[thirdparty/systemd.git] / src / basic / capability-util.c

/* SPDX-License-Identifier: LGPL-2.1+ */

#include <errno.h>
#include <stdio.h>
#include <stdlib.h>
#include <sys/prctl.h>
#include <unistd.h>

#include "alloc-util.h"
#include "capability-util.h"
#include "cap-list.h"
#include "fileio.h"
#include "log.h"
#include "macro.h"
#include "missing_prctl.h"
#include "parse-util.h"
#include "user-util.h"
#include "util.h"

int have_effective_cap(int value) {
        _cleanup_cap_free_ cap_t cap;
        cap_flag_value_t fv;

        cap = cap_get_proc();
        if (!cap)
                return -errno;

        if (cap_get_flag(cap, value, CAP_EFFECTIVE, &fv) < 0)
                return -errno;

        return fv == CAP_SET;
}

unsigned cap_last_cap(void) {
        static thread_local unsigned saved;
        static thread_local bool valid = false;
        _cleanup_free_ char *content = NULL;
        unsigned long p = 0;
        int r;

        if (valid)
                return saved;

        /* available since linux-3.2 */
        r = read_one_line_file("/proc/sys/kernel/cap_last_cap", &content);
        if (r >= 0) {
                r = safe_atolu(content, &p);
                if (r >= 0) {

                        if (p > 63) /* Safety for the future: if one day the kernel learns more than 64 caps,
                                     * then we are in trouble (since we, as much userspace and kernel space
                                     * store capability masks in uint64_t types). Let's hence protect
                                     * ourselves against that and always cap at 63 for now. */
                                p = 63;

                        saved = p;
                        valid = true;
                        return p;
                }
        }

        /* fall back to syscall-probing for pre linux-3.2 */
        p = MIN((unsigned long) CAP_LAST_CAP, 63U);

        if (prctl(PR_CAPBSET_READ, p) < 0) {

                /* Hmm, look downwards, until we find one that works */
                for (p--; p > 0; p--)
                        if (prctl(PR_CAPBSET_READ, p) >= 0)
                                break;

        } else {

                /* Hmm, look upwards, until we find one that doesn't work */
                for (; p < 63; p++)
                        if (prctl(PR_CAPBSET_READ, p+1) < 0)
                                break;
        }

        saved = p;
        valid = true;

        return p;
}

int capability_update_inherited_set(cap_t caps, uint64_t set) {
        /* Add capabilities in the set to the inherited caps, drops capabilities not in the set.
         * Do not apply them yet. */

        for (unsigned i = 0; i <= cap_last_cap(); i++) {
                cap_flag_value_t flag = set & (UINT64_C(1) << i) ? CAP_SET : CAP_CLEAR;
                cap_value_t v;

                v = (cap_value_t) i;

                if (cap_set_flag(caps, CAP_INHERITABLE, 1, &v, flag) < 0)
                        return -errno;
        }

        return 0;
}

int capability_ambient_set_apply(uint64_t set, bool also_inherit) {
        _cleanup_cap_free_ cap_t caps = NULL;
        int r;

        /* Remove capabilities requested in ambient set, but not in the bounding set */
        for (unsigned i = 0; i <= cap_last_cap(); i++) {
                if (set == 0)
                        break;

                if (FLAGS_SET(set, (UINT64_C(1) << i)) && prctl(PR_CAPBSET_READ, i) != 1) {
                        log_debug("Ambient capability %s requested but missing from bounding set,"
                                        " suppressing automatically.", capability_to_name(i));
                        set &= ~(UINT64_C(1) << i);
                }
        }

        /* Add the capabilities to the ambient set (an possibly also the inheritable set) */

        /* Check that we can use PR_CAP_AMBIENT or quit early. */
        if (!ambient_capabilities_supported())
                return (set & all_capabilities()) == 0 ?
                        0 : -EOPNOTSUPP; /* if actually no ambient caps are to be set, be silent,
                                          * otherwise fail recognizably */

        if (also_inherit) {
                caps = cap_get_proc();
                if (!caps)
                        return -errno;

                r = capability_update_inherited_set(caps, set);
                if (r < 0)
                        return -errno;

                if (cap_set_proc(caps) < 0)
                        return -errno;
        }

        for (unsigned i = 0; i <= cap_last_cap(); i++) {

                if (set & (UINT64_C(1) << i)) {

                        /* Add the capability to the ambient set. */
                        if (prctl(PR_CAP_AMBIENT, PR_CAP_AMBIENT_RAISE, i, 0, 0) < 0)
                                return -errno;
                } else {

                        /* Drop the capability so we don't inherit capabilities we didn't ask for. */
                        r = prctl(PR_CAP_AMBIENT, PR_CAP_AMBIENT_IS_SET, i, 0, 0);
                        if (r < 0)
                                return -errno;

                        if (r)
                                if (prctl(PR_CAP_AMBIENT, PR_CAP_AMBIENT_LOWER, i, 0, 0) < 0)
                                        return -errno;

                }
        }

        return 0;
}

int capability_bounding_set_drop(uint64_t keep, bool right_now) {
        _cleanup_cap_free_ cap_t before_cap = NULL, after_cap = NULL;
        cap_flag_value_t fv;
        int r;

        /* If we are run as PID 1 we will lack CAP_SETPCAP by default
         * in the effective set (yes, the kernel drops that when
         * executing init!), so get it back temporarily so that we can
         * call PR_CAPBSET_DROP. */

        before_cap = cap_get_proc();
        if (!before_cap)
                return -errno;

        if (cap_get_flag(before_cap, CAP_SETPCAP, CAP_EFFECTIVE, &fv) < 0)
                return -errno;

        if (fv != CAP_SET) {
                _cleanup_cap_free_ cap_t temp_cap = NULL;
                static const cap_value_t v = CAP_SETPCAP;

                temp_cap = cap_dup(before_cap);
                if (!temp_cap)
                        return -errno;

                if (cap_set_flag(temp_cap, CAP_EFFECTIVE, 1, &v, CAP_SET) < 0)
                        return -errno;

                if (cap_set_proc(temp_cap) < 0)
                        log_debug_errno(errno, "Can't acquire effective CAP_SETPCAP bit, ignoring: %m");

                /* If we didn't manage to acquire the CAP_SETPCAP bit, we continue anyway, after all this just means
                 * we'll fail later, when we actually intend to drop some capabilities. */
        }

        after_cap = cap_dup(before_cap);
        if (!after_cap)
                return -errno;

        for (unsigned i = 0; i <= cap_last_cap(); i++) {
                cap_value_t v;

                if ((keep & (UINT64_C(1) << i)))
                        continue;

                /* Drop it from the bounding set */
                if (prctl(PR_CAPBSET_DROP, i) < 0) {
                        r = -errno;

                        /* If dropping the capability failed, let's see if we didn't have it in the first place. If so,
                         * continue anyway, as dropping a capability we didn't have in the first place doesn't really
                         * matter anyway. */
                        if (prctl(PR_CAPBSET_READ, i) != 0)
                                goto finish;
                }
                v = (cap_value_t) i;

                /* Also drop it from the inheritable set, so
                 * that anything we exec() loses the
                 * capability for good. */
                if (cap_set_flag(after_cap, CAP_INHERITABLE, 1, &v, CAP_CLEAR) < 0) {
                        r = -errno;
                        goto finish;
                }

                /* If we shall apply this right now drop it
                 * also from our own capability sets. */
                if (right_now) {
                        if (cap_set_flag(after_cap, CAP_PERMITTED, 1, &v, CAP_CLEAR) < 0 ||
                            cap_set_flag(after_cap, CAP_EFFECTIVE, 1, &v, CAP_CLEAR) < 0) {
                                r = -errno;
                                goto finish;
                        }
                }
        }

        r = 0;

finish:
        if (cap_set_proc(after_cap) < 0) {
                /* If there are no actual changes anyway then let's ignore this error. */
                if (cap_compare(before_cap, after_cap) != 0)
                        r = -errno;
        }

        return r;
}

static int drop_from_file(const char *fn, uint64_t keep) {
        _cleanup_free_ char *p = NULL;
        uint64_t current, after;
        uint32_t hi, lo;
        int r, k;

        r = read_one_line_file(fn, &p);
        if (r < 0)
                return r;

        k = sscanf(p, "%" PRIu32 " %" PRIu32, &lo, &hi);
        if (k != 2)
                return -EIO;

        current = (uint64_t) lo | ((uint64_t) hi << 32);
        after = current & keep;

        if (current == after)
                return 0;

        lo = after & UINT32_C(0xFFFFFFFF);
        hi = (after >> 32) & UINT32_C(0xFFFFFFFF);

        return write_string_filef(fn, 0, "%" PRIu32 " %" PRIu32, lo, hi);
}

int capability_bounding_set_drop_usermode(uint64_t keep) {
        int r;

        r = drop_from_file("/proc/sys/kernel/usermodehelper/inheritable", keep);
        if (r < 0)
                return r;

        r = drop_from_file("/proc/sys/kernel/usermodehelper/bset", keep);
        if (r < 0)
                return r;

        return r;
}

int drop_privileges(uid_t uid, gid_t gid, uint64_t keep_capabilities) {
        int r;

        /* Unfortunately we cannot leave privilege dropping to PID 1 here, since we want to run as user but
         * want to keep some capabilities. Since file capabilities have been introduced this cannot be done
         * across exec() anymore, unless our binary has the capability configured in the file system, which
         * we want to avoid. */

        if (setresgid(gid, gid, gid) < 0)
                return log_error_errno(errno, "Failed to change group ID: %m");

        r = maybe_setgroups(0, NULL);
        if (r < 0)
                return log_error_errno(r, "Failed to drop auxiliary groups list: %m");

        /* Ensure we keep the permitted caps across the setresuid(). Note that we do this even if we actually
         * don't want to keep any capabilities, since we want to be able to drop them from the bounding set
         * too, and we can only do that if we have capabilities. */
        if (prctl(PR_SET_KEEPCAPS, 1) < 0)
                return log_error_errno(errno, "Failed to enable keep capabilities flag: %m");

        if (setresuid(uid, uid, uid) < 0)
                return log_error_errno(errno, "Failed to change user ID: %m");

        if (prctl(PR_SET_KEEPCAPS, 0) < 0)
                return log_error_errno(errno, "Failed to disable keep capabilities flag: %m");

        /* Drop all caps from the bounding set (as well as the inheritable/permitted/effective sets), except
         * the ones we want to keep */
        r = capability_bounding_set_drop(keep_capabilities, true);
        if (r < 0)
                return log_error_errno(r, "Failed to drop capabilities: %m");

        /* Now upgrade the permitted caps we still kept to effective caps */
        if (keep_capabilities != 0) {
                cap_value_t bits[u64log2(keep_capabilities) + 1];
                _cleanup_cap_free_ cap_t d = NULL;
                unsigned i, j = 0;

                d = cap_init();
                if (!d)
                        return log_oom();

                for (i = 0; i < ELEMENTSOF(bits); i++)
                        if (keep_capabilities & (1ULL << i))
                                bits[j++] = i;

                /* use enough bits */
                assert(i == 64 || (keep_capabilities >> i) == 0);
                /* don't use too many bits */
                assert(keep_capabilities & (UINT64_C(1) << (i - 1)));

                if (cap_set_flag(d, CAP_EFFECTIVE, j, bits, CAP_SET) < 0 ||
                    cap_set_flag(d, CAP_PERMITTED, j, bits, CAP_SET) < 0)
                        return log_error_errno(errno, "Failed to enable capabilities bits: %m");

                if (cap_set_proc(d) < 0)
                        return log_error_errno(errno, "Failed to increase capabilities: %m");
        }

        return 0;
}

int drop_capability(cap_value_t cv) {
        _cleanup_cap_free_ cap_t tmp_cap = NULL;

        tmp_cap = cap_get_proc();
        if (!tmp_cap)
                return -errno;

        if ((cap_set_flag(tmp_cap, CAP_INHERITABLE, 1, &cv, CAP_CLEAR) < 0) ||
            (cap_set_flag(tmp_cap, CAP_PERMITTED, 1, &cv, CAP_CLEAR) < 0) ||
            (cap_set_flag(tmp_cap, CAP_EFFECTIVE, 1, &cv, CAP_CLEAR) < 0))
                return -errno;

        if (cap_set_proc(tmp_cap) < 0)
                return -errno;

        return 0;
}

bool ambient_capabilities_supported(void) {
        static int cache = -1;

        if (cache >= 0)
                return cache;

        /* If PR_CAP_AMBIENT returns something valid, or an unexpected error code we assume that ambient caps are
         * available. */

        cache = prctl(PR_CAP_AMBIENT, PR_CAP_AMBIENT_IS_SET, CAP_KILL, 0, 0) >= 0 ||
                !IN_SET(errno, EINVAL, EOPNOTSUPP, ENOSYS);

        return cache;
}

bool capability_quintet_mangle(CapabilityQuintet *q) {
        uint64_t combined, drop = 0;
        bool ambient_supported;

        assert(q);

        combined = q->effective | q->bounding | q->inheritable | q->permitted;

        ambient_supported = q->ambient != (uint64_t) -1;
        if (ambient_supported)
                combined |= q->ambient;

        for (unsigned i = 0; i <= cap_last_cap(); i++) {
                unsigned long bit = UINT64_C(1) << i;
                if (!FLAGS_SET(combined, bit))
                        continue;

                if (prctl(PR_CAPBSET_READ, i) > 0)
                        continue;

                drop |= bit;

                log_debug("Not in the current bounding set: %s", capability_to_name(i));
        }

        q->effective &= ~drop;
        q->bounding &= ~drop;
        q->inheritable &= ~drop;
        q->permitted &= ~drop;

        if (ambient_supported)
                q->ambient &= ~drop;

        return drop != 0; /* Let the caller know we changed something */
}

int capability_quintet_enforce(const CapabilityQuintet *q) {
        _cleanup_cap_free_ cap_t c = NULL, modified = NULL;
        int r;

        if (q->ambient != (uint64_t) -1) {
                bool changed = false;

                c = cap_get_proc();
                if (!c)
                        return -errno;

                /* In order to raise the ambient caps set we first need to raise the matching
                 * inheritable + permitted cap */
                for (unsigned i = 0; i <= cap_last_cap(); i++) {
                        uint64_t m = UINT64_C(1) << i;
                        cap_value_t cv = (cap_value_t) i;
                        cap_flag_value_t old_value_inheritable, old_value_permitted;

                        if ((q->ambient & m) == 0)
                                continue;

                        if (cap_get_flag(c, cv, CAP_INHERITABLE, &old_value_inheritable) < 0)
                                return -errno;
                        if (cap_get_flag(c, cv, CAP_PERMITTED, &old_value_permitted) < 0)
                                return -errno;

                        if (old_value_inheritable == CAP_SET && old_value_permitted == CAP_SET)
                                continue;

                        if (cap_set_flag(c, CAP_INHERITABLE, 1, &cv, CAP_SET) < 0)
                                return -errno;
                        if (cap_set_flag(c, CAP_PERMITTED, 1, &cv, CAP_SET) < 0)
                                return -errno;

                        changed = true;
                }

                if (changed)
                        if (cap_set_proc(c) < 0)
                                return -errno;

                r = capability_ambient_set_apply(q->ambient, false);
                if (r < 0)
                        return r;
        }

        if (q->inheritable != (uint64_t) -1 || q->permitted != (uint64_t) -1 || q->effective != (uint64_t) -1) {
                bool changed = false;

                if (!c) {
                        c = cap_get_proc();
                        if (!c)
                                return -errno;
                }

                for (unsigned i = 0; i <= cap_last_cap(); i++) {
                        uint64_t m = UINT64_C(1) << i;
                        cap_value_t cv = (cap_value_t) i;

                        if (q->inheritable != (uint64_t) -1) {
                                cap_flag_value_t old_value, new_value;

                                if (cap_get_flag(c, cv, CAP_INHERITABLE, &old_value) < 0) {
                                        if (errno == EINVAL) /* If the kernel knows more caps than this
                                                              * version of libcap, then this will return
                                                              * EINVAL. In that case, simply ignore it,
                                                              * pretend it doesn't exist. */
                                                continue;

                                        return -errno;
                                }

                                new_value = (q->inheritable & m) ? CAP_SET : CAP_CLEAR;

                                if (old_value != new_value) {
                                        changed = true;

                                        if (cap_set_flag(c, CAP_INHERITABLE, 1, &cv, new_value) < 0)
                                                return -errno;
                                }
                        }

                        if (q->permitted != (uint64_t) -1) {
                                cap_flag_value_t old_value, new_value;

                                if (cap_get_flag(c, cv, CAP_PERMITTED, &old_value) < 0) {
                                        if (errno == EINVAL)
                                                continue;

                                        return -errno;
                                }

                                new_value = (q->permitted & m) ? CAP_SET : CAP_CLEAR;

                                if (old_value != new_value) {
                                        changed = true;

                                        if (cap_set_flag(c, CAP_PERMITTED, 1, &cv, new_value) < 0)
                                                return -errno;
                                }
                        }

                        if (q->effective != (uint64_t) -1) {
                                cap_flag_value_t old_value, new_value;

                                if (cap_get_flag(c, cv, CAP_EFFECTIVE, &old_value) < 0) {
                                        if (errno == EINVAL)
                                                continue;

                                        return -errno;
                                }

                                new_value = (q->effective & m) ? CAP_SET : CAP_CLEAR;

                                if (old_value != new_value) {
                                        changed = true;

                                        if (cap_set_flag(c, CAP_EFFECTIVE, 1, &cv, new_value) < 0)
                                                return -errno;
                                }
                        }
                }

                if (changed) {
                        /* In order to change the bounding caps, we need to keep CAP_SETPCAP for a bit
                         * longer. Let's add it to our list hence for now. */
                        if (q->bounding != (uint64_t) -1) {
                                cap_value_t cv = CAP_SETPCAP;

                                modified = cap_dup(c);
                                if (!modified)
                                        return -ENOMEM;

                                if (cap_set_flag(modified, CAP_PERMITTED, 1, &cv, CAP_SET) < 0)
                                        return -errno;
                                if (cap_set_flag(modified, CAP_EFFECTIVE, 1, &cv, CAP_SET) < 0)
                                        return -errno;

                                if (cap_compare(modified, c) == 0) {
                                        /* No change? then drop this nonsense again */
                                        cap_free(modified);
                                        modified = NULL;
                                }
                        }

                        /* Now, let's enforce the caps for the first time. Note that this is where we acquire
                         * caps in any of the sets we currently don't have. We have to do this before
                         * dropping the bounding caps below, since at that point we can never acquire new
                         * caps in inherited/permitted/effective anymore, but only lose them. */
                        if (cap_set_proc(modified ?: c) < 0)
                                return -errno;
                }
        }

        if (q->bounding != (uint64_t) -1) {
                r = capability_bounding_set_drop(q->bounding, false);
                if (r < 0)
                        return r;
        }

        /* If needed, let's now set the caps again, this time in the final version, which differs from what
         * we have already set only in the CAP_SETPCAP bit, which we needed for dropping the bounding
         * bits. This call only undoes bits and doesn't acquire any which means the bounding caps don't
         * matter. */
        if (modified)
                if (cap_set_proc(c) < 0)
                        return -errno;

        return 0;
}
Commit	Line	Data
53e1b683	1	/* SPDX-License-Identifier: LGPL-2.1+ */
d7832d2c	2
d7832d2c KS	3	#include <errno.h>
d7832d2c KS	4	#include <stdio.h>
11c3a366	5	#include <stdlib.h>
d7832d2c	6	#include <sys/prctl.h>
3ffd4af2	7	#include <unistd.h>
d7832d2c	8
b5efdb8a	9	#include "alloc-util.h"
430f0182	10	#include "capability-util.h"
f66ad460	11	#include "cap-list.h"
3ffd4af2 LP	12	#include "fileio.h"
3ffd4af2 LP	13	#include "log.h"
d7832d2c	14	#include "macro.h"
a22692d7	15	#include "missing_prctl.h"
6bedfcbb	16	#include "parse-util.h"
36d85478	17	#include "user-util.h"
d7832d2c	18	#include "util.h"
d7832d2c KS	19
d7832d2c KS	20	int have_effective_cap(int value) {
5ce70e5b	21	_cleanup_cap_free_ cap_t cap;
d7832d2c	22	cap_flag_value_t fv;
d7832d2c	23
ec8927ca LP	24	cap = cap_get_proc();
ec8927ca LP	25	if (!cap)
d7832d2c KS	26	return -errno;
	27
	28	if (cap_get_flag(cap, value, CAP_EFFECTIVE, &fv) < 0)
5ce70e5b	29	return -errno;
15ce3f26 LP	30
15ce3f26 LP	31	return fv == CAP_SET;
d7832d2c KS	32	}
d7832d2c KS	33
864a25d9 ZJS	34	unsigned cap_last_cap(void) {
864a25d9 ZJS	35	static thread_local unsigned saved;
ec202eae	36	static thread_local bool valid = false;
80b43783	37	_cleanup_free_ char *content = NULL;
a7f7d1bd	38	unsigned long p = 0;
80b43783	39	int r;
d7832d2c KS	40
	41	if (valid)
	42	return saved;
	43
80b43783 DH	44	/* available since linux-3.2 */
	45	r = read_one_line_file("/proc/sys/kernel/cap_last_cap", &content);
	46	if (r >= 0) {
	47	r = safe_atolu(content, &p);
	48	if (r >= 0) {
5211445e LP	49
	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
d09a2585	52	* store capability masks in uint64_t types). Let's hence protect
5211445e LP	53	* ourselves against that and always cap at 63 for now. */
	54	p = 63;
	55
80b43783 DH	56	saved = p;
	57	valid = true;
	58	return p;
	59	}
	60	}
	61
	62	/* fall back to syscall-probing for pre linux-3.2 */
46eda043	63	p = MIN((unsigned long) CAP_LAST_CAP, 63U);
d7832d2c KS	64
	65	if (prctl(PR_CAPBSET_READ, p) < 0) {
	66
5211445e	67	/* Hmm, look downwards, until we find one that works */
864a25d9	68	for (p--; p > 0; p--)
d7832d2c KS	69	if (prctl(PR_CAPBSET_READ, p) >= 0)
	70	break;
	71
	72	} else {
	73
5211445e LP	74	/* Hmm, look upwards, until we find one that doesn't work */
5211445e LP	75	for (; p < 63; p++)
d7832d2c KS	76	if (prctl(PR_CAPBSET_READ, p+1) < 0)
	77	break;
	78	}
	79
	80	saved = p;
	81	valid = true;
	82
	83	return p;
	84	}
ec8927ca	85
755d4b67	86	int capability_update_inherited_set(cap_t caps, uint64_t set) {
82d832b4 KK	87	/* Add capabilities in the set to the inherited caps, drops capabilities not in the set.
82d832b4 KK	88	* Do not apply them yet. */
755d4b67	89
864a25d9	90	for (unsigned i = 0; i <= cap_last_cap(); i++) {
82d832b4 KK	91	cap_flag_value_t flag = set & (UINT64_C(1) << i) ? CAP_SET : CAP_CLEAR;
82d832b4 KK	92	cap_value_t v;
755d4b67	93
82d832b4	94	v = (cap_value_t) i;
755d4b67	95
82d832b4 KK	96	if (cap_set_flag(caps, CAP_INHERITABLE, 1, &v, flag) < 0)
82d832b4 KK	97	return -errno;
755d4b67 IP	98	}
	99
	100	return 0;
	101	}
	102
	103	int capability_ambient_set_apply(uint64_t set, bool also_inherit) {
755d4b67	104	_cleanup_cap_free_ cap_t caps = NULL;
15ce3f26	105	int r;
755d4b67	106
8acb11a6	107	/* Remove capabilities requested in ambient set, but not in the bounding set */
864a25d9	108	for (unsigned i = 0; i <= cap_last_cap(); i++) {
8acb11a6	109	if (set == 0)
	110	break;
	111
	112	if (FLAGS_SET(set, (UINT64_C(1) << i)) && prctl(PR_CAPBSET_READ, i) != 1) {
	113	log_debug("Ambient capability %s requested but missing from bounding set,"
	114	" suppressing automatically.", capability_to_name(i));
	115	set &= ~(UINT64_C(1) << i);
	116	}
	117	}
	118
c4b23712 LP	119	/* Add the capabilities to the ambient set (an possibly also the inheritable set) */
c4b23712 LP	120
7ea4392f KK	121	/* Check that we can use PR_CAP_AMBIENT or quit early. */
7ea4392f KK	122	if (!ambient_capabilities_supported())
c4b23712 LP	123	return (set & all_capabilities()) == 0 ?
	124	0 : -EOPNOTSUPP; /* if actually no ambient caps are to be set, be silent,
	125	* otherwise fail recognizably */
755d4b67 IP	126
755d4b67 IP	127	if (also_inherit) {
755d4b67 IP	128	caps = cap_get_proc();
	129	if (!caps)
	130	return -errno;
	131
	132	r = capability_update_inherited_set(caps, set);
	133	if (r < 0)
	134	return -errno;
	135
	136	if (cap_set_proc(caps) < 0)
	137	return -errno;
	138	}
	139
864a25d9	140	for (unsigned i = 0; i <= cap_last_cap(); i++) {
755d4b67 IP	141
	142	if (set & (UINT64_C(1) << i)) {
	143
	144	/* Add the capability to the ambient set. */
	145	if (prctl(PR_CAP_AMBIENT, PR_CAP_AMBIENT_RAISE, i, 0, 0) < 0)
	146	return -errno;
82d832b4 KK	147	} else {
	148
	149	/* Drop the capability so we don't inherit capabilities we didn't ask for. */
	150	r = prctl(PR_CAP_AMBIENT, PR_CAP_AMBIENT_IS_SET, i, 0, 0);
	151	if (r < 0)
	152	return -errno;
	153
	154	if (r)
	155	if (prctl(PR_CAP_AMBIENT, PR_CAP_AMBIENT_LOWER, i, 0, 0) < 0)
	156	return -errno;
	157
755d4b67 IP	158	}
	159	}
	160
	161	return 0;
	162	}
	163
a103496c	164	int capability_bounding_set_drop(uint64_t keep, bool right_now) {
6067611a	165	_cleanup_cap_free_ cap_t before_cap = NULL, after_cap = NULL;
ec8927ca LP	166	cap_flag_value_t fv;
	167	int r;
	168
	169	/* If we are run as PID 1 we will lack CAP_SETPCAP by default
	170	* in the effective set (yes, the kernel drops that when
	171	* executing init!), so get it back temporarily so that we can
	172	* call PR_CAPBSET_DROP. */
	173
6067611a LP	174	before_cap = cap_get_proc();
6067611a LP	175	if (!before_cap)
ec8927ca LP	176	return -errno;
ec8927ca LP	177
6067611a	178	if (cap_get_flag(before_cap, CAP_SETPCAP, CAP_EFFECTIVE, &fv) < 0)
ec8927ca	179	return -errno;
ec8927ca LP	180
ec8927ca LP	181	if (fv != CAP_SET) {
6a010ac9	182	_cleanup_cap_free_ cap_t temp_cap = NULL;
ec8927ca LP	183	static const cap_value_t v = CAP_SETPCAP;
ec8927ca LP	184
6067611a LP	185	temp_cap = cap_dup(before_cap);
	186	if (!temp_cap)
	187	return -errno;
ec8927ca	188
6067611a LP	189	if (cap_set_flag(temp_cap, CAP_EFFECTIVE, 1, &v, CAP_SET) < 0)
6067611a LP	190	return -errno;
ec8927ca	191
6067611a LP	192	if (cap_set_proc(temp_cap) < 0)
	193	log_debug_errno(errno, "Can't acquire effective CAP_SETPCAP bit, ignoring: %m");
	194
	195	/* If we didn't manage to acquire the CAP_SETPCAP bit, we continue anyway, after all this just means
	196	* we'll fail later, when we actually intend to drop some capabilities. */
ec8927ca LP	197	}
ec8927ca LP	198
6067611a LP	199	after_cap = cap_dup(before_cap);
	200	if (!after_cap)
	201	return -errno;
	202
864a25d9	203	for (unsigned i = 0; i <= cap_last_cap(); i++) {
6067611a	204	cap_value_t v;
ec8927ca	205
6067611a LP	206	if ((keep & (UINT64_C(1) << i)))
6067611a LP	207	continue;
ec8927ca	208
6067611a LP	209	/* Drop it from the bounding set */
	210	if (prctl(PR_CAPBSET_DROP, i) < 0) {
	211	r = -errno;
	212
	213	/* If dropping the capability failed, let's see if we didn't have it in the first place. If so,
	214	* continue anyway, as dropping a capability we didn't have in the first place doesn't really
	215	* matter anyway. */
	216	if (prctl(PR_CAPBSET_READ, i) != 0)
ec8927ca	217	goto finish;
6067611a LP	218	}
	219	v = (cap_value_t) i;
	220
	221	/* Also drop it from the inheritable set, so
	222	* that anything we exec() loses the
	223	* capability for good. */
	224	if (cap_set_flag(after_cap, CAP_INHERITABLE, 1, &v, CAP_CLEAR) < 0) {
	225	r = -errno;
	226	goto finish;
	227	}
ec8927ca	228
6067611a LP	229	/* If we shall apply this right now drop it
	230	* also from our own capability sets. */
	231	if (right_now) {
	232	if (cap_set_flag(after_cap, CAP_PERMITTED, 1, &v, CAP_CLEAR) < 0 \|\|
	233	cap_set_flag(after_cap, CAP_EFFECTIVE, 1, &v, CAP_CLEAR) < 0) {
ec8927ca LP	234	r = -errno;
	235	goto finish;
	236	}
ec8927ca LP	237	}
	238	}
	239
	240	r = 0;
	241
	242	finish:
6067611a LP	243	if (cap_set_proc(after_cap) < 0) {
	244	/* If there are no actual changes anyway then let's ignore this error. */
	245	if (cap_compare(before_cap, after_cap) != 0)
	246	r = -errno;
	247	}
ec8927ca LP	248
	249	return r;
	250	}
939b8f14	251
a103496c	252	static int drop_from_file(const char *fn, uint64_t keep) {
3130fca5	253	_cleanup_free_ char *p = NULL;
939b8f14	254	uint64_t current, after;
3130fca5 LP	255	uint32_t hi, lo;
3130fca5 LP	256	int r, k;
939b8f14 LP	257
	258	r = read_one_line_file(fn, &p);
	259	if (r < 0)
	260	return r;
	261
7345109e	262	k = sscanf(p, "%" PRIu32 " %" PRIu32, &lo, &hi);
939b8f14 LP	263	if (k != 2)
	264	return -EIO;
	265
7345109e	266	current = (uint64_t) lo \| ((uint64_t) hi << 32);
a103496c	267	after = current & keep;
939b8f14 LP	268
	269	if (current == after)
	270	return 0;
	271
7345109e LP	272	lo = after & UINT32_C(0xFFFFFFFF);
7345109e LP	273	hi = (after >> 32) & UINT32_C(0xFFFFFFFF);
939b8f14	274
7345109e	275	return write_string_filef(fn, 0, "%" PRIu32 " %" PRIu32, lo, hi);
939b8f14 LP	276	}
939b8f14 LP	277
a103496c	278	int capability_bounding_set_drop_usermode(uint64_t keep) {
939b8f14 LP	279	int r;
939b8f14 LP	280
a103496c	281	r = drop_from_file("/proc/sys/kernel/usermodehelper/inheritable", keep);
939b8f14 LP	282	if (r < 0)
	283	return r;
	284
a103496c	285	r = drop_from_file("/proc/sys/kernel/usermodehelper/bset", keep);
939b8f14 LP	286	if (r < 0)
	287	return r;
	288
	289	return r;
	290	}
966bff26	291
ed617ec2	292	int drop_privileges(uid_t uid, gid_t gid, uint64_t keep_capabilities) {
966bff26 LP	293	int r;
966bff26 LP	294
bff74941 LP	295	/* Unfortunately we cannot leave privilege dropping to PID 1 here, since we want to run as user but
	296	* want to keep some capabilities. Since file capabilities have been introduced this cannot be done
	297	* across exec() anymore, unless our binary has the capability configured in the file system, which
	298	* we want to avoid. */
966bff26	299
4a62c710 MS	300	if (setresgid(gid, gid, gid) < 0)
4a62c710 MS	301	return log_error_errno(errno, "Failed to change group ID: %m");
966bff26	302
97f0e76f LP	303	r = maybe_setgroups(0, NULL);
	304	if (r < 0)
	305	return log_error_errno(r, "Failed to drop auxiliary groups list: %m");
966bff26	306
bff74941 LP	307	/* Ensure we keep the permitted caps across the setresuid(). Note that we do this even if we actually
	308	* don't want to keep any capabilities, since we want to be able to drop them from the bounding set
	309	* too, and we can only do that if we have capabilities. */
51ddf615	310	if (prctl(PR_SET_KEEPCAPS, 1) < 0)
4a62c710	311	return log_error_errno(errno, "Failed to enable keep capabilities flag: %m");
966bff26	312
b1c05b98	313	if (setresuid(uid, uid, uid) < 0)
4a62c710	314	return log_error_errno(errno, "Failed to change user ID: %m");
966bff26	315
4a62c710 MS	316	if (prctl(PR_SET_KEEPCAPS, 0) < 0)
4a62c710 MS	317	return log_error_errno(errno, "Failed to disable keep capabilities flag: %m");
966bff26	318
bff74941 LP	319	/* Drop all caps from the bounding set (as well as the inheritable/permitted/effective sets), except
bff74941 LP	320	* the ones we want to keep */
a103496c	321	r = capability_bounding_set_drop(keep_capabilities, true);
f647962d MS	322	if (r < 0)
f647962d MS	323	return log_error_errno(r, "Failed to drop capabilities: %m");
966bff26	324
f11943c5	325	/* Now upgrade the permitted caps we still kept to effective caps */
bff74941	326	if (keep_capabilities != 0) {
057255fb	327	cap_value_t bits[u64log2(keep_capabilities) + 1];
bff74941 LP	328	_cleanup_cap_free_ cap_t d = NULL;
	329	unsigned i, j = 0;
	330
	331	d = cap_init();
	332	if (!d)
	333	return log_oom();
6a010ac9	334
7d328b54	335	for (i = 0; i < ELEMENTSOF(bits); i++)
51ddf615 ZJS	336	if (keep_capabilities & (1ULL << i))
51ddf615 ZJS	337	bits[j++] = i;
057255fb	338
2c9fc266 ZJS	339	/* use enough bits */
	340	assert(i == 64 \|\| (keep_capabilities >> i) == 0);
	341	/* don't use too many bits */
246bb370	342	assert(keep_capabilities & (UINT64_C(1) << (i - 1)));
966bff26	343
51ddf615	344	if (cap_set_flag(d, CAP_EFFECTIVE, j, bits, CAP_SET) < 0 \|\|
e1427b13 MS	345	cap_set_flag(d, CAP_PERMITTED, j, bits, CAP_SET) < 0)
e1427b13 MS	346	return log_error_errno(errno, "Failed to enable capabilities bits: %m");
51ddf615 ZJS	347
	348	if (cap_set_proc(d) < 0)
	349	return log_error_errno(errno, "Failed to increase capabilities: %m");
	350	}
966bff26 LP	351
	352	return 0;
	353	}
dd5ae4c3 PK	354
	355	int drop_capability(cap_value_t cv) {
	356	_cleanup_cap_free_ cap_t tmp_cap = NULL;
	357
	358	tmp_cap = cap_get_proc();
	359	if (!tmp_cap)
	360	return -errno;
	361
	362	if ((cap_set_flag(tmp_cap, CAP_INHERITABLE, 1, &cv, CAP_CLEAR) < 0) \|\|
	363	(cap_set_flag(tmp_cap, CAP_PERMITTED, 1, &cv, CAP_CLEAR) < 0) \|\|
	364	(cap_set_flag(tmp_cap, CAP_EFFECTIVE, 1, &cv, CAP_CLEAR) < 0))
	365	return -errno;
	366
	367	if (cap_set_proc(tmp_cap) < 0)
	368	return -errno;
	369
	370	return 0;
	371	}
39f608e4 LP	372
	373	bool ambient_capabilities_supported(void) {
	374	static int cache = -1;
	375
	376	if (cache >= 0)
	377	return cache;
	378
	379	/* If PR_CAP_AMBIENT returns something valid, or an unexpected error code we assume that ambient caps are
	380	* available. */
	381
	382	cache = prctl(PR_CAP_AMBIENT, PR_CAP_AMBIENT_IS_SET, CAP_KILL, 0, 0) >= 0 \|\|
	383	!IN_SET(errno, EINVAL, EOPNOTSUPP, ENOSYS);
	384
	385	return cache;
	386	}
d7391698	387
f66ad460	388	bool capability_quintet_mangle(CapabilityQuintet *q) {
f66ad460 AZ	389	uint64_t combined, drop = 0;
	390	bool ambient_supported;
	391
	392	assert(q);
	393
	394	combined = q->effective \| q->bounding \| q->inheritable \| q->permitted;
	395
	396	ambient_supported = q->ambient != (uint64_t) -1;
	397	if (ambient_supported)
	398	combined \|= q->ambient;
	399
864a25d9	400	for (unsigned i = 0; i <= cap_last_cap(); i++) {
f66ad460 AZ	401	unsigned long bit = UINT64_C(1) << i;
	402	if (!FLAGS_SET(combined, bit))
	403	continue;
	404
	405	if (prctl(PR_CAPBSET_READ, i) > 0)
	406	continue;
	407
	408	drop \|= bit;
	409
	410	log_debug("Not in the current bounding set: %s", capability_to_name(i));
	411	}
	412
	413	q->effective &= ~drop;
	414	q->bounding &= ~drop;
	415	q->inheritable &= ~drop;
	416	q->permitted &= ~drop;
	417
	418	if (ambient_supported)
	419	q->ambient &= ~drop;
	420
	421	return drop != 0; /* Let the caller know we changed something */
	422	}
	423
d7391698	424	int capability_quintet_enforce(const CapabilityQuintet *q) {
6757a013	425	_cleanup_cap_free_ cap_t c = NULL, modified = NULL;
d7391698 LP	426	int r;
	427
	428	if (q->ambient != (uint64_t) -1) {
d7391698 LP	429	bool changed = false;
	430
	431	c = cap_get_proc();
	432	if (!c)
	433	return -errno;
	434
864a25d9 ZJS	435	/* In order to raise the ambient caps set we first need to raise the matching
	436	* inheritable + permitted cap */
	437	for (unsigned i = 0; i <= cap_last_cap(); i++) {
d7391698 LP	438	uint64_t m = UINT64_C(1) << i;
	439	cap_value_t cv = (cap_value_t) i;
	440	cap_flag_value_t old_value_inheritable, old_value_permitted;
	441
	442	if ((q->ambient & m) == 0)
	443	continue;
	444
	445	if (cap_get_flag(c, cv, CAP_INHERITABLE, &old_value_inheritable) < 0)
	446	return -errno;
	447	if (cap_get_flag(c, cv, CAP_PERMITTED, &old_value_permitted) < 0)
	448	return -errno;
	449
	450	if (old_value_inheritable == CAP_SET && old_value_permitted == CAP_SET)
	451	continue;
	452
	453	if (cap_set_flag(c, CAP_INHERITABLE, 1, &cv, CAP_SET) < 0)
	454	return -errno;
d7391698 LP	455	if (cap_set_flag(c, CAP_PERMITTED, 1, &cv, CAP_SET) < 0)
	456	return -errno;
	457
	458	changed = true;
	459	}
	460
	461	if (changed)
	462	if (cap_set_proc(c) < 0)
	463	return -errno;
	464
	465	r = capability_ambient_set_apply(q->ambient, false);
	466	if (r < 0)
	467	return r;
	468	}
	469
	470	if (q->inheritable != (uint64_t) -1 \|\| q->permitted != (uint64_t) -1 \|\| q->effective != (uint64_t) -1) {
	471	bool changed = false;
d7391698 LP	472
	473	if (!c) {
	474	c = cap_get_proc();
	475	if (!c)
	476	return -errno;
	477	}
	478
864a25d9	479	for (unsigned i = 0; i <= cap_last_cap(); i++) {
d7391698 LP	480	uint64_t m = UINT64_C(1) << i;
	481	cap_value_t cv = (cap_value_t) i;
	482
	483	if (q->inheritable != (uint64_t) -1) {
	484	cap_flag_value_t old_value, new_value;
	485
248dd941 LP	486	if (cap_get_flag(c, cv, CAP_INHERITABLE, &old_value) < 0) {
	487	if (errno == EINVAL) /* If the kernel knows more caps than this
	488	* version of libcap, then this will return
	489	* EINVAL. In that case, simply ignore it,
	490	* pretend it doesn't exist. */
	491	continue;
	492
d7391698	493	return -errno;
248dd941	494	}
d7391698 LP	495
	496	new_value = (q->inheritable & m) ? CAP_SET : CAP_CLEAR;
	497
	498	if (old_value != new_value) {
	499	changed = true;
	500
	501	if (cap_set_flag(c, CAP_INHERITABLE, 1, &cv, new_value) < 0)
	502	return -errno;
	503	}
	504	}
	505
	506	if (q->permitted != (uint64_t) -1) {
	507	cap_flag_value_t old_value, new_value;
	508
248dd941 LP	509	if (cap_get_flag(c, cv, CAP_PERMITTED, &old_value) < 0) {
	510	if (errno == EINVAL)
	511	continue;
	512
d7391698	513	return -errno;
248dd941	514	}
d7391698 LP	515
	516	new_value = (q->permitted & m) ? CAP_SET : CAP_CLEAR;
	517
	518	if (old_value != new_value) {
	519	changed = true;
	520
	521	if (cap_set_flag(c, CAP_PERMITTED, 1, &cv, new_value) < 0)
	522	return -errno;
	523	}
	524	}
	525
	526	if (q->effective != (uint64_t) -1) {
	527	cap_flag_value_t old_value, new_value;
	528
248dd941 LP	529	if (cap_get_flag(c, cv, CAP_EFFECTIVE, &old_value) < 0) {
	530	if (errno == EINVAL)
	531	continue;
	532
d7391698	533	return -errno;
248dd941	534	}
d7391698 LP	535
	536	new_value = (q->effective & m) ? CAP_SET : CAP_CLEAR;
	537
	538	if (old_value != new_value) {
	539	changed = true;
	540
	541	if (cap_set_flag(c, CAP_EFFECTIVE, 1, &cv, new_value) < 0)
	542	return -errno;
	543	}
	544	}
	545	}
	546
9a2c5911	547	if (changed) {
9a2c5911 LP	548	/* In order to change the bounding caps, we need to keep CAP_SETPCAP for a bit
	549	* longer. Let's add it to our list hence for now. */
	550	if (q->bounding != (uint64_t) -1) {
	551	cap_value_t cv = CAP_SETPCAP;
	552
	553	modified = cap_dup(c);
	554	if (!modified)
	555	return -ENOMEM;
	556
	557	if (cap_set_flag(modified, CAP_PERMITTED, 1, &cv, CAP_SET) < 0)
	558	return -errno;
	559	if (cap_set_flag(modified, CAP_EFFECTIVE, 1, &cv, CAP_SET) < 0)
	560	return -errno;
	561
	562	if (cap_compare(modified, c) == 0) {
	563	/* No change? then drop this nonsense again */
	564	cap_free(modified);
	565	modified = NULL;
	566	}
	567	}
	568
	569	/* Now, let's enforce the caps for the first time. Note that this is where we acquire
	570	* caps in any of the sets we currently don't have. We have to do this before
664ff517 LP	571	* dropping the bounding caps below, since at that point we can never acquire new
664ff517 LP	572	* caps in inherited/permitted/effective anymore, but only lose them. */
9a2c5911	573	if (cap_set_proc(modified ?: c) < 0)
d7391698	574	return -errno;
9a2c5911	575	}
d7391698 LP	576	}
	577
	578	if (q->bounding != (uint64_t) -1) {
	579	r = capability_bounding_set_drop(q->bounding, false);
	580	if (r < 0)
	581	return r;
	582	}
	583
9a2c5911 LP	584	/* If needed, let's now set the caps again, this time in the final version, which differs from what
	585	* we have already set only in the CAP_SETPCAP bit, which we needed for dropping the bounding
	586	* bits. This call only undoes bits and doesn't acquire any which means the bounding caps don't
	587	* matter. */
6757a013	588	if (modified)
9a2c5911 LP	589	if (cap_set_proc(c) < 0)
	590	return -errno;
	591
d7391698 LP	592	return 0;
d7391698 LP	593	}