1 /* SPDX-License-Identifier: LGPL-2.1+ */
7 #include <sys/capability.h>
11 #include "alloc-util.h"
12 #include "capability-util.h"
16 #include "missing_prctl.h"
17 #include "parse-util.h"
18 #include "user-util.h"
21 int have_effective_cap(int value
) {
22 _cleanup_cap_free_ cap_t cap
;
29 if (cap_get_flag(cap
, value
, CAP_EFFECTIVE
, &fv
) < 0)
35 unsigned long cap_last_cap(void) {
36 static thread_local
unsigned long saved
;
37 static thread_local
bool valid
= false;
38 _cleanup_free_
char *content
= NULL
;
45 /* available since linux-3.2 */
46 r
= read_one_line_file("/proc/sys/kernel/cap_last_cap", &content
);
48 r
= safe_atolu(content
, &p
);
51 if (p
> 63) /* Safety for the future: if one day the kernel learns more than 64 caps,
52 * then we are in trouble (since we, as much userspace and kernel space
53 * store capability masks in uint64_t types). Let's hence protect
54 * ourselves against that and always cap at 63 for now. */
63 /* fall back to syscall-probing for pre linux-3.2 */
64 p
= MIN((unsigned long) CAP_LAST_CAP
, 63U);
66 if (prctl(PR_CAPBSET_READ
, p
) < 0) {
68 /* Hmm, look downwards, until we find one that works */
69 for (p
--; p
> 0; p
--)
70 if (prctl(PR_CAPBSET_READ
, p
) >= 0)
75 /* Hmm, look upwards, until we find one that doesn't work */
77 if (prctl(PR_CAPBSET_READ
, p
+1) < 0)
87 int capability_update_inherited_set(cap_t caps
, uint64_t set
) {
90 /* Add capabilities in the set to the inherited caps. Do not apply
93 for (i
= 0; i
< cap_last_cap(); i
++) {
95 if (set
& (UINT64_C(1) << i
)) {
100 /* Make the capability inheritable. */
101 if (cap_set_flag(caps
, CAP_INHERITABLE
, 1, &v
, CAP_SET
) < 0)
109 int capability_ambient_set_apply(uint64_t set
, bool also_inherit
) {
110 _cleanup_cap_free_ cap_t caps
= NULL
;
114 /* Add the capabilities to the ambient set. */
117 caps
= cap_get_proc();
121 r
= capability_update_inherited_set(caps
, set
);
125 if (cap_set_proc(caps
) < 0)
129 for (i
= 0; i
< cap_last_cap(); i
++) {
131 if (set
& (UINT64_C(1) << i
)) {
133 /* Add the capability to the ambient set. */
134 if (prctl(PR_CAP_AMBIENT
, PR_CAP_AMBIENT_RAISE
, i
, 0, 0) < 0)
142 int capability_bounding_set_drop(uint64_t keep
, bool right_now
) {
143 _cleanup_cap_free_ cap_t before_cap
= NULL
, after_cap
= NULL
;
148 /* If we are run as PID 1 we will lack CAP_SETPCAP by default
149 * in the effective set (yes, the kernel drops that when
150 * executing init!), so get it back temporarily so that we can
151 * call PR_CAPBSET_DROP. */
153 before_cap
= cap_get_proc();
157 if (cap_get_flag(before_cap
, CAP_SETPCAP
, CAP_EFFECTIVE
, &fv
) < 0)
161 _cleanup_cap_free_ cap_t temp_cap
= NULL
;
162 static const cap_value_t v
= CAP_SETPCAP
;
164 temp_cap
= cap_dup(before_cap
);
168 if (cap_set_flag(temp_cap
, CAP_EFFECTIVE
, 1, &v
, CAP_SET
) < 0)
171 if (cap_set_proc(temp_cap
) < 0)
172 log_debug_errno(errno
, "Can't acquire effective CAP_SETPCAP bit, ignoring: %m");
174 /* If we didn't manage to acquire the CAP_SETPCAP bit, we continue anyway, after all this just means
175 * we'll fail later, when we actually intend to drop some capabilities. */
178 after_cap
= cap_dup(before_cap
);
182 for (i
= 0; i
<= cap_last_cap(); i
++) {
185 if ((keep
& (UINT64_C(1) << i
)))
188 /* Drop it from the bounding set */
189 if (prctl(PR_CAPBSET_DROP
, i
) < 0) {
192 /* If dropping the capability failed, let's see if we didn't have it in the first place. If so,
193 * continue anyway, as dropping a capability we didn't have in the first place doesn't really
195 if (prctl(PR_CAPBSET_READ
, i
) != 0)
200 /* Also drop it from the inheritable set, so
201 * that anything we exec() loses the
202 * capability for good. */
203 if (cap_set_flag(after_cap
, CAP_INHERITABLE
, 1, &v
, CAP_CLEAR
) < 0) {
208 /* If we shall apply this right now drop it
209 * also from our own capability sets. */
211 if (cap_set_flag(after_cap
, CAP_PERMITTED
, 1, &v
, CAP_CLEAR
) < 0 ||
212 cap_set_flag(after_cap
, CAP_EFFECTIVE
, 1, &v
, CAP_CLEAR
) < 0) {
222 if (cap_set_proc(after_cap
) < 0) {
223 /* If there are no actual changes anyway then let's ignore this error. */
224 if (cap_compare(before_cap
, after_cap
) != 0)
231 static int drop_from_file(const char *fn
, uint64_t keep
) {
232 _cleanup_free_
char *p
= NULL
;
233 uint64_t current
, after
;
237 r
= read_one_line_file(fn
, &p
);
241 k
= sscanf(p
, "%" PRIu32
" %" PRIu32
, &lo
, &hi
);
245 current
= (uint64_t) lo
| ((uint64_t) hi
<< 32);
246 after
= current
& keep
;
248 if (current
== after
)
251 lo
= after
& UINT32_C(0xFFFFFFFF);
252 hi
= (after
>> 32) & UINT32_C(0xFFFFFFFF);
254 return write_string_filef(fn
, 0, "%" PRIu32
" %" PRIu32
, lo
, hi
);
257 int capability_bounding_set_drop_usermode(uint64_t keep
) {
260 r
= drop_from_file("/proc/sys/kernel/usermodehelper/inheritable", keep
);
264 r
= drop_from_file("/proc/sys/kernel/usermodehelper/bset", keep
);
271 int drop_privileges(uid_t uid
, gid_t gid
, uint64_t keep_capabilities
) {
274 /* Unfortunately we cannot leave privilege dropping to PID 1 here, since we want to run as user but
275 * want to keep some capabilities. Since file capabilities have been introduced this cannot be done
276 * across exec() anymore, unless our binary has the capability configured in the file system, which
277 * we want to avoid. */
279 if (setresgid(gid
, gid
, gid
) < 0)
280 return log_error_errno(errno
, "Failed to change group ID: %m");
282 r
= maybe_setgroups(0, NULL
);
284 return log_error_errno(r
, "Failed to drop auxiliary groups list: %m");
286 /* Ensure we keep the permitted caps across the setresuid(). Note that we do this even if we actually
287 * don't want to keep any capabilities, since we want to be able to drop them from the bounding set
288 * too, and we can only do that if we have capabilities. */
289 if (prctl(PR_SET_KEEPCAPS
, 1) < 0)
290 return log_error_errno(errno
, "Failed to enable keep capabilities flag: %m");
292 if (setresuid(uid
, uid
, uid
) < 0)
293 return log_error_errno(errno
, "Failed to change user ID: %m");
295 if (prctl(PR_SET_KEEPCAPS
, 0) < 0)
296 return log_error_errno(errno
, "Failed to disable keep capabilities flag: %m");
298 /* Drop all caps from the bounding set (as well as the inheritable/permitted/effective sets), except
299 * the ones we want to keep */
300 r
= capability_bounding_set_drop(keep_capabilities
, true);
302 return log_error_errno(r
, "Failed to drop capabilities: %m");
304 /* Now upgrade the permitted caps we still kept to effective caps */
305 if (keep_capabilities
!= 0) {
306 cap_value_t bits
[u64log2(keep_capabilities
) + 1];
307 _cleanup_cap_free_ cap_t d
= NULL
;
314 for (i
= 0; i
< ELEMENTSOF(bits
); i
++)
315 if (keep_capabilities
& (1ULL << i
))
318 /* use enough bits */
319 assert(i
== 64 || (keep_capabilities
>> i
) == 0);
320 /* don't use too many bits */
321 assert(keep_capabilities
& (UINT64_C(1) << (i
- 1)));
323 if (cap_set_flag(d
, CAP_EFFECTIVE
, j
, bits
, CAP_SET
) < 0 ||
324 cap_set_flag(d
, CAP_PERMITTED
, j
, bits
, CAP_SET
) < 0)
325 return log_error_errno(errno
, "Failed to enable capabilities bits: %m");
327 if (cap_set_proc(d
) < 0)
328 return log_error_errno(errno
, "Failed to increase capabilities: %m");
334 int drop_capability(cap_value_t cv
) {
335 _cleanup_cap_free_ cap_t tmp_cap
= NULL
;
337 tmp_cap
= cap_get_proc();
341 if ((cap_set_flag(tmp_cap
, CAP_INHERITABLE
, 1, &cv
, CAP_CLEAR
) < 0) ||
342 (cap_set_flag(tmp_cap
, CAP_PERMITTED
, 1, &cv
, CAP_CLEAR
) < 0) ||
343 (cap_set_flag(tmp_cap
, CAP_EFFECTIVE
, 1, &cv
, CAP_CLEAR
) < 0))
346 if (cap_set_proc(tmp_cap
) < 0)
352 bool ambient_capabilities_supported(void) {
353 static int cache
= -1;
358 /* If PR_CAP_AMBIENT returns something valid, or an unexpected error code we assume that ambient caps are
361 cache
= prctl(PR_CAP_AMBIENT
, PR_CAP_AMBIENT_IS_SET
, CAP_KILL
, 0, 0) >= 0 ||
362 !IN_SET(errno
, EINVAL
, EOPNOTSUPP
, ENOSYS
);
367 int capability_quintet_enforce(const CapabilityQuintet
*q
) {
368 _cleanup_cap_free_ cap_t c
= NULL
, modified
= NULL
;
371 if (q
->ambient
!= (uint64_t) -1) {
373 bool changed
= false;
379 /* In order to raise the ambient caps set we first need to raise the matching inheritable + permitted
381 for (i
= 0; i
<= cap_last_cap(); i
++) {
382 uint64_t m
= UINT64_C(1) << i
;
383 cap_value_t cv
= (cap_value_t
) i
;
384 cap_flag_value_t old_value_inheritable
, old_value_permitted
;
386 if ((q
->ambient
& m
) == 0)
389 if (cap_get_flag(c
, cv
, CAP_INHERITABLE
, &old_value_inheritable
) < 0)
391 if (cap_get_flag(c
, cv
, CAP_PERMITTED
, &old_value_permitted
) < 0)
394 if (old_value_inheritable
== CAP_SET
&& old_value_permitted
== CAP_SET
)
397 if (cap_set_flag(c
, CAP_INHERITABLE
, 1, &cv
, CAP_SET
) < 0)
399 if (cap_set_flag(c
, CAP_PERMITTED
, 1, &cv
, CAP_SET
) < 0)
406 if (cap_set_proc(c
) < 0)
409 r
= capability_ambient_set_apply(q
->ambient
, false);
414 if (q
->inheritable
!= (uint64_t) -1 || q
->permitted
!= (uint64_t) -1 || q
->effective
!= (uint64_t) -1) {
415 bool changed
= false;
424 for (i
= 0; i
<= cap_last_cap(); i
++) {
425 uint64_t m
= UINT64_C(1) << i
;
426 cap_value_t cv
= (cap_value_t
) i
;
428 if (q
->inheritable
!= (uint64_t) -1) {
429 cap_flag_value_t old_value
, new_value
;
431 if (cap_get_flag(c
, cv
, CAP_INHERITABLE
, &old_value
) < 0) {
432 if (errno
== EINVAL
) /* If the kernel knows more caps than this
433 * version of libcap, then this will return
434 * EINVAL. In that case, simply ignore it,
435 * pretend it doesn't exist. */
441 new_value
= (q
->inheritable
& m
) ? CAP_SET
: CAP_CLEAR
;
443 if (old_value
!= new_value
) {
446 if (cap_set_flag(c
, CAP_INHERITABLE
, 1, &cv
, new_value
) < 0)
451 if (q
->permitted
!= (uint64_t) -1) {
452 cap_flag_value_t old_value
, new_value
;
454 if (cap_get_flag(c
, cv
, CAP_PERMITTED
, &old_value
) < 0) {
461 new_value
= (q
->permitted
& m
) ? CAP_SET
: CAP_CLEAR
;
463 if (old_value
!= new_value
) {
466 if (cap_set_flag(c
, CAP_PERMITTED
, 1, &cv
, new_value
) < 0)
471 if (q
->effective
!= (uint64_t) -1) {
472 cap_flag_value_t old_value
, new_value
;
474 if (cap_get_flag(c
, cv
, CAP_EFFECTIVE
, &old_value
) < 0) {
481 new_value
= (q
->effective
& m
) ? CAP_SET
: CAP_CLEAR
;
483 if (old_value
!= new_value
) {
486 if (cap_set_flag(c
, CAP_EFFECTIVE
, 1, &cv
, new_value
) < 0)
493 /* In order to change the bounding caps, we need to keep CAP_SETPCAP for a bit
494 * longer. Let's add it to our list hence for now. */
495 if (q
->bounding
!= (uint64_t) -1) {
496 cap_value_t cv
= CAP_SETPCAP
;
498 modified
= cap_dup(c
);
502 if (cap_set_flag(modified
, CAP_PERMITTED
, 1, &cv
, CAP_SET
) < 0)
504 if (cap_set_flag(modified
, CAP_EFFECTIVE
, 1, &cv
, CAP_SET
) < 0)
507 if (cap_compare(modified
, c
) == 0) {
508 /* No change? then drop this nonsense again */
514 /* Now, let's enforce the caps for the first time. Note that this is where we acquire
515 * caps in any of the sets we currently don't have. We have to do this before
516 * dropping the bounding caps below, since at that point we can never acquire new
517 * caps in inherited/permitted/effective anymore, but only lose them. */
518 if (cap_set_proc(modified
?: c
) < 0)
523 if (q
->bounding
!= (uint64_t) -1) {
524 r
= capability_bounding_set_drop(q
->bounding
, false);
529 /* If needed, let's now set the caps again, this time in the final version, which differs from what
530 * we have already set only in the CAP_SETPCAP bit, which we needed for dropping the bounding
531 * bits. This call only undoes bits and doesn't acquire any which means the bounding caps don't
534 if (cap_set_proc(c
) < 0)