1 /* SPDX-License-Identifier: LGPL-2.1+ */
9 #include "alloc-util.h"
10 #include "capability-util.h"
15 #include "missing_prctl.h"
16 #include "parse-util.h"
17 #include "user-util.h"
20 int have_effective_cap(int value
) {
21 _cleanup_cap_free_ cap_t cap
;
28 if (cap_get_flag(cap
, value
, CAP_EFFECTIVE
, &fv
) < 0)
34 unsigned long cap_last_cap(void) {
35 static thread_local
unsigned long saved
;
36 static thread_local
bool valid
= false;
37 _cleanup_free_
char *content
= NULL
;
44 /* available since linux-3.2 */
45 r
= read_one_line_file("/proc/sys/kernel/cap_last_cap", &content
);
47 r
= safe_atolu(content
, &p
);
50 if (p
> 63) /* Safety for the future: if one day the kernel learns more than 64 caps,
51 * then we are in trouble (since we, as much userspace and kernel space
52 * store capability masks in uint64_t types). Let's hence protect
53 * ourselves against that and always cap at 63 for now. */
62 /* fall back to syscall-probing for pre linux-3.2 */
63 p
= MIN((unsigned long) CAP_LAST_CAP
, 63U);
65 if (prctl(PR_CAPBSET_READ
, p
) < 0) {
67 /* Hmm, look downwards, until we find one that works */
68 for (p
--; p
> 0; p
--)
69 if (prctl(PR_CAPBSET_READ
, p
) >= 0)
74 /* Hmm, look upwards, until we find one that doesn't work */
76 if (prctl(PR_CAPBSET_READ
, p
+1) < 0)
86 int capability_update_inherited_set(cap_t caps
, uint64_t set
) {
89 /* Add capabilities in the set to the inherited caps, drops capabilities not in the set.
90 * Do not apply them yet. */
92 for (i
= 0; i
<= cap_last_cap(); i
++) {
93 cap_flag_value_t flag
= set
& (UINT64_C(1) << i
) ? CAP_SET
: CAP_CLEAR
;
98 if (cap_set_flag(caps
, CAP_INHERITABLE
, 1, &v
, flag
) < 0)
105 int capability_ambient_set_apply(uint64_t set
, bool also_inherit
) {
106 _cleanup_cap_free_ cap_t caps
= NULL
;
110 /* Check that we can use PR_CAP_AMBIENT or quit early. */
111 if (!ambient_capabilities_supported())
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)
138 /* Drop the capability so we don't inherit capabilities we didn't ask for. */
139 r
= prctl(PR_CAP_AMBIENT
, PR_CAP_AMBIENT_IS_SET
, i
, 0, 0);
144 if (prctl(PR_CAP_AMBIENT
, PR_CAP_AMBIENT_LOWER
, i
, 0, 0) < 0)
153 int capability_bounding_set_drop(uint64_t keep
, bool right_now
) {
154 _cleanup_cap_free_ cap_t before_cap
= NULL
, after_cap
= NULL
;
159 /* If we are run as PID 1 we will lack CAP_SETPCAP by default
160 * in the effective set (yes, the kernel drops that when
161 * executing init!), so get it back temporarily so that we can
162 * call PR_CAPBSET_DROP. */
164 before_cap
= cap_get_proc();
168 if (cap_get_flag(before_cap
, CAP_SETPCAP
, CAP_EFFECTIVE
, &fv
) < 0)
172 _cleanup_cap_free_ cap_t temp_cap
= NULL
;
173 static const cap_value_t v
= CAP_SETPCAP
;
175 temp_cap
= cap_dup(before_cap
);
179 if (cap_set_flag(temp_cap
, CAP_EFFECTIVE
, 1, &v
, CAP_SET
) < 0)
182 if (cap_set_proc(temp_cap
) < 0)
183 log_debug_errno(errno
, "Can't acquire effective CAP_SETPCAP bit, ignoring: %m");
185 /* If we didn't manage to acquire the CAP_SETPCAP bit, we continue anyway, after all this just means
186 * we'll fail later, when we actually intend to drop some capabilities. */
189 after_cap
= cap_dup(before_cap
);
193 for (i
= 0; i
<= cap_last_cap(); i
++) {
196 if ((keep
& (UINT64_C(1) << i
)))
199 /* Drop it from the bounding set */
200 if (prctl(PR_CAPBSET_DROP
, i
) < 0) {
203 /* If dropping the capability failed, let's see if we didn't have it in the first place. If so,
204 * continue anyway, as dropping a capability we didn't have in the first place doesn't really
206 if (prctl(PR_CAPBSET_READ
, i
) != 0)
211 /* Also drop it from the inheritable set, so
212 * that anything we exec() loses the
213 * capability for good. */
214 if (cap_set_flag(after_cap
, CAP_INHERITABLE
, 1, &v
, CAP_CLEAR
) < 0) {
219 /* If we shall apply this right now drop it
220 * also from our own capability sets. */
222 if (cap_set_flag(after_cap
, CAP_PERMITTED
, 1, &v
, CAP_CLEAR
) < 0 ||
223 cap_set_flag(after_cap
, CAP_EFFECTIVE
, 1, &v
, CAP_CLEAR
) < 0) {
233 if (cap_set_proc(after_cap
) < 0) {
234 /* If there are no actual changes anyway then let's ignore this error. */
235 if (cap_compare(before_cap
, after_cap
) != 0)
242 static int drop_from_file(const char *fn
, uint64_t keep
) {
243 _cleanup_free_
char *p
= NULL
;
244 uint64_t current
, after
;
248 r
= read_one_line_file(fn
, &p
);
252 k
= sscanf(p
, "%" PRIu32
" %" PRIu32
, &lo
, &hi
);
256 current
= (uint64_t) lo
| ((uint64_t) hi
<< 32);
257 after
= current
& keep
;
259 if (current
== after
)
262 lo
= after
& UINT32_C(0xFFFFFFFF);
263 hi
= (after
>> 32) & UINT32_C(0xFFFFFFFF);
265 return write_string_filef(fn
, 0, "%" PRIu32
" %" PRIu32
, lo
, hi
);
268 int capability_bounding_set_drop_usermode(uint64_t keep
) {
271 r
= drop_from_file("/proc/sys/kernel/usermodehelper/inheritable", keep
);
275 r
= drop_from_file("/proc/sys/kernel/usermodehelper/bset", keep
);
282 int drop_privileges(uid_t uid
, gid_t gid
, uint64_t keep_capabilities
) {
285 /* Unfortunately we cannot leave privilege dropping to PID 1 here, since we want to run as user but
286 * want to keep some capabilities. Since file capabilities have been introduced this cannot be done
287 * across exec() anymore, unless our binary has the capability configured in the file system, which
288 * we want to avoid. */
290 if (setresgid(gid
, gid
, gid
) < 0)
291 return log_error_errno(errno
, "Failed to change group ID: %m");
293 r
= maybe_setgroups(0, NULL
);
295 return log_error_errno(r
, "Failed to drop auxiliary groups list: %m");
297 /* Ensure we keep the permitted caps across the setresuid(). Note that we do this even if we actually
298 * don't want to keep any capabilities, since we want to be able to drop them from the bounding set
299 * too, and we can only do that if we have capabilities. */
300 if (prctl(PR_SET_KEEPCAPS
, 1) < 0)
301 return log_error_errno(errno
, "Failed to enable keep capabilities flag: %m");
303 if (setresuid(uid
, uid
, uid
) < 0)
304 return log_error_errno(errno
, "Failed to change user ID: %m");
306 if (prctl(PR_SET_KEEPCAPS
, 0) < 0)
307 return log_error_errno(errno
, "Failed to disable keep capabilities flag: %m");
309 /* Drop all caps from the bounding set (as well as the inheritable/permitted/effective sets), except
310 * the ones we want to keep */
311 r
= capability_bounding_set_drop(keep_capabilities
, true);
313 return log_error_errno(r
, "Failed to drop capabilities: %m");
315 /* Now upgrade the permitted caps we still kept to effective caps */
316 if (keep_capabilities
!= 0) {
317 cap_value_t bits
[u64log2(keep_capabilities
) + 1];
318 _cleanup_cap_free_ cap_t d
= NULL
;
325 for (i
= 0; i
< ELEMENTSOF(bits
); i
++)
326 if (keep_capabilities
& (1ULL << i
))
329 /* use enough bits */
330 assert(i
== 64 || (keep_capabilities
>> i
) == 0);
331 /* don't use too many bits */
332 assert(keep_capabilities
& (UINT64_C(1) << (i
- 1)));
334 if (cap_set_flag(d
, CAP_EFFECTIVE
, j
, bits
, CAP_SET
) < 0 ||
335 cap_set_flag(d
, CAP_PERMITTED
, j
, bits
, CAP_SET
) < 0)
336 return log_error_errno(errno
, "Failed to enable capabilities bits: %m");
338 if (cap_set_proc(d
) < 0)
339 return log_error_errno(errno
, "Failed to increase capabilities: %m");
345 int drop_capability(cap_value_t cv
) {
346 _cleanup_cap_free_ cap_t tmp_cap
= NULL
;
348 tmp_cap
= cap_get_proc();
352 if ((cap_set_flag(tmp_cap
, CAP_INHERITABLE
, 1, &cv
, CAP_CLEAR
) < 0) ||
353 (cap_set_flag(tmp_cap
, CAP_PERMITTED
, 1, &cv
, CAP_CLEAR
) < 0) ||
354 (cap_set_flag(tmp_cap
, CAP_EFFECTIVE
, 1, &cv
, CAP_CLEAR
) < 0))
357 if (cap_set_proc(tmp_cap
) < 0)
363 bool ambient_capabilities_supported(void) {
364 static int cache
= -1;
369 /* If PR_CAP_AMBIENT returns something valid, or an unexpected error code we assume that ambient caps are
372 cache
= prctl(PR_CAP_AMBIENT
, PR_CAP_AMBIENT_IS_SET
, CAP_KILL
, 0, 0) >= 0 ||
373 !IN_SET(errno
, EINVAL
, EOPNOTSUPP
, ENOSYS
);
378 bool capability_quintet_mangle(CapabilityQuintet
*q
) {
380 uint64_t combined
, drop
= 0;
381 bool ambient_supported
;
385 combined
= q
->effective
| q
->bounding
| q
->inheritable
| q
->permitted
;
387 ambient_supported
= q
->ambient
!= (uint64_t) -1;
388 if (ambient_supported
)
389 combined
|= q
->ambient
;
391 for (i
= 0; i
<= cap_last_cap(); i
++) {
392 unsigned long bit
= UINT64_C(1) << i
;
393 if (!FLAGS_SET(combined
, bit
))
396 if (prctl(PR_CAPBSET_READ
, i
) > 0)
401 log_debug("Not in the current bounding set: %s", capability_to_name(i
));
404 q
->effective
&= ~drop
;
405 q
->bounding
&= ~drop
;
406 q
->inheritable
&= ~drop
;
407 q
->permitted
&= ~drop
;
409 if (ambient_supported
)
412 return drop
!= 0; /* Let the caller know we changed something */
415 int capability_quintet_enforce(const CapabilityQuintet
*q
) {
416 _cleanup_cap_free_ cap_t c
= NULL
, modified
= NULL
;
419 if (q
->ambient
!= (uint64_t) -1) {
421 bool changed
= false;
427 /* In order to raise the ambient caps set we first need to raise the matching inheritable + permitted
429 for (i
= 0; i
<= cap_last_cap(); i
++) {
430 uint64_t m
= UINT64_C(1) << i
;
431 cap_value_t cv
= (cap_value_t
) i
;
432 cap_flag_value_t old_value_inheritable
, old_value_permitted
;
434 if ((q
->ambient
& m
) == 0)
437 if (cap_get_flag(c
, cv
, CAP_INHERITABLE
, &old_value_inheritable
) < 0)
439 if (cap_get_flag(c
, cv
, CAP_PERMITTED
, &old_value_permitted
) < 0)
442 if (old_value_inheritable
== CAP_SET
&& old_value_permitted
== CAP_SET
)
445 if (cap_set_flag(c
, CAP_INHERITABLE
, 1, &cv
, CAP_SET
) < 0)
447 if (cap_set_flag(c
, CAP_PERMITTED
, 1, &cv
, CAP_SET
) < 0)
454 if (cap_set_proc(c
) < 0)
457 r
= capability_ambient_set_apply(q
->ambient
, false);
462 if (q
->inheritable
!= (uint64_t) -1 || q
->permitted
!= (uint64_t) -1 || q
->effective
!= (uint64_t) -1) {
463 bool changed
= false;
472 for (i
= 0; i
<= cap_last_cap(); i
++) {
473 uint64_t m
= UINT64_C(1) << i
;
474 cap_value_t cv
= (cap_value_t
) i
;
476 if (q
->inheritable
!= (uint64_t) -1) {
477 cap_flag_value_t old_value
, new_value
;
479 if (cap_get_flag(c
, cv
, CAP_INHERITABLE
, &old_value
) < 0) {
480 if (errno
== EINVAL
) /* If the kernel knows more caps than this
481 * version of libcap, then this will return
482 * EINVAL. In that case, simply ignore it,
483 * pretend it doesn't exist. */
489 new_value
= (q
->inheritable
& m
) ? CAP_SET
: CAP_CLEAR
;
491 if (old_value
!= new_value
) {
494 if (cap_set_flag(c
, CAP_INHERITABLE
, 1, &cv
, new_value
) < 0)
499 if (q
->permitted
!= (uint64_t) -1) {
500 cap_flag_value_t old_value
, new_value
;
502 if (cap_get_flag(c
, cv
, CAP_PERMITTED
, &old_value
) < 0) {
509 new_value
= (q
->permitted
& m
) ? CAP_SET
: CAP_CLEAR
;
511 if (old_value
!= new_value
) {
514 if (cap_set_flag(c
, CAP_PERMITTED
, 1, &cv
, new_value
) < 0)
519 if (q
->effective
!= (uint64_t) -1) {
520 cap_flag_value_t old_value
, new_value
;
522 if (cap_get_flag(c
, cv
, CAP_EFFECTIVE
, &old_value
) < 0) {
529 new_value
= (q
->effective
& m
) ? CAP_SET
: CAP_CLEAR
;
531 if (old_value
!= new_value
) {
534 if (cap_set_flag(c
, CAP_EFFECTIVE
, 1, &cv
, new_value
) < 0)
541 /* In order to change the bounding caps, we need to keep CAP_SETPCAP for a bit
542 * longer. Let's add it to our list hence for now. */
543 if (q
->bounding
!= (uint64_t) -1) {
544 cap_value_t cv
= CAP_SETPCAP
;
546 modified
= cap_dup(c
);
550 if (cap_set_flag(modified
, CAP_PERMITTED
, 1, &cv
, CAP_SET
) < 0)
552 if (cap_set_flag(modified
, CAP_EFFECTIVE
, 1, &cv
, CAP_SET
) < 0)
555 if (cap_compare(modified
, c
) == 0) {
556 /* No change? then drop this nonsense again */
562 /* Now, let's enforce the caps for the first time. Note that this is where we acquire
563 * caps in any of the sets we currently don't have. We have to do this before
564 * dropping the bounding caps below, since at that point we can never acquire new
565 * caps in inherited/permitted/effective anymore, but only lose them. */
566 if (cap_set_proc(modified
?: c
) < 0)
571 if (q
->bounding
!= (uint64_t) -1) {
572 r
= capability_bounding_set_drop(q
->bounding
, false);
577 /* If needed, let's now set the caps again, this time in the final version, which differs from what
578 * we have already set only in the CAP_SETPCAP bit, which we needed for dropping the bounding
579 * bits. This call only undoes bits and doesn't acquire any which means the bounding caps don't
582 if (cap_set_proc(c
) < 0)