Merge pull request #2265 from ipuustin/ambient

capabilities: added support for ambient capabilities.

man/systemd.exec.xml

@@ -806,6 +806,35 @@
         settings.</para></listitem>
       </varlistentry>
 
+      <varlistentry>
+        <term><varname>AmbientCapabilities=</varname></term>
+
+        <listitem><para>Controls which capabilities to include in the
+        ambient capability set for the executed process. Takes a
+        whitespace-separated list of capability names as read by
+        <citerefentry project='mankier'><refentrytitle>cap_from_name</refentrytitle><manvolnum>3</manvolnum></citerefentry>,
+        e.g. <constant>CAP_SYS_ADMIN</constant>,
+        <constant>CAP_DAC_OVERRIDE</constant>,
+        <constant>CAP_SYS_PTRACE</constant>. This option may appear more than
+        once in which case the ambient capability sets are merged.
+        If the list of capabilities is prefixed with <literal>~</literal>, all
+        but the listed capabilities will be included, the effect of the
+        assignment inverted. If the empty string is
+        assigned to this option, the ambient capability set is reset to
+        the empty capability set, and all prior settings have no effect.
+        If set to <literal>~</literal> (without any further argument), the
+        ambient capability set is reset to the full set of available
+        capabilities, also undoing any previous settings. Note that adding
+        capabilities to ambient capability set adds them to the process's
+        inherited capability set.
+        </para><para>
+        Ambient capability sets are useful if you want to execute a process
+        as a non-privileged user but still want to give it some capabilities.
+        Note that in this case option <constant>keep-caps</constant> is
+        automatically added to <varname>SecureBits=</varname> to retain the
+        capabilities over the user change.</para></listitem>
+      </varlistentry>
+
       <varlistentry>
         <term><varname>SecureBits=</varname></term>
         <listitem><para>Controls the secure bits set for the executed

src/basic/capability-util.c

@@ -96,7 +96,62 @@ unsigned long cap_last_cap(void) {
         return p;
 }
 
-int capability_bounding_set_drop(uint64_t drop, bool right_now) {
+int capability_update_inherited_set(cap_t caps, uint64_t set) {
+        unsigned long i;
+
+        /* Add capabilities in the set to the inherited caps. Do not apply
+         * them yet. */
+
+        for (i = 0; i < cap_last_cap(); i++) {
+
+                if (set & (UINT64_C(1) << i)) {
+                        cap_value_t v;
+
+                        v = (cap_value_t) i;
+
+                        /* Make the capability inheritable. */
+                        if (cap_set_flag(caps, CAP_INHERITABLE, 1, &v, CAP_SET) < 0)
+                                return -errno;
+                }
+        }
+
+        return 0;
+}
+
+int capability_ambient_set_apply(uint64_t set, bool also_inherit) {
+        unsigned long i;
+        _cleanup_cap_free_ cap_t caps = NULL;
+
+        /* Add the capabilities to the ambient set. */
+
+        if (also_inherit) {
+                int r;
+                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 (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;
+                }
+        }
+
+        return 0;
+}
+
+int capability_bounding_set_drop(uint64_t keep, bool right_now) {
         _cleanup_cap_free_ cap_t after_cap = NULL;
         cap_flag_value_t fv;
         unsigned long i;
@@ -137,7 +192,7 @@ int capability_bounding_set_drop(uint64_t drop, bool right_now) {
 
         for (i = 0; i <= cap_last_cap(); i++) {
 
-                if (drop & ((uint64_t) 1ULL << (uint64_t) i)) {
+                if (!(keep & (UINT64_C(1) << i))) {
                         cap_value_t v;
 
                         /* Drop it from the bounding set */
@@ -176,7 +231,7 @@ finish:
         return r;
 }
 
-static int drop_from_file(const char *fn, uint64_t drop) {
+static int drop_from_file(const char *fn, uint64_t keep) {
         int r, k;
         uint32_t hi, lo;
         uint64_t current, after;
@@ -196,7 +251,7 @@ static int drop_from_file(const char *fn, uint64_t drop) {
                 return -EIO;
 
         current = (uint64_t) lo | ((uint64_t) hi << 32ULL);
-        after = current & ~drop;
+        after = current & keep;
 
         if (current == after)
                 return 0;
@@ -213,14 +268,14 @@ static int drop_from_file(const char *fn, uint64_t drop) {
         return r;
 }
 
-int capability_bounding_set_drop_usermode(uint64_t drop) {
+int capability_bounding_set_drop_usermode(uint64_t keep) {
         int r;
 
-        r = drop_from_file("/proc/sys/kernel/usermodehelper/inheritable", drop);
+        r = drop_from_file("/proc/sys/kernel/usermodehelper/inheritable", keep);
         if (r < 0)
                 return r;
 
-        r = drop_from_file("/proc/sys/kernel/usermodehelper/bset", drop);
+        r = drop_from_file("/proc/sys/kernel/usermodehelper/bset", keep);
         if (r < 0)
                 return r;
 
@@ -257,7 +312,7 @@ int drop_privileges(uid_t uid, gid_t gid, uint64_t keep_capabilities) {
                 return log_error_errno(errno, "Failed to disable keep capabilities flag: %m");
 
         /* Drop all caps from the bounding set, except the ones we want */
-        r = capability_bounding_set_drop(~keep_capabilities, true);
+        r = capability_bounding_set_drop(keep_capabilities, true);
         if (r < 0)
                 return log_error_errno(r, "Failed to drop capabilities: %m");
 

src/basic/capability-util.h

@@ -29,10 +29,15 @@
 #include "macro.h"
 #include "util.h"
 
+#define CAP_ALL (uint64_t) -1
+
 unsigned long cap_last_cap(void);
 int have_effective_cap(int value);
-int capability_bounding_set_drop(uint64_t drop, bool right_now);
-int capability_bounding_set_drop_usermode(uint64_t drop);
+int capability_bounding_set_drop(uint64_t keep, bool right_now);
+int capability_bounding_set_drop_usermode(uint64_t keep);
+
+int capability_ambient_set_apply(uint64_t set, bool also_inherit);
+int capability_update_inherited_set(cap_t caps, uint64_t ambient_set);
 
 int drop_privileges(uid_t uid, gid_t gid, uint64_t keep_capabilities);
 
@@ -46,3 +51,9 @@ static inline void cap_free_charpp(char **p) {
                 cap_free(*p);
 }
 #define _cleanup_cap_free_charp_ _cleanup_(cap_free_charpp)
+
+static inline bool cap_test_all(uint64_t caps) {
+        uint64_t m;
+        m = (UINT64_C(1) << (cap_last_cap() + 1)) - 1;
+        return (caps & m) == m;
+}

src/basic/missing.h

@@ -1129,3 +1129,19 @@ static inline key_serial_t request_key(const char *type, const char *description
 #ifndef KEY_SPEC_USER_KEYRING
 #define KEY_SPEC_USER_KEYRING -4
 #endif
+
+#ifndef PR_CAP_AMBIENT
+#define PR_CAP_AMBIENT 47
+#endif
+
+#ifndef PR_CAP_AMBIENT_IS_SET
+#define PR_CAP_AMBIENT_IS_SET 1
+#endif
+
+#ifndef PR_CAP_AMBIENT_RAISE
+#define PR_CAP_AMBIENT_RAISE 2
+#endif
+
+#ifndef PR_CAP_AMBIENT_CLEAR_ALL
+#define PR_CAP_AMBIENT_CLEAR_ALL 4
+#endif

src/core/dbus-execute.c

@@ -293,9 +293,25 @@ static int property_get_capability_bounding_set(
         assert(reply);
         assert(c);
 
-        /* We store this negated internally, to match the kernel, but
-         * we expose it normalized. */
-        return sd_bus_message_append(reply, "t", ~c->capability_bounding_set_drop);
+        return sd_bus_message_append(reply, "t", c->capability_bounding_set);
+}
+
+static int property_get_ambient_capabilities(
+                sd_bus *bus,
+                const char *path,
+                const char *interface,
+                const char *property,
+                sd_bus_message *reply,
+                void *userdata,
+                sd_bus_error *error) {
+
+        ExecContext *c = userdata;
+
+        assert(bus);
+        assert(reply);
+        assert(c);
+
+        return sd_bus_message_append(reply, "t", c->capability_ambient_set);
 }
 
 static int property_get_capabilities(
@@ -689,6 +705,7 @@ const sd_bus_vtable bus_exec_vtable[] = {
         SD_BUS_PROPERTY("Capabilities", "s", property_get_capabilities, 0, SD_BUS_VTABLE_PROPERTY_CONST),
         SD_BUS_PROPERTY("SecureBits", "i", bus_property_get_int, offsetof(ExecContext, secure_bits), SD_BUS_VTABLE_PROPERTY_CONST),
         SD_BUS_PROPERTY("CapabilityBoundingSet", "t", property_get_capability_bounding_set, 0, SD_BUS_VTABLE_PROPERTY_CONST),
+        SD_BUS_PROPERTY("AmbientCapabilities", "t", property_get_ambient_capabilities, 0, SD_BUS_VTABLE_PROPERTY_CONST),
         SD_BUS_PROPERTY("User", "s", NULL, offsetof(ExecContext, user), SD_BUS_VTABLE_PROPERTY_CONST),
         SD_BUS_PROPERTY("Group", "s", NULL, offsetof(ExecContext, group), SD_BUS_VTABLE_PROPERTY_CONST),
         SD_BUS_PROPERTY("SupplementaryGroups", "as", NULL, offsetof(ExecContext, supplementary_groups), SD_BUS_VTABLE_PROPERTY_CONST),

src/core/execute.c

@@ -737,12 +737,7 @@ static int enforce_user(const ExecContext *context, uid_t uid) {
         /* Sets (but doesn't lookup) the uid and make sure we keep the
          * capabilities while doing so. */
 
-        if (context->capabilities) {
-                _cleanup_cap_free_ cap_t d = NULL;
-                static const cap_value_t bits[] = {
-                        CAP_SETUID,   /* Necessary so that we can run setresuid() below */
-                        CAP_SETPCAP   /* Necessary so that we can set PR_SET_SECUREBITS later on */
-                };
+        if (context->capabilities || context->capability_ambient_set != 0) {
 
                 /* First step: If we need to keep capabilities but
                  * drop privileges we need to make sure we keep our
@@ -758,16 +753,24 @@ static int enforce_user(const ExecContext *context, uid_t uid) {
                 /* Second step: set the capabilities. This will reduce
                  * the capabilities to the minimum we need. */
 
-                d = cap_dup(context->capabilities);
-                if (!d)
-                        return -errno;
+                if (context->capabilities) {
+                        _cleanup_cap_free_ cap_t d = NULL;
+                        static const cap_value_t bits[] = {
+                                CAP_SETUID,   /* Necessary so that we can run setresuid() below */
+                                CAP_SETPCAP   /* Necessary so that we can set PR_SET_SECUREBITS later on */
+                        };
 
-                if (cap_set_flag(d, CAP_EFFECTIVE, ELEMENTSOF(bits), bits, CAP_SET) < 0 ||
-                    cap_set_flag(d, CAP_PERMITTED, ELEMENTSOF(bits), bits, CAP_SET) < 0)
-                        return -errno;
+                        d = cap_dup(context->capabilities);
+                        if (!d)
+                                return -errno;
 
-                if (cap_set_proc(d) < 0)
-                        return -errno;
+                        if (cap_set_flag(d, CAP_EFFECTIVE, ELEMENTSOF(bits), bits, CAP_SET) < 0 ||
+                            cap_set_flag(d, CAP_PERMITTED, ELEMENTSOF(bits), bits, CAP_SET) < 0)
+                                return -errno;
+
+                        if (cap_set_proc(d) < 0)
+                                return -errno;
+                }
         }
 
         /* Third step: actually set the uids */
@@ -1856,6 +1859,8 @@ static int exec_child(
 
         if (params->apply_permissions) {
 
+                int secure_bits = context->secure_bits;
+
                 for (i = 0; i < _RLIMIT_MAX; i++) {
                         if (!context->rlimit[i])
                                 continue;
@@ -1866,28 +1871,71 @@ static int exec_child(
                         }
                 }
 
-                if (context->capability_bounding_set_drop) {
-                        r = capability_bounding_set_drop(context->capability_bounding_set_drop, false);
+                if (!cap_test_all(context->capability_bounding_set)) {
+                        r = capability_bounding_set_drop(context->capability_bounding_set, false);
                         if (r < 0) {
                                 *exit_status = EXIT_CAPABILITIES;
                                 return r;
                         }
                 }
 
+                /* This is done before enforce_user, but ambient set
+                 * does not survive over setresuid() if keep_caps is not set. */
+                if (context->capability_ambient_set != 0) {
+                        r = capability_ambient_set_apply(context->capability_ambient_set, true);
+                        if (r < 0) {
+                                *exit_status = EXIT_CAPABILITIES;
+                                return r;
+                        }
+
+                        if (context->capabilities) {
+
+                                /* The capabilities in ambient set need to be also in the inherited
+                                 * set. If they aren't, trying to get them will fail. Add the ambient
+                                 * set inherited capabilities to the capability set in the context.
+                                 * This is needed because if capabilities are set (using "Capabilities="
+                                 * keyword), they will override whatever we set now. */
+
+                                r = capability_update_inherited_set(context->capabilities, context->capability_ambient_set);
+                                if (r < 0) {
+                                        *exit_status = EXIT_CAPABILITIES;
+                                        return r;
+                                }
+                        }
+                }
+
                 if (context->user) {
                         r = enforce_user(context, uid);
                         if (r < 0) {
                                 *exit_status = EXIT_USER;
                                 return r;
                         }
+                        if (context->capability_ambient_set != 0) {
+
+                                /* Fix the ambient capabilities after user change. */
+                                r = capability_ambient_set_apply(context->capability_ambient_set, false);
+                                if (r < 0) {
+                                        *exit_status = EXIT_CAPABILITIES;
+                                        return r;
+                                }
+
+                                /* If we were asked to change user and ambient capabilities
+                                 * were requested, we had to add keep-caps to the securebits
+                                 * so that we would maintain the inherited capability set
+                                 * through the setresuid(). Make sure that the bit is added
+                                 * also to the context secure_bits so that we don't try to
+                                 * drop the bit away next. */
+
+                                 secure_bits |= 1<<SECURE_KEEP_CAPS;
+                        }
                 }
 
                 /* PR_GET_SECUREBITS is not privileged, while
                  * PR_SET_SECUREBITS is. So to suppress
                  * potential EPERMs we'll try not to call
                  * PR_SET_SECUREBITS unless necessary. */
-                if (prctl(PR_GET_SECUREBITS) != context->secure_bits)
-                        if (prctl(PR_SET_SECUREBITS, context->secure_bits) < 0) {
+                if (prctl(PR_GET_SECUREBITS) != secure_bits)
+                        if (prctl(PR_SET_SECUREBITS, secure_bits) < 0) {
                                 *exit_status = EXIT_SECUREBITS;
                                 return -errno;
                         }
@@ -2114,6 +2162,7 @@ void exec_context_init(ExecContext *c) {
         c->timer_slack_nsec = NSEC_INFINITY;
         c->personality = PERSONALITY_INVALID;
         c->runtime_directory_mode = 0755;
+        c->capability_bounding_set = CAP_ALL;
 }
 
 void exec_context_done(ExecContext *c) {
@@ -2517,12 +2566,23 @@ void exec_context_dump(ExecContext *c, FILE* f, const char *prefix) {
                         (c->secure_bits & 1<<SECURE_NOROOT) ? " noroot" : "",
                         (c->secure_bits & 1<<SECURE_NOROOT_LOCKED) ? "noroot-locked" : "");
 
-        if (c->capability_bounding_set_drop) {
+        if (c->capability_bounding_set != CAP_ALL) {
                 unsigned long l;
                 fprintf(f, "%sCapabilityBoundingSet:", prefix);
 
                 for (l = 0; l <= cap_last_cap(); l++)
-                        if (!(c->capability_bounding_set_drop & ((uint64_t) 1ULL << (uint64_t) l)))
+                        if (c->capability_bounding_set & (UINT64_C(1) << l))
+                                fprintf(f, " %s", strna(capability_to_name(l)));
+
+                fputs("\n", f);
+        }
+
+        if (c->capability_ambient_set != 0) {
+                unsigned long l;
+                fprintf(f, "%sAmbientCapabilities:", prefix);
+
+                for (l = 0; l <= cap_last_cap(); l++)
+                        if (c->capability_ambient_set & (UINT64_C(1) << l))
                                 fprintf(f, " %s", strna(capability_to_name(l)));
 
                 fputs("\n", f);

src/core/execute.h

@@ -155,7 +155,9 @@ struct ExecContext {
         char **read_write_dirs, **read_only_dirs, **inaccessible_dirs;
         unsigned long mount_flags;
 
-        uint64_t capability_bounding_set_drop;
+        uint64_t capability_bounding_set;
+
+        uint64_t capability_ambient_set;
 
         cap_t capabilities;
         int secure_bits;

src/core/load-fragment-gperf.gperf.m4

@@ -47,7 +47,8 @@ $1.SyslogLevel,                  config_parse_log_level,             0,
 $1.SyslogLevelPrefix,            config_parse_bool,                  0,                             offsetof($1, exec_context.syslog_level_prefix)
 $1.Capabilities,                 config_parse_exec_capabilities,     0,                             offsetof($1, exec_context)
 $1.SecureBits,                   config_parse_exec_secure_bits,      0,                             offsetof($1, exec_context)
-$1.CapabilityBoundingSet,        config_parse_bounding_set,          0,                             offsetof($1, exec_context.capability_bounding_set_drop)
+$1.CapabilityBoundingSet,        config_parse_capability_set,        0,                             offsetof($1, exec_context.capability_bounding_set)
+$1.AmbientCapabilities,          config_parse_capability_set,        0,                             offsetof($1, exec_context.capability_ambient_set)
 $1.TimerSlackNSec,               config_parse_nsec,                  0,                             offsetof($1, exec_context.timer_slack_nsec)
 $1.NoNewPrivileges,              config_parse_no_new_privileges,     0,                             offsetof($1, exec_context)
 m4_ifdef(`HAVE_SECCOMP',

src/core/load-fragment.c

@@ -38,6 +38,7 @@
 #include "bus-internal.h"
 #include "bus-util.h"
 #include "cap-list.h"
+#include "capability-util.h"
 #include "cgroup.h"
 #include "conf-parser.h"
 #include "cpu-set-util.h"
@@ -1024,7 +1025,7 @@ int config_parse_exec_secure_bits(const char *unit,
         return 0;
 }
 
-int config_parse_bounding_set(
+int config_parse_capability_set(
                 const char *unit,
                 const char *filename,
                 unsigned line,
@@ -1036,8 +1037,8 @@ int config_parse_bounding_set(
                 void *data,
                 void *userdata) {
 
-        uint64_t *capability_bounding_set_drop = data;
-        uint64_t capability_bounding_set, sum = 0;
+        uint64_t *capability_set = data;
+        uint64_t sum = 0, initial = 0;
         bool invert = false;
         const char *p;
 
@@ -1051,10 +1052,9 @@ int config_parse_bounding_set(
                 rvalue++;
         }
 
-        /* Note that we store this inverted internally, since the
-         * kernel wants it like this. But we actually expose it
-         * non-inverted everywhere to have a fully normalized
-         * interface. */
+        if (strcmp(lvalue, "CapabilityBoundingSet") == 0)
+                initial = CAP_ALL; /* initialized to all bits on */
+        /* else "AmbientCapabilities" initialized to all bits off */
 
         p = rvalue;
         for (;;) {
@@ -1073,18 +1073,21 @@ int config_parse_bounding_set(
 
                 cap = capability_from_name(word);
                 if (cap < 0) {
-                        log_syntax(unit, LOG_ERR, filename, line, 0, "Failed to parse capability in bounding set, ignoring: %s", word);
+                        log_syntax(unit, LOG_ERR, filename, line, 0, "Failed to parse capability in bounding/ambient set, ignoring: %s", word);
                         continue;
                 }
 
                 sum |= ((uint64_t) UINT64_C(1)) << (uint64_t) cap;
         }
 
-        capability_bounding_set = invert ? ~sum : sum;
-        if (*capability_bounding_set_drop != 0 && capability_bounding_set != 0)
-                *capability_bounding_set_drop = ~(~*capability_bounding_set_drop | capability_bounding_set);
+        sum = invert ? ~sum : sum;
+
+        if (sum == 0 || *capability_set == initial)
+                /* "" or uninitialized data -> replace */
+                *capability_set = sum;
         else
-                *capability_bounding_set_drop = ~capability_bounding_set;
+                /* previous data -> merge */
+                *capability_set |= sum;
 
         return 0;
 }
@@ -4002,7 +4005,7 @@ void unit_dump_config_items(FILE *f) {
                 { config_parse_log_level,             "LEVEL" },
                 { config_parse_exec_capabilities,     "CAPABILITIES" },
                 { config_parse_exec_secure_bits,      "SECUREBITS" },
-                { config_parse_bounding_set,          "BOUNDINGSET" },
+                { config_parse_capability_set,        "BOUNDINGSET" },
                 { config_parse_limit,                 "LIMIT" },
                 { config_parse_unit_deps,             "UNIT [...]" },
                 { config_parse_exec,                  "PATH [ARGUMENT [...]]" },

src/core/load-fragment.h

@@ -56,7 +56,7 @@ int config_parse_exec_cpu_sched_prio(const char *unit, const char *filename, uns
 int config_parse_exec_cpu_affinity(const char *unit, const char *filename, unsigned line, const char *section, unsigned section_line, const char *lvalue, int ltype, const char *rvalue, void *data, void *userdata);
 int config_parse_exec_capabilities(const char *unit, const char *filename, unsigned line, const char *section, unsigned section_line, const char *lvalue, int ltype, const char *rvalue, void *data, void *userdata);
 int config_parse_exec_secure_bits(const char *unit, const char *filename, unsigned line, const char *section, unsigned section_line, const char *lvalue, int ltype, const char *rvalue, void *data, void *userdata);
-int config_parse_bounding_set(const char *unit, const char *filename, unsigned line, const char *section, unsigned section_line, const char *lvalue, int ltype, const char *rvalue, void *data, void *userdata);
+int config_parse_capability_set(const char *unit, const char *filename, unsigned line, const char *section, unsigned section_line, const char *lvalue, int ltype, const char *rvalue, void *data, void *userdata);
 int config_parse_limit(const char *unit, const char *filename, unsigned line, const char *section, unsigned section_line, const char *lvalue, int ltype, const char *rvalue, void *data, void *userdata);
 int config_parse_bytes_limit(const char *unit, const char *filename, unsigned line, const char *section, unsigned section_line, const char *lvalue, int ltype, const char *rvalue, void *data, void *userdata);
 int config_parse_sec_limit(const char *unit, const char *filename, unsigned line, const char *section, unsigned section_line, const char *lvalue, int ltype, const char *rvalue, void *data, void *userdata);

src/core/main.c

@@ -117,7 +117,7 @@ static usec_t arg_runtime_watchdog = 0;
 static usec_t arg_shutdown_watchdog = 10 * USEC_PER_MINUTE;
 static char **arg_default_environment = NULL;
 static struct rlimit *arg_default_rlimit[_RLIMIT_MAX] = {};
-static uint64_t arg_capability_bounding_set_drop = 0;
+static uint64_t arg_capability_bounding_set = CAP_ALL;
 static nsec_t arg_timer_slack_nsec = NSEC_INFINITY;
 static usec_t arg_default_timer_accuracy_usec = 1 * USEC_PER_MINUTE;
 static Set* arg_syscall_archs = NULL;
@@ -644,7 +644,7 @@ static int parse_config_file(void) {
                 { "Manager", "JoinControllers",           config_parse_join_controllers, 0, &arg_join_controllers                  },
                 { "Manager", "RuntimeWatchdogSec",        config_parse_sec,              0, &arg_runtime_watchdog                  },
                 { "Manager", "ShutdownWatchdogSec",       config_parse_sec,              0, &arg_shutdown_watchdog                 },
-                { "Manager", "CapabilityBoundingSet",     config_parse_bounding_set,     0, &arg_capability_bounding_set_drop      },
+                { "Manager", "CapabilityBoundingSet",     config_parse_capability_set,   0, &arg_capability_bounding_set           },
 #ifdef HAVE_SECCOMP
                 { "Manager", "SystemCallArchitectures",   config_parse_syscall_archs,    0, &arg_syscall_archs                     },
 #endif
@@ -1631,14 +1631,14 @@ int main(int argc, char *argv[]) {
                 if (prctl(PR_SET_TIMERSLACK, arg_timer_slack_nsec) < 0)
                         log_error_errno(errno, "Failed to adjust timer slack: %m");
 
-        if (arg_capability_bounding_set_drop) {
-                r = capability_bounding_set_drop_usermode(arg_capability_bounding_set_drop);
+        if (!cap_test_all(arg_capability_bounding_set)) {
+                r = capability_bounding_set_drop_usermode(arg_capability_bounding_set);
                 if (r < 0) {
                         log_emergency_errno(r, "Failed to drop capability bounding set of usermode helpers: %m");
                         error_message = "Failed to drop capability bounding set of usermode helpers";
                         goto finish;
                 }
-                r = capability_bounding_set_drop(arg_capability_bounding_set_drop, true);
+                r = capability_bounding_set_drop(arg_capability_bounding_set, true);
                 if (r < 0) {
                         log_emergency_errno(r, "Failed to drop capability bounding set: %m");
                         error_message = "Failed to drop capability bounding set";

src/core/unit.c

@@ -3231,7 +3231,7 @@ int unit_patch_contexts(Unit *u) {
                         ec->no_new_privileges = true;
 
                 if (ec->private_devices)
-                        ec->capability_bounding_set_drop |= (uint64_t) 1ULL << (uint64_t) CAP_MKNOD;
+                        ec->capability_bounding_set &= ~(UINT64_C(1) << CAP_MKNOD);
         }
 
         cc = unit_get_cgroup_context(u);

src/import/import-common.c

@@ -134,7 +134,7 @@ int import_fork_tar_x(const char *path, pid_t *ret) {
                 if (unshare(CLONE_NEWNET) < 0)
                         log_error_errno(errno, "Failed to lock tar into network namespace, ignoring: %m");
 
-                r = capability_bounding_set_drop(~retain, true);
+                r = capability_bounding_set_drop(retain, true);
                 if (r < 0)
                         log_error_errno(r, "Failed to drop capabilities, ignoring: %m");
 
@@ -208,7 +208,7 @@ int import_fork_tar_c(const char *path, pid_t *ret) {
                 if (unshare(CLONE_NEWNET) < 0)
                         log_error_errno(errno, "Failed to lock tar into network namespace, ignoring: %m");
 
-                r = capability_bounding_set_drop(~retain, true);
+                r = capability_bounding_set_drop(retain, true);
                 if (r < 0)
                         log_error_errno(r, "Failed to drop capabilities, ignoring: %m");
 

src/nspawn/nspawn.c

@@ -1482,7 +1482,7 @@ static int setup_journal(const char *directory) {
 }
 
 static int drop_capabilities(void) {
-        return capability_bounding_set_drop(~arg_retain, false);
+        return capability_bounding_set_drop(arg_retain, false);
 }
 
 static int reset_audit_loginuid(void) {

src/test/test-capability.c

@@ -20,6 +20,7 @@
 #include <netinet/in.h>
 #include <pwd.h>
 #include <sys/capability.h>
+#include <sys/prctl.h>
 #include <sys/socket.h>
 #include <sys/wait.h>
 #include <unistd.h>
@@ -66,8 +67,9 @@ static void show_capabilities(void) {
         cap_free(text);
 }
 
-static int setup_tests(void) {
+static int setup_tests(bool *run_ambient) {
         struct passwd *nobody;
+        int r;
 
         nobody = getpwnam("nobody");
         if (!nobody) {
@@ -77,6 +79,18 @@ static int setup_tests(void) {
         test_uid = nobody->pw_uid;
         test_gid = nobody->pw_gid;
 
+        *run_ambient = false;
+
+        r = prctl(PR_CAP_AMBIENT, PR_CAP_AMBIENT_CLEAR_ALL, 0, 0, 0);
+
+        /* There's support for PR_CAP_AMBIENT if the prctl() call
+         * succeeded or error code was something else than EINVAL. The
+         * EINVAL check should be good enough to rule out false
+         * positives. */
+
+        if (r >= 0 || errno != EINVAL)
+                *run_ambient = true;
+
         return 0;
 }
 
@@ -140,8 +154,53 @@ static void test_have_effective_cap(void) {
         assert_se(!have_effective_cap(CAP_CHOWN));
 }
 
+static void test_update_inherited_set(void) {
+        cap_t caps;
+        uint64_t set = 0;
+        cap_flag_value_t fv;
+
+        caps = cap_get_proc();
+        assert_se(caps);
+        assert_se(!cap_get_flag(caps, CAP_CHOWN, CAP_INHERITABLE, &fv));
+        assert(fv == CAP_CLEAR);
+
+        set = (UINT64_C(1) << CAP_CHOWN);
+
+        assert_se(!capability_update_inherited_set(caps, set));
+        assert_se(!cap_get_flag(caps, CAP_CHOWN, CAP_INHERITABLE, &fv));
+        assert(fv == CAP_SET);
+
+        cap_free(caps);
+}
+
+static void test_set_ambient_caps(void) {
+        cap_t caps;
+        uint64_t set = 0;
+        cap_flag_value_t fv;
+
+        caps = cap_get_proc();
+        assert_se(caps);
+        assert_se(!cap_get_flag(caps, CAP_CHOWN, CAP_INHERITABLE, &fv));
+        assert(fv == CAP_CLEAR);
+        cap_free(caps);
+
+        assert_se(prctl(PR_CAP_AMBIENT, PR_CAP_AMBIENT_IS_SET, CAP_CHOWN, 0, 0) == 0);
+
+        set = (UINT64_C(1) << CAP_CHOWN);
+
+        assert_se(!capability_ambient_set_apply(set, true));
+
+        caps = cap_get_proc();
+        assert_se(!cap_get_flag(caps, CAP_CHOWN, CAP_INHERITABLE, &fv));
+        assert(fv == CAP_SET);
+        cap_free(caps);
+
+        assert_se(prctl(PR_CAP_AMBIENT, PR_CAP_AMBIENT_IS_SET, CAP_CHOWN, 0, 0) == 1);
+}
+
 int main(int argc, char *argv[]) {
         int r;
+        bool run_ambient;
 
         log_parse_environment();
         log_open();
@@ -149,14 +208,19 @@ int main(int argc, char *argv[]) {
         if (getuid() != 0)
                 return EXIT_TEST_SKIP;
 
-        r = setup_tests();
+        r = setup_tests(&run_ambient);
         if (r < 0)
                 return -r;
 
         show_capabilities();
 
         test_drop_privileges();
+        test_update_inherited_set();
+
         fork_test(test_have_effective_cap);
 
+        if (run_ambient)
+                fork_test(test_set_ambient_caps);
+
         return 0;
 }

src/test/test-execute.c

@@ -20,6 +20,7 @@
 #include <grp.h>
 #include <pwd.h>
 #include <stdio.h>
+#include <sys/prctl.h>
 #include <sys/types.h>
 
 #include "fileio.h"
@@ -224,6 +225,20 @@ static void test_exec_capabilityboundingset(Manager *m) {
         test(m, "exec-capabilityboundingset-invert.service", 0, CLD_EXITED);
 }
 
+static void test_exec_capabilityambientset(Manager *m) {
+        int r;
+
+        /* Check if the kernel has support for ambient capabilities. Run
+         * the tests only if that's the case. Clearing all ambient
+         * capabilities is fine, since we are expecting them to be unset
+         * in the first place for the tests. */
+        r = prctl(PR_CAP_AMBIENT, PR_CAP_AMBIENT_CLEAR_ALL, 0, 0, 0);
+        if (r >= 0 || errno != EINVAL) {
+                test(m, "exec-capabilityambientset.service", 0, CLD_EXITED);
+                test(m, "exec-capabilityambientset-merge.service", 0, CLD_EXITED);
+        }
+}
+
 static void test_exec_privatenetwork(Manager *m) {
         int r;
 
@@ -266,6 +281,7 @@ int main(int argc, char *argv[]) {
                 test_exec_umask,
                 test_exec_runtimedirectory,
                 test_exec_capabilityboundingset,
+                test_exec_capabilityambientset,
                 test_exec_oomscoreadjust,
                 test_exec_ioschedulingclass,
                 NULL,

src/test/test-unit-file.c

@@ -28,6 +28,7 @@
 #include <unistd.h>
 
 #include "alloc-util.h"
+#include "capability-util.h"
 #include "fd-util.h"
 #include "fileio.h"
 #include "hashmap.h"
@@ -625,8 +626,8 @@ static uint64_t make_cap(int cap) {
         return ((uint64_t) 1ULL << (uint64_t) cap);
 }
 
-static void test_config_parse_bounding_set(void) {
-        /* int config_parse_bounding_set(
+static void test_config_parse_capability_set(void) {
+        /* int config_parse_capability_set(
                  const char *unit,
                  const char *filename,
                  unsigned line,
@@ -638,38 +639,38 @@ static void test_config_parse_bounding_set(void) {
                  void *data,
                  void *userdata) */
         int r;
-        uint64_t capability_bounding_set_drop = 0;
+        uint64_t capability_bounding_set = 0;
 
-        r = config_parse_bounding_set(NULL, "fake", 1, "section", 1,
+        r = config_parse_capability_set(NULL, "fake", 1, "section", 1,
                               "CapabilityBoundingSet", 0, "CAP_NET_RAW",
-                              &capability_bounding_set_drop, NULL);
+                              &capability_bounding_set, NULL);
         assert_se(r >= 0);
-        assert_se(capability_bounding_set_drop == ~make_cap(CAP_NET_RAW));
+        assert_se(capability_bounding_set == make_cap(CAP_NET_RAW));
 
-        r = config_parse_bounding_set(NULL, "fake", 1, "section", 1,
+        r = config_parse_capability_set(NULL, "fake", 1, "section", 1,
                               "CapabilityBoundingSet", 0, "CAP_NET_ADMIN",
-                              &capability_bounding_set_drop, NULL);
+                              &capability_bounding_set, NULL);
         assert_se(r >= 0);
-        assert_se(capability_bounding_set_drop == ~(make_cap(CAP_NET_RAW) | make_cap(CAP_NET_ADMIN)));
+        assert_se(capability_bounding_set == (make_cap(CAP_NET_RAW) | make_cap(CAP_NET_ADMIN)));
 
-        r = config_parse_bounding_set(NULL, "fake", 1, "section", 1,
+        r = config_parse_capability_set(NULL, "fake", 1, "section", 1,
                               "CapabilityBoundingSet", 0, "",
-                              &capability_bounding_set_drop, NULL);
+                              &capability_bounding_set, NULL);
         assert_se(r >= 0);
-        assert_se(capability_bounding_set_drop == ~((uint64_t) 0ULL));
+        assert_se(capability_bounding_set == UINT64_C(0));
 
-        r = config_parse_bounding_set(NULL, "fake", 1, "section", 1,
+        r = config_parse_capability_set(NULL, "fake", 1, "section", 1,
                               "CapabilityBoundingSet", 0, "~",
-                              &capability_bounding_set_drop, NULL);
+                              &capability_bounding_set, NULL);
         assert_se(r >= 0);
-        assert_se(capability_bounding_set_drop == (uint64_t) 0ULL);
+        assert_se(cap_test_all(capability_bounding_set));
 
-        capability_bounding_set_drop = 0;
-        r = config_parse_bounding_set(NULL, "fake", 1, "section", 1,
+        capability_bounding_set = 0;
+        r = config_parse_capability_set(NULL, "fake", 1, "section", 1,
                               "CapabilityBoundingSet", 0, "  'CAP_NET_RAW' WAT_CAP??? CAP_NET_ADMIN CAP'_trailing_garbage",
-                              &capability_bounding_set_drop, NULL);
+                              &capability_bounding_set, NULL);
         assert_se(r >= 0);
-        assert_se(capability_bounding_set_drop == ~(make_cap(CAP_NET_RAW) | make_cap(CAP_NET_ADMIN)));
+        assert_se(capability_bounding_set == (make_cap(CAP_NET_RAW) | make_cap(CAP_NET_ADMIN)));
 }
 
 static void test_config_parse_rlimit(void) {
@@ -829,7 +830,7 @@ int main(int argc, char *argv[]) {
 
         r = test_unit_file_get_set();
         test_config_parse_exec();
-        test_config_parse_bounding_set();
+        test_config_parse_capability_set();
         test_config_parse_rlimit();
         test_config_parse_pass_environ();
         test_load_env_file_1();

test/test-execute/exec-capabilityambientset-merge.service

@@ -0,0 +1,9 @@
+[Unit]
+Description=Test for AmbientCapabilities
+
+[Service]
+ExecStart=/bin/sh -x -c 'c=$$(grep "CapAmb:" /proc/self/status); test "$$c" = "CapAmb:	0000000000003000"'
+Type=oneshot
+User=nobody
+AmbientCapabilities=CAP_NET_ADMIN
+AmbientCapabilities=CAP_NET_RAW

test/test-execute/exec-capabilityambientset.service

@@ -0,0 +1,8 @@
+[Unit]
+Description=Test for AmbientCapabilities
+
+[Service]
+ExecStart=/bin/sh -x -c 'c=$$(grep "CapAmb:" /proc/self/status); test "$$c" = "CapAmb:	0000000000003000"'
+Type=oneshot
+User=nobody
+AmbientCapabilities=CAP_NET_ADMIN CAP_NET_RAW