 0)
402 if (context
->supplementary_groups
) {
407 /* Final step, initialize any manually set supplementary groups */
408 ngroups_max
= (int) sysconf(_SC_NGROUPS_MAX
);
410 if (!(gids
= new(gid_t
, ngroups_max
)))
414 if ((k
= getgroups(ngroups_max
, gids
)) < 0) {
421 STRV_FOREACH(i
, context
->supplementary_groups
) {
423 if (k
>= ngroups_max
) {
428 if ((r
= get_group_creds(*i
, gids
+k
)) < 0) {
436 if (setgroups(k
, gids
) < 0) {
447 static int enforce_user(const ExecContext
*context
, uid_t uid
) {
451 /* Sets (but doesn't lookup) the uid and make sure we keep the
452 * capabilities while doing so. */
454 if (context
->capabilities
) {
456 static const cap_value_t bits
[] = {
457 CAP_SETUID
, /* Necessary so that we can run setresuid() below */
458 CAP_SETPCAP
/* Necessary so that we can set PR_SET_SECUREBITS later on */
461 /* First step: If we need to keep capabilities but
462 * drop privileges we need to make sure we keep our
463 * caps, whiel we drop priviliges. */
465 if (prctl(PR_SET_SECUREBITS
, context
->secure_bits
|SECURE_KEEP_CAPS
) < 0)
468 /* Second step: set the capabilites. This will reduce
469 * the capabilities to the minimum we need. */
471 if (!(d
= cap_dup(context
->capabilities
)))
474 if (cap_set_flag(d
, CAP_EFFECTIVE
, ELEMENTSOF(bits
), bits
, CAP_SET
) < 0 ||
475 cap_set_flag(d
, CAP_PERMITTED
, ELEMENTSOF(bits
), bits
, CAP_SET
) < 0) {
481 if (cap_set_proc(d
) < 0) {
490 /* Third step: actually set the uids */
491 if (setresuid(uid
, uid
, uid
) < 0)
494 /* At this point we should have all necessary capabilities but
495 are otherwise a normal user. However, the caps might got
496 corrupted due to the setresuid() so we need clean them up
497 later. This is done outside of this call. */
502 int exec_spawn(const ExecCommand
*command
,
503 const ExecContext
*context
,
504 int *fds
, unsigned n_fds
,
505 bool apply_permissions
,
514 assert(fds
|| n_fds
<= 0);
516 log_debug("About to execute %s", command
->path
);
518 if ((pid
= fork()) < 0)
524 const char *username
= NULL
, *home
= NULL
;
525 uid_t uid
= (uid_t
) -1;
526 gid_t gid
= (gid_t
) -1;
527 char **our_env
= NULL
, **final_env
= NULL
;
532 if (sigemptyset(&ss
) < 0 ||
533 sigprocmask(SIG_SETMASK
, &ss
, NULL
) < 0) {
534 r
= EXIT_SIGNAL_MASK
;
538 if (setpgid(0, 0) < 0) {
543 umask(context
->umask
);
545 if (setup_input(context
) < 0) {
550 if (setup_output(context
, file_name_from_path(command
->path
)) < 0) {
555 if (context
->oom_adjust_set
) {
558 snprintf(t
, sizeof(t
), "%i", context
->oom_adjust
);
561 if (write_one_line_file("/proc/self/oom_adj", t
) < 0) {
567 if (context
->nice_set
)
568 if (setpriority(PRIO_PROCESS
, 0, context
->nice
) < 0) {
573 if (context
->cpu_sched_set
) {
574 struct sched_param param
;
577 param
.sched_priority
= context
->cpu_sched_priority
;
579 if (sched_setscheduler(0, context
->cpu_sched_policy
|
580 (context
->cpu_sched_reset_on_fork
? SCHED_RESET_ON_FORK
: 0), ¶m
) < 0) {
581 r
= EXIT_SETSCHEDULER
;
586 if (context
->cpu_affinity_set
)
587 if (sched_setaffinity(0, sizeof(context
->cpu_affinity
), &context
->cpu_affinity
) < 0) {
588 r
= EXIT_CPUAFFINITY
;
592 if (context
->ioprio_set
)
593 if (ioprio_set(IOPRIO_WHO_PROCESS
, 0, context
->ioprio
) < 0) {
598 if (context
->timer_slack_ns_set
)
599 if (prctl(PR_SET_TIMERSLACK
, context
->timer_slack_ns_set
) < 0) {
605 username
= context
->user
;
606 if (get_user_creds(&username
, &uid
, &gid
, &home
) < 0) {
612 if (apply_permissions
)
613 if (enforce_groups(context
, username
, uid
) < 0) {
619 if (context
->root_directory
)
620 if (chroot(context
->root_directory
) < 0) {
625 if (chdir(context
->working_directory
? context
->working_directory
: "/") < 0) {
633 if (asprintf(&d
, "%s/%s",
634 context
->root_directory
? context
->root_directory
: "",
635 context
->working_directory
? context
->working_directory
: "") < 0) {
649 if (close_fds(fds
, n_fds
) < 0 ||
650 shift_fds(fds
, n_fds
) < 0 ||
651 flags_fds(fds
, n_fds
, context
->non_blocking
) < 0) {
656 if (apply_permissions
) {
658 for (i
= 0; i
< RLIMIT_NLIMITS
; i
++) {
659 if (!context
->rlimit
[i
])
662 if (setrlimit(i
, context
->rlimit
[i
]) < 0) {
669 if (enforce_user(context
, uid
) < 0) {
674 if (prctl(PR_SET_SECUREBITS
, context
->secure_bits
) < 0) {
679 if (context
->capabilities
)
680 if (cap_set_proc(context
->capabilities
) < 0) {
681 r
= EXIT_CAPABILITIES
;
686 if (!(our_env
= new0(char*, 6))) {
692 if (asprintf(our_env
+ n_env
++, "LISTEN_PID=%llu", (unsigned long long) getpid()) < 0 ||
693 asprintf(our_env
+ n_env
++, "LISTEN_FDS=%u", n_fds
) < 0) {
699 if (asprintf(our_env
+ n_env
++, "HOME=%s", home
) < 0) {
705 if (asprintf(our_env
+ n_env
++, "LOGNAME=%s", username
) < 0 ||
706 asprintf(our_env
+ n_env
++, "USER=%s", username
) < 0) {
711 if (!(final_env
= strv_env_merge(environ
, our_env
, context
->environment
, NULL
))) {
716 execve(command
->path
, command
->argv
, final_env
);
721 strv_free(final_env
);
727 log_debug("Forked %s as %llu", command
->path
, (unsigned long long) pid
);
733 void exec_context_init(ExecContext
*c
) {
738 c
->oom_adjust_set
= false;
741 c
->ioprio
= IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE
, 0);
742 c
->ioprio_set
= false;
743 c
->cpu_sched_policy
= SCHED_OTHER
;
744 c
->cpu_sched_priority
= 0;
745 c
->cpu_sched_set
= false;
746 CPU_ZERO(&c
->cpu_affinity
);
747 c
->cpu_affinity_set
= false;
751 c
->syslog_priority
= LOG_DAEMON
|LOG_INFO
;
754 c
->capability_bounding_set_drop
= 0;
757 void exec_context_done(ExecContext
*c
) {
762 strv_free(c
->environment
);
763 c
->environment
= NULL
;
765 for (l
= 0; l
< ELEMENTSOF(c
->rlimit
); l
++) {
770 free(c
->working_directory
);
771 c
->working_directory
= NULL
;
772 free(c
->root_directory
);
773 c
->root_directory
= NULL
;
775 free(c
->syslog_identifier
);
776 c
->syslog_identifier
= NULL
;
784 strv_free(c
->supplementary_groups
);
785 c
->supplementary_groups
= NULL
;
787 if (c
->capabilities
) {
788 cap_free(c
->capabilities
);
789 c
->capabilities
= NULL
;
793 void exec_command_free_list(ExecCommand
*c
) {
797 LIST_REMOVE(ExecCommand
, command
, c
, i
);
805 void exec_command_free_array(ExecCommand
**c
, unsigned n
) {
808 for (i
= 0; i
< n
; i
++) {
809 exec_command_free_list(c
[i
]);
814 void exec_context_dump(ExecContext
*c
, FILE* f
, const char *prefix
) {
826 "%sWorkingDirectory: %s\n"
827 "%sRootDirectory: %s\n"
828 "%sNonBlocking: %s\n",
830 prefix
, c
->working_directory
? c
->working_directory
: "/",
831 prefix
, c
->root_directory
? c
->root_directory
: "/",
832 prefix
, yes_no(c
->non_blocking
));
835 for (e
= c
->environment
; *e
; e
++)
836 fprintf(f
, "%sEnvironment: %s\n", prefix
, *e
);
843 if (c
->oom_adjust_set
)
846 prefix
, c
->oom_adjust
);
848 for (i
= 0; i
< RLIM_NLIMITS
; i
++)
850 fprintf(f
, "%s%s: %llu\n", prefix
, rlimit_to_string(i
), (unsigned long long) c
->rlimit
[i
]->rlim_max
);
854 "%sIOSchedulingClass: %s\n"
855 "%sIOPriority: %i\n",
856 prefix
, ioprio_class_to_string(IOPRIO_PRIO_CLASS(c
->ioprio
)),
857 prefix
, (int) IOPRIO_PRIO_DATA(c
->ioprio
));
859 if (c
->cpu_sched_set
)
861 "%sCPUSchedulingPolicy: %s\n"
862 "%sCPUSchedulingPriority: %i\n"
863 "%sCPUSchedulingResetOnFork: %s\n",
864 prefix
, sched_policy_to_string(c
->cpu_sched_policy
),
865 prefix
, c
->cpu_sched_priority
,
866 prefix
, yes_no(c
->cpu_sched_reset_on_fork
));
868 if (c
->cpu_affinity_set
) {
869 fprintf(f
, "%sCPUAffinity:", prefix
);
870 for (i
= 0; i
< CPU_SETSIZE
; i
++)
871 if (CPU_ISSET(i
, &c
->cpu_affinity
))
872 fprintf(f
, " %i", i
);
876 if (c
->timer_slack_ns_set
)
877 fprintf(f
, "%sTimerSlackNS: %lu\n", prefix
, c
->timer_slack_ns
);
882 prefix
, exec_input_to_string(c
->input
),
883 prefix
, exec_output_to_string(c
->output
));
885 if (c
->output
== EXEC_OUTPUT_SYSLOG
|| c
->output
== EXEC_OUTPUT_KERNEL
)
887 "%sSyslogFacility: %s\n"
888 "%sSyslogLevel: %s\n",
889 prefix
, log_facility_to_string(LOG_FAC(c
->syslog_priority
)),
890 prefix
, log_level_to_string(LOG_PRI(c
->syslog_priority
)));
892 if (c
->capabilities
) {
894 if ((t
= cap_to_text(c
->capabilities
, NULL
))) {
895 fprintf(f
, "%sCapabilities: %s\n",
902 fprintf(f
, "%sSecure Bits:%s%s%s%s%s%s\n",
904 (c
->secure_bits
& SECURE_KEEP_CAPS
) ? " keep-caps" : "",
905 (c
->secure_bits
& SECURE_KEEP_CAPS_LOCKED
) ? " keep-caps-locked" : "",
906 (c
->secure_bits
& SECURE_NO_SETUID_FIXUP
) ? " no-setuid-fixup" : "",
907 (c
->secure_bits
& SECURE_NO_SETUID_FIXUP_LOCKED
) ? " no-setuid-fixup-locked" : "",
908 (c
->secure_bits
& SECURE_NOROOT
) ? " noroot" : "",
909 (c
->secure_bits
& SECURE_NOROOT_LOCKED
) ? "noroot-locked" : "");
911 if (c
->capability_bounding_set_drop
) {
912 fprintf(f
, "%sCapabilityBoundingSetDrop:", prefix
);
914 for (i
= 0; i
<= CAP_LAST_CAP
; i
++)
915 if (c
->capability_bounding_set_drop
& (1 << i
)) {
918 if ((t
= cap_to_name(i
))) {
919 fprintf(f
, " %s", t
);
928 fprintf(f
, "%sUser: %s", prefix
, c
->user
);
930 fprintf(f
, "%sGroup: %s", prefix
, c
->group
);
932 if (c
->supplementary_groups
) {
935 fprintf(f
, "%sSupplementaryGroups:", prefix
);
937 STRV_FOREACH(g
, c
->supplementary_groups
)
938 fprintf(f
, " %s", *g
);
944 void exec_status_fill(ExecStatus
*s
, pid_t pid
, int code
, int status
) {
950 s
->timestamp
= now(CLOCK_REALTIME
);
953 char *exec_command_line(ExecCommand
*c
) {
962 STRV_FOREACH(a
, c
->argv
)
965 if (!(n
= new(char, k
)))
969 STRV_FOREACH(a
, c
->argv
) {
976 if (strpbrk(*a
, WHITESPACE
)) {
987 /* FIXME: this doesn't really handle arguments that have
988 * spaces and ticks in them */
993 void exec_command_dump(ExecCommand
*c
, FILE *f
, const char *prefix
) {
1002 cmd
= exec_command_line(c
);
1005 "%sCommand Line: %s\n",
1006 prefix
, cmd
? cmd
: strerror(ENOMEM
));
1011 void exec_command_dump_list(ExecCommand
*c
, FILE *f
, const char *prefix
) {
1017 LIST_FOREACH(command
, c
, c
)
1018 exec_command_dump(c
, f
, prefix
);
1021 void exec_command_append_list(ExecCommand
**l
, ExecCommand
*e
) {
1028 /* It's kinda important that we keep the order here */
1029 LIST_FIND_TAIL(ExecCommand
, command
, *l
, end
);
1030 LIST_INSERT_AFTER(ExecCommand
, command
, *l
, end
, e
);
1035 static const char* const exec_output_table
[_EXEC_OUTPUT_MAX
] = {
1036 [EXEC_OUTPUT_CONSOLE
] = "console",
1037 [EXEC_OUTPUT_NULL
] = "null",
1038 [EXEC_OUTPUT_SYSLOG
] = "syslog",
1039 [EXEC_OUTPUT_KERNEL
] = "kernel"
1042 DEFINE_STRING_TABLE_LOOKUP(exec_output
, ExecOutput
);
1044 static const char* const exec_input_table
[_EXEC_INPUT_MAX
] = {
1045 [EXEC_INPUT_NULL
] = "null",
1046 [EXEC_INPUT_CONSOLE
] = "console"
1049 DEFINE_STRING_TABLE_LOOKUP(exec_input
, ExecInput
);