chiark / gitweb /
exec: introduce PrivateNetwork= process option to turn off network access to specific...
[elogind.git] / src / execute.c
index 1e376ff89e08beacdf46aa344e3a2128f9bde6ac..c73b0c6c04866704d3af9f750ea5143669daebfa 100644 (file)
@@ -56,6 +56,7 @@
 #include "missing.h"
 #include "utmp-wtmp.h"
 #include "def.h"
+#include "loopback-setup.h"
 
 /* This assumes there is a 'tty' group */
 #define TTY_MODE 0620
@@ -140,6 +141,19 @@ static const char *tty_path(const ExecContext *context) {
         return "/dev/console";
 }
 
+void exec_context_tty_reset(const ExecContext *context) {
+        assert(context);
+
+        if (context->tty_vhangup)
+                terminal_vhangup(tty_path(context));
+
+        if (context->tty_reset)
+                reset_terminal(tty_path(context));
+
+        if (context->tty_vt_disallocate && context->tty_path)
+                vt_disallocate(context->tty_path);
+}
+
 static int open_null_as(int flags, int nfd) {
         int fd, r;
 
@@ -536,82 +550,6 @@ static int restore_confirm_stdio(const ExecContext *context,
         return 0;
 }
 
-static int get_group_creds(const char *groupname, gid_t *gid) {
-        struct group *g;
-        unsigned long lu;
-
-        assert(groupname);
-        assert(gid);
-
-        /* We enforce some special rules for gid=0: in order to avoid
-         * NSS lookups for root we hardcode its data. */
-
-        if (streq(groupname, "root") || streq(groupname, "0")) {
-                *gid = 0;
-                return 0;
-        }
-
-        if (safe_atolu(groupname, &lu) >= 0) {
-                errno = 0;
-                g = getgrgid((gid_t) lu);
-        } else {
-                errno = 0;
-                g = getgrnam(groupname);
-        }
-
-        if (!g)
-                return errno != 0 ? -errno : -ESRCH;
-
-        *gid = g->gr_gid;
-        return 0;
-}
-
-static int get_user_creds(const char **username, uid_t *uid, gid_t *gid, const char **home) {
-        struct passwd *p;
-        unsigned long lu;
-
-        assert(username);
-        assert(*username);
-        assert(uid);
-        assert(gid);
-        assert(home);
-
-        /* We enforce some special rules for uid=0: in order to avoid
-         * NSS lookups for root we hardcode its data. */
-
-        if (streq(*username, "root") || streq(*username, "0")) {
-                *username = "root";
-                *uid = 0;
-                *gid = 0;
-                *home = "/root";
-                return 0;
-        }
-
-        if (safe_atolu(*username, &lu) >= 0) {
-                errno = 0;
-                p = getpwuid((uid_t) lu);
-
-                /* If there are multiple users with the same id, make
-                 * sure to leave $USER to the configured value instead
-                 * of the first occurrence in the database. However if
-                 * the uid was configured by a numeric uid, then let's
-                 * pick the real username from /etc/passwd. */
-                if (*username && p)
-                        *username = p->pw_name;
-        } else {
-                errno = 0;
-                p = getpwnam(*username);
-        }
-
-        if (!p)
-                return errno != 0 ? -errno : -ESRCH;
-
-        *uid = p->pw_uid;
-        *gid = p->pw_gid;
-        *home = p->pw_dir;
-        return 0;
-}
-
 static int enforce_groups(const ExecContext *context, const char *username, gid_t gid) {
         bool keep_groups = false;
         int r;
@@ -623,9 +561,12 @@ static int enforce_groups(const ExecContext *context, const char *username, gid_
 
         if (context->group || username) {
 
-                if (context->group)
-                        if ((r = get_group_creds(context->group, &gid)) < 0)
+                if (context->group) {
+                        const char *g = context->group;
+
+                        if ((r = get_group_creds(&g, &gid)) < 0)
                                 return r;
+                }
 
                 /* First step, initialize groups from /etc/groups */
                 if (username && gid != 0) {
@@ -660,13 +601,16 @@ static int enforce_groups(const ExecContext *context, const char *username, gid_
                         k = 0;
 
                 STRV_FOREACH(i, context->supplementary_groups) {
+                        const char *g;
 
                         if (k >= ngroups_max) {
                                 free(gids);
                                 return -E2BIG;
                         }
 
-                        if ((r = get_group_creds(*i, gids+k)) < 0) {
+                        g = *i;
+                        r = get_group_creds(&g, gids+k);
+                        if (r < 0) {
                                 free(gids);
                                 return r;
                         }
@@ -804,9 +748,6 @@ static int setup_pam(
 
         close_session = true;
 
-        if ((pam_code = pam_setcred(handle, PAM_ESTABLISH_CRED | PAM_SILENT)) != PAM_SUCCESS)
-                goto fail;
-
         if ((!(e = pam_getenvlist(handle)))) {
                 pam_code = PAM_BUF_ERR;
                 goto fail;
@@ -833,7 +774,7 @@ static int setup_pam(
 
                 /* This string must fit in 10 chars (i.e. the length
                  * of "/sbin/init") */
-                rename_process("sd:pam");
+                rename_process("sd(PAM)");
 
                 /* Make sure we don't keep open the passed fds in this
                 child. We assume that otherwise only those fds are
@@ -851,13 +792,20 @@ static int setup_pam(
                 /* Check if our parent process might already have
                  * died? */
                 if (getppid() == parent_pid) {
-                        if (sigwait(&ss, &sig) < 0)
-                                goto child_finish;
+                        for (;;) {
+                                if (sigwait(&ss, &sig) < 0) {
+                                        if (errno == EINTR)
+                                                continue;
+
+                                        goto child_finish;
+                                }
 
-                        assert(sig == SIGTERM);
+                                assert(sig == SIGTERM);
+                                break;
+                        }
                 }
 
-                /* Only if our parent died we'll end the session */
+                /* If our parent died we'll end the session */
                 if (getppid() != parent_pid)
                         if ((pam_code = pam_close_session(handle, PAM_DATA_SILENT)) != PAM_SUCCESS)
                                 goto child_finish;
@@ -873,7 +821,7 @@ static int setup_pam(
          * cleanups, so forget about the handle here. */
         handle = NULL;
 
-        /* Unblock SIGSUR1 again in the parent */
+        /* Unblock SIGTERM again in the parent */
         if (sigprocmask(SIG_SETMASK, &old_ss, NULL) < 0)
                 goto fail;
 
@@ -881,6 +829,9 @@ static int setup_pam(
          * might have opened it, but we don't want this fd around. */
         closelog();
 
+        *pam_env = e;
+        e = NULL;
+
         return 0;
 
 fail:
@@ -904,6 +855,71 @@ fail:
 }
 #endif
 
+static int do_capability_bounding_set_drop(uint64_t drop) {
+        unsigned long i;
+        cap_t old_cap = NULL, new_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. */
+
+        old_cap = cap_get_proc();
+        if (!old_cap)
+                return -errno;
+
+        if (cap_get_flag(old_cap, CAP_SETPCAP, CAP_EFFECTIVE, &fv) < 0) {
+                r = -errno;
+                goto finish;
+        }
+
+        if (fv != CAP_SET) {
+                static const cap_value_t v = CAP_SETPCAP;
+
+                new_cap = cap_dup(old_cap);
+                if (!new_cap) {
+                        r = -errno;
+                        goto finish;
+                }
+
+                if (cap_set_flag(new_cap, CAP_EFFECTIVE, 1, &v, CAP_SET) < 0) {
+                        r = -errno;
+                        goto finish;
+                }
+
+                if (cap_set_proc(new_cap) < 0) {
+                        r = -errno;
+                        goto finish;
+                }
+        }
+
+        for (i = 0; i <= MAX(63LU, (unsigned long) CAP_LAST_CAP); i++)
+                if (drop & ((uint64_t) 1ULL << (uint64_t) i)) {
+                        if (prctl(PR_CAPBSET_DROP, i) < 0) {
+                                if (errno == EINVAL)
+                                        break;
+
+                                r = -errno;
+                                goto finish;
+                        }
+                }
+
+        r = 0;
+
+finish:
+        if (new_cap)
+                cap_free(new_cap);
+
+        if (old_cap) {
+                cap_set_proc(old_cap);
+                cap_free(old_cap);
+        }
+
+        return r;
+}
+
 int exec_spawn(ExecCommand *command,
                char **argv,
                const ExecContext *context,
@@ -981,7 +997,7 @@ int exec_spawn(ExecCommand *command,
 
                 /* This string must fit in 10 chars (i.e. the length
                  * of "/sbin/init") */
-                rename_process("sd.exec");
+                rename_process("sd(EXEC)");
 
                 /* We reset exactly these signals, since they are the
                  * only ones we set to SIG_IGN in the main daemon. All
@@ -1027,6 +1043,8 @@ int exec_spawn(ExecCommand *command,
                         }
                 }
 
+                exec_context_tty_reset(context);
+
                 /* We skip the confirmation step if we shall not apply the TTY */
                 if (confirm_spawn &&
                     (!is_terminal_input(context->std_input) || apply_tty_stdin)) {
@@ -1166,24 +1184,39 @@ int exec_spawn(ExecCommand *command,
                                         r = EXIT_STDIN;
                                         goto fail_child;
                                 }
+
+                        if (cgroup_bondings && context->control_group_modify)
+                                if (cgroup_bonding_set_group_access_list(cgroup_bondings, 0755, uid, gid) < 0 ||
+                                    cgroup_bonding_set_task_access_list(cgroup_bondings, 0644, uid, gid) < 0) {
+                                        r = EXIT_CGROUP;
+                                        goto fail_child;
+                                }
                 }
 
+                if (apply_permissions)
+                        if (enforce_groups(context, username, gid) < 0) {
+                                r = EXIT_GROUP;
+                                goto fail_child;
+                        }
+
+                umask(context->umask);
+
 #ifdef HAVE_PAM
                 if (context->pam_name && username) {
-                        if (setup_pam(context->pam_name, username, context->tty_path, &pam_env, fds, n_fds) < 0) {
+                        if (setup_pam(context->pam_name, username, context->tty_path, &pam_env, fds, n_fds) != 0) {
                                 r = EXIT_PAM;
                                 goto fail_child;
                         }
                 }
 #endif
-
-                if (apply_permissions)
-                        if (enforce_groups(context, username, uid) < 0) {
-                                r = EXIT_GROUP;
+                if (context->private_network) {
+                        if (unshare(CLONE_NEWNET) < 0) {
+                                r = EXIT_NETWORK;
                                 goto fail_child;
                         }
 
-                umask(context->umask);
+                        loopback_setup();
+                }
 
                 if (strv_length(context->read_write_dirs) > 0 ||
                     strv_length(context->read_only_dirs) > 0 ||
@@ -1251,13 +1284,10 @@ int exec_spawn(ExecCommand *command,
                         }
 
                         if (context->capability_bounding_set_drop)
-                                for (i = 0; i <= CAP_LAST_CAP; i++)
-                                        if (context->capability_bounding_set_drop & ((uint64_t) 1ULL << (uint64_t) i)) {
-                                                if (prctl(PR_CAPBSET_DROP, i) < 0) {
-                                                        r = EXIT_CAPABILITIES;
-                                                        goto fail_child;
-                                                }
-                                        }
+                                if (do_capability_bounding_set_drop(context->capability_bounding_set_drop) < 0) {
+                                        r = EXIT_CAPABILITIES;
+                                        goto fail_child;
+                                }
 
                         if (context->user)
                                 if (enforce_user(context, uid) < 0) {
@@ -1381,7 +1411,7 @@ fail_parent:
 void exec_context_init(ExecContext *c) {
         assert(c);
 
-        c->umask = 0002;
+        c->umask = 0022;
         c->ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 0);
         c->cpu_sched_policy = SCHED_OTHER;
         c->syslog_priority = LOG_DAEMON|LOG_INFO;
@@ -1572,12 +1602,16 @@ void exec_context_dump(ExecContext *c, FILE* f, const char *prefix) {
                 "%sWorkingDirectory: %s\n"
                 "%sRootDirectory: %s\n"
                 "%sNonBlocking: %s\n"
-                "%sPrivateTmp: %s\n",
+                "%sPrivateTmp: %s\n"
+                "%sControlGroupModify: %s\n"
+                "%sPrivateNetwork: %s\n",
                 prefix, c->umask,
                 prefix, c->working_directory ? c->working_directory : "/",
                 prefix, c->root_directory ? c->root_directory : "/",
                 prefix, yes_no(c->non_blocking),
-                prefix, yes_no(c->private_tmp));
+                prefix, yes_no(c->private_tmp),
+                prefix, yes_no(c->control_group_modify),
+                prefix, yes_no(c->private_network));
 
         STRV_FOREACH(e, c->environment)
                 fprintf(f, "%sEnvironment: %s\n", prefix, *e);
@@ -1641,8 +1675,14 @@ void exec_context_dump(ExecContext *c, FILE* f, const char *prefix) {
 
         if (c->tty_path)
                 fprintf(f,
-                        "%sTTYPath: %s\n",
-                        prefix, c->tty_path);
+                        "%sTTYPath: %s\n"
+                        "%sTTYReset: %s\n"
+                        "%sTTYVHangup: %s\n"
+                        "%sTTYVTDisallocate: %s\n",
+                        prefix, c->tty_path,
+                        prefix, yes_no(c->tty_reset),
+                        prefix, yes_no(c->tty_vhangup),
+                        prefix, yes_no(c->tty_vt_disallocate));
 
         if (c->std_output == EXEC_OUTPUT_SYSLOG || c->std_output == EXEC_OUTPUT_KMSG ||
             c->std_output == EXEC_OUTPUT_SYSLOG_AND_CONSOLE || c->std_output == EXEC_OUTPUT_KMSG_AND_CONSOLE ||
@@ -1674,13 +1714,14 @@ void exec_context_dump(ExecContext *c, FILE* f, const char *prefix) {
                         (c->secure_bits & SECURE_NOROOT_LOCKED) ? "noroot-locked" : "");
 
         if (c->capability_bounding_set_drop) {
+                unsigned long l;
                 fprintf(f, "%sCapabilityBoundingSet:", prefix);
 
-                for (i = 0; i <= CAP_LAST_CAP; i++)
-                        if (!(c->capability_bounding_set_drop & ((uint64_t) 1ULL << (uint64_t) i))) {
+                for (l = 0; l <= (unsigned long) CAP_LAST_CAP; l++)
+                        if (!(c->capability_bounding_set_drop & ((uint64_t) 1ULL << (uint64_t) l))) {
                                 char *t;
 
-                                if ((t = cap_to_name(i))) {
+                                if ((t = cap_to_name(l))) {
                                         fprintf(f, " %s", t);
                                         cap_free(t);
                                 }
@@ -1743,7 +1784,7 @@ void exec_status_start(ExecStatus *s, pid_t pid) {
         dual_timestamp_get(&s->start_timestamp);
 }
 
-void exec_status_exit(ExecStatus *s, pid_t pid, int code, int status, const char *utmp_id) {
+void exec_status_exit(ExecStatus *s, ExecContext *context, pid_t pid, int code, int status) {
         assert(s);
 
         if ((s->pid && s->pid != pid) ||
@@ -1756,8 +1797,12 @@ void exec_status_exit(ExecStatus *s, pid_t pid, int code, int status, const char
         s->code = code;
         s->status = status;
 
-        if (utmp_id)
-                utmp_put_dead_process(utmp_id, pid, code, status);
+        if (context) {
+                if (context->utmp_id)
+                        utmp_put_dead_process(context->utmp_id, pid, code, status);
+
+                exec_context_tty_reset(context);
+        }
 }
 
 void exec_status_dump(ExecStatus *s, FILE *f, const char *prefix) {