chiark / gitweb /
Remove src/core
[elogind.git] / src / core / execute.c
diff --git a/src/core/execute.c b/src/core/execute.c
deleted file mode 100644 (file)
index 027e331..0000000
+++ /dev/null
@@ -1,2912 +0,0 @@
-/*-*- Mode: C; c-basic-offset: 8; indent-tabs-mode: nil -*-*/
-
-/***
-  This file is part of systemd.
-
-  Copyright 2010 Lennart Poettering
-
-  systemd is free software; you can redistribute it and/or modify it
-  under the terms of the GNU Lesser General Public License as published by
-  the Free Software Foundation; either version 2.1 of the License, or
-  (at your option) any later version.
-
-  systemd is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
-  Lesser General Public License for more details.
-
-  You should have received a copy of the GNU Lesser General Public License
-  along with systemd; If not, see <http://www.gnu.org/licenses/>.
-***/
-
-#include <errno.h>
-#include <fcntl.h>
-#include <unistd.h>
-#include <string.h>
-#include <signal.h>
-#include <sys/socket.h>
-#include <sys/un.h>
-#include <sys/prctl.h>
-#include <sys/stat.h>
-#include <grp.h>
-#include <poll.h>
-#include <glob.h>
-#include <sys/personality.h>
-
-#ifdef HAVE_PAM
-#include <security/pam_appl.h>
-#endif
-
-#ifdef HAVE_SELINUX
-#include <selinux/selinux.h>
-#endif
-
-#ifdef HAVE_SECCOMP
-#include <seccomp.h>
-#endif
-
-#ifdef HAVE_APPARMOR
-#include <sys/apparmor.h>
-#endif
-
-#include "execute.h"
-#include "strv.h"
-#include "macro.h"
-#include "capability.h"
-#include "util.h"
-#include "log.h"
-#include "sd-messages.h"
-#include "ioprio.h"
-#include "securebits.h"
-#include "namespace.h"
-#include "exit-status.h"
-#include "missing.h"
-#include "def.h"
-#include "path-util.h"
-#include "env-util.h"
-#include "fileio.h"
-#include "unit.h"
-#include "async.h"
-#include "selinux-util.h"
-#include "errno-list.h"
-#include "af-list.h"
-#include "mkdir.h"
-#include "smack-util.h"
-#include "bus-endpoint.h"
-#include "cap-list.h"
-
-#ifdef HAVE_APPARMOR
-#include "apparmor-util.h"
-#endif
-
-#ifdef HAVE_SECCOMP
-#include "seccomp-util.h"
-#endif
-
-#define IDLE_TIMEOUT_USEC (5*USEC_PER_SEC)
-#define IDLE_TIMEOUT2_USEC (1*USEC_PER_SEC)
-
-/* This assumes there is a 'tty' group */
-#define TTY_MODE 0620
-
-#define SNDBUF_SIZE (8*1024*1024)
-
-static int shift_fds(int fds[], unsigned n_fds) {
-        int start, restart_from;
-
-        if (n_fds <= 0)
-                return 0;
-
-        /* Modifies the fds array! (sorts it) */
-
-        assert(fds);
-
-        start = 0;
-        for (;;) {
-                int i;
-
-                restart_from = -1;
-
-                for (i = start; i < (int) n_fds; i++) {
-                        int nfd;
-
-                        /* Already at right index? */
-                        if (fds[i] == i+3)
-                                continue;
-
-                        if ((nfd = fcntl(fds[i], F_DUPFD, i+3)) < 0)
-                                return -errno;
-
-                        safe_close(fds[i]);
-                        fds[i] = nfd;
-
-                        /* Hmm, the fd we wanted isn't free? Then
-                         * let's remember that and try again from here */
-                        if (nfd != i+3 && restart_from < 0)
-                                restart_from = i;
-                }
-
-                if (restart_from < 0)
-                        break;
-
-                start = restart_from;
-        }
-
-        return 0;
-}
-
-static int flags_fds(const int fds[], unsigned n_fds, bool nonblock) {
-        unsigned i;
-        int r;
-
-        if (n_fds <= 0)
-                return 0;
-
-        assert(fds);
-
-        /* Drops/Sets O_NONBLOCK and FD_CLOEXEC from the file flags */
-
-        for (i = 0; i < n_fds; i++) {
-
-                if ((r = fd_nonblock(fds[i], nonblock)) < 0)
-                        return r;
-
-                /* We unconditionally drop FD_CLOEXEC from the fds,
-                 * since after all we want to pass these fds to our
-                 * children */
-
-                if ((r = fd_cloexec(fds[i], false)) < 0)
-                        return r;
-        }
-
-        return 0;
-}
-
-_pure_ static const char *tty_path(const ExecContext *context) {
-        assert(context);
-
-        if (context->tty_path)
-                return context->tty_path;
-
-        return "/dev/console";
-}
-
-static 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 bool is_terminal_output(ExecOutput o) {
-        return
-                o == EXEC_OUTPUT_TTY ||
-                o == EXEC_OUTPUT_SYSLOG_AND_CONSOLE ||
-                o == EXEC_OUTPUT_KMSG_AND_CONSOLE ||
-                o == EXEC_OUTPUT_JOURNAL_AND_CONSOLE;
-}
-
-static int open_null_as(int flags, int nfd) {
-        int fd, r;
-
-        assert(nfd >= 0);
-
-        fd = open("/dev/null", flags|O_NOCTTY);
-        if (fd < 0)
-                return -errno;
-
-        if (fd != nfd) {
-                r = dup2(fd, nfd) < 0 ? -errno : nfd;
-                safe_close(fd);
-        } else
-                r = nfd;
-
-        return r;
-}
-
-static int connect_journal_socket(int fd, uid_t uid, gid_t gid) {
-        union sockaddr_union sa = {
-                .un.sun_family = AF_UNIX,
-                .un.sun_path = "/run/systemd/journal/stdout",
-        };
-        uid_t olduid = UID_INVALID;
-        gid_t oldgid = GID_INVALID;
-        int r;
-
-        if (gid != GID_INVALID) {
-                oldgid = getgid();
-
-                r = setegid(gid);
-                if (r < 0)
-                        return -errno;
-        }
-
-        if (uid != UID_INVALID) {
-                olduid = getuid();
-
-                r = seteuid(uid);
-                if (r < 0) {
-                        r = -errno;
-                        goto restore_gid;
-                }
-        }
-
-        r = connect(fd, &sa.sa, offsetof(struct sockaddr_un, sun_path) + strlen(sa.un.sun_path));
-        if (r < 0)
-                r = -errno;
-
-        /* If we fail to restore the uid or gid, things will likely
-           fail later on. This should only happen if an LSM interferes. */
-
-        if (uid != UID_INVALID)
-                (void) seteuid(olduid);
-
- restore_gid:
-        if (gid != GID_INVALID)
-                (void) setegid(oldgid);
-
-        return r;
-}
-
-static int connect_logger_as(const ExecContext *context, ExecOutput output, const char *ident, const char *unit_id, int nfd, uid_t uid, gid_t gid) {
-        int fd, r;
-
-        assert(context);
-        assert(output < _EXEC_OUTPUT_MAX);
-        assert(ident);
-        assert(nfd >= 0);
-
-        fd = socket(AF_UNIX, SOCK_STREAM, 0);
-        if (fd < 0)
-                return -errno;
-
-        r = connect_journal_socket(fd, uid, gid);
-        if (r < 0)
-                return r;
-
-        if (shutdown(fd, SHUT_RD) < 0) {
-                safe_close(fd);
-                return -errno;
-        }
-
-        fd_inc_sndbuf(fd, SNDBUF_SIZE);
-
-        dprintf(fd,
-                "%s\n"
-                "%s\n"
-                "%i\n"
-                "%i\n"
-                "%i\n"
-                "%i\n"
-                "%i\n",
-                context->syslog_identifier ? context->syslog_identifier : ident,
-                unit_id,
-                context->syslog_priority,
-                !!context->syslog_level_prefix,
-                output == EXEC_OUTPUT_SYSLOG || output == EXEC_OUTPUT_SYSLOG_AND_CONSOLE,
-                output == EXEC_OUTPUT_KMSG || output == EXEC_OUTPUT_KMSG_AND_CONSOLE,
-                is_terminal_output(output));
-
-        if (fd != nfd) {
-                r = dup2(fd, nfd) < 0 ? -errno : nfd;
-                safe_close(fd);
-        } else
-                r = nfd;
-
-        return r;
-}
-static int open_terminal_as(const char *path, mode_t mode, int nfd) {
-        int fd, r;
-
-        assert(path);
-        assert(nfd >= 0);
-
-        if ((fd = open_terminal(path, mode | O_NOCTTY)) < 0)
-                return fd;
-
-        if (fd != nfd) {
-                r = dup2(fd, nfd) < 0 ? -errno : nfd;
-                safe_close(fd);
-        } else
-                r = nfd;
-
-        return r;
-}
-
-static bool is_terminal_input(ExecInput i) {
-        return
-                i == EXEC_INPUT_TTY ||
-                i == EXEC_INPUT_TTY_FORCE ||
-                i == EXEC_INPUT_TTY_FAIL;
-}
-
-static int fixup_input(ExecInput std_input, int socket_fd, bool apply_tty_stdin) {
-
-        if (is_terminal_input(std_input) && !apply_tty_stdin)
-                return EXEC_INPUT_NULL;
-
-        if (std_input == EXEC_INPUT_SOCKET && socket_fd < 0)
-                return EXEC_INPUT_NULL;
-
-        return std_input;
-}
-
-static int fixup_output(ExecOutput std_output, int socket_fd) {
-
-        if (std_output == EXEC_OUTPUT_SOCKET && socket_fd < 0)
-                return EXEC_OUTPUT_INHERIT;
-
-        return std_output;
-}
-
-static int setup_input(const ExecContext *context, int socket_fd, bool apply_tty_stdin) {
-        ExecInput i;
-
-        assert(context);
-
-        i = fixup_input(context->std_input, socket_fd, apply_tty_stdin);
-
-        switch (i) {
-
-        case EXEC_INPUT_NULL:
-                return open_null_as(O_RDONLY, STDIN_FILENO);
-
-        case EXEC_INPUT_TTY:
-        case EXEC_INPUT_TTY_FORCE:
-        case EXEC_INPUT_TTY_FAIL: {
-                int fd, r;
-
-                fd = acquire_terminal(tty_path(context),
-                                      i == EXEC_INPUT_TTY_FAIL,
-                                      i == EXEC_INPUT_TTY_FORCE,
-                                      false,
-                                      USEC_INFINITY);
-                if (fd < 0)
-                        return fd;
-
-                if (fd != STDIN_FILENO) {
-                        r = dup2(fd, STDIN_FILENO) < 0 ? -errno : STDIN_FILENO;
-                        safe_close(fd);
-                } else
-                        r = STDIN_FILENO;
-
-                return r;
-        }
-
-        case EXEC_INPUT_SOCKET:
-                return dup2(socket_fd, STDIN_FILENO) < 0 ? -errno : STDIN_FILENO;
-
-        default:
-                assert_not_reached("Unknown input type");
-        }
-}
-
-static int setup_output(const ExecContext *context, int fileno, int socket_fd, const char *ident, const char *unit_id, bool apply_tty_stdin, uid_t uid, gid_t gid) {
-        ExecOutput o;
-        ExecInput i;
-        int r;
-
-        assert(context);
-        assert(ident);
-
-        i = fixup_input(context->std_input, socket_fd, apply_tty_stdin);
-        o = fixup_output(context->std_output, socket_fd);
-
-        if (fileno == STDERR_FILENO) {
-                ExecOutput e;
-                e = fixup_output(context->std_error, socket_fd);
-
-                /* This expects the input and output are already set up */
-
-                /* Don't change the stderr file descriptor if we inherit all
-                 * the way and are not on a tty */
-                if (e == EXEC_OUTPUT_INHERIT &&
-                    o == EXEC_OUTPUT_INHERIT &&
-                    i == EXEC_INPUT_NULL &&
-                    !is_terminal_input(context->std_input) &&
-                    getppid () != 1)
-                        return fileno;
-
-                /* Duplicate from stdout if possible */
-                if (e == o || e == EXEC_OUTPUT_INHERIT)
-                        return dup2(STDOUT_FILENO, fileno) < 0 ? -errno : fileno;
-
-                o = e;
-
-        } else if (o == EXEC_OUTPUT_INHERIT) {
-                /* If input got downgraded, inherit the original value */
-                if (i == EXEC_INPUT_NULL && is_terminal_input(context->std_input))
-                        return open_terminal_as(tty_path(context), O_WRONLY, fileno);
-
-                /* If the input is connected to anything that's not a /dev/null, inherit that... */
-                if (i != EXEC_INPUT_NULL)
-                        return dup2(STDIN_FILENO, fileno) < 0 ? -errno : fileno;
-
-                /* If we are not started from PID 1 we just inherit STDOUT from our parent process. */
-                if (getppid() != 1)
-                        return fileno;
-
-                /* We need to open /dev/null here anew, to get the right access mode. */
-                return open_null_as(O_WRONLY, fileno);
-        }
-
-        switch (o) {
-
-        case EXEC_OUTPUT_NULL:
-                return open_null_as(O_WRONLY, fileno);
-
-        case EXEC_OUTPUT_TTY:
-                if (is_terminal_input(i))
-                        return dup2(STDIN_FILENO, fileno) < 0 ? -errno : fileno;
-
-                /* We don't reset the terminal if this is just about output */
-                return open_terminal_as(tty_path(context), O_WRONLY, fileno);
-
-        case EXEC_OUTPUT_SYSLOG:
-        case EXEC_OUTPUT_SYSLOG_AND_CONSOLE:
-        case EXEC_OUTPUT_KMSG:
-        case EXEC_OUTPUT_KMSG_AND_CONSOLE:
-        case EXEC_OUTPUT_JOURNAL:
-        case EXEC_OUTPUT_JOURNAL_AND_CONSOLE:
-                r = connect_logger_as(context, o, ident, unit_id, fileno, uid, gid);
-                if (r < 0) {
-                        log_unit_struct(unit_id,
-                                        LOG_ERR,
-                                        LOG_MESSAGE("Failed to connect %s of %s to the journal socket: %s",
-                                                    fileno == STDOUT_FILENO ? "stdout" : "stderr",
-                                                    unit_id, strerror(-r)),
-                                        LOG_ERRNO(-r),
-                                        NULL);
-                        r = open_null_as(O_WRONLY, fileno);
-                }
-                return r;
-
-        case EXEC_OUTPUT_SOCKET:
-                assert(socket_fd >= 0);
-                return dup2(socket_fd, fileno) < 0 ? -errno : fileno;
-
-        default:
-                assert_not_reached("Unknown error type");
-        }
-}
-
-static int chown_terminal(int fd, uid_t uid) {
-        struct stat st;
-
-        assert(fd >= 0);
-
-        /* This might fail. What matters are the results. */
-        (void) fchown(fd, uid, -1);
-        (void) fchmod(fd, TTY_MODE);
-
-        if (fstat(fd, &st) < 0)
-                return -errno;
-
-        if (st.st_uid != uid || (st.st_mode & 0777) != TTY_MODE)
-                return -EPERM;
-
-        return 0;
-}
-
-static int setup_confirm_stdio(int *_saved_stdin,
-                               int *_saved_stdout) {
-        int fd = -1, saved_stdin, saved_stdout = -1, r;
-
-        assert(_saved_stdin);
-        assert(_saved_stdout);
-
-        saved_stdin = fcntl(STDIN_FILENO, F_DUPFD, 3);
-        if (saved_stdin < 0)
-                return -errno;
-
-        saved_stdout = fcntl(STDOUT_FILENO, F_DUPFD, 3);
-        if (saved_stdout < 0) {
-                r = errno;
-                goto fail;
-        }
-
-        fd = acquire_terminal(
-                        "/dev/console",
-                        false,
-                        false,
-                        false,
-                        DEFAULT_CONFIRM_USEC);
-        if (fd < 0) {
-                r = fd;
-                goto fail;
-        }
-
-        r = chown_terminal(fd, getuid());
-        if (r < 0)
-                goto fail;
-
-        if (dup2(fd, STDIN_FILENO) < 0) {
-                r = -errno;
-                goto fail;
-        }
-
-        if (dup2(fd, STDOUT_FILENO) < 0) {
-                r = -errno;
-                goto fail;
-        }
-
-        if (fd >= 2)
-                safe_close(fd);
-
-        *_saved_stdin = saved_stdin;
-        *_saved_stdout = saved_stdout;
-
-        return 0;
-
-fail:
-        safe_close(saved_stdout);
-        safe_close(saved_stdin);
-        safe_close(fd);
-
-        return r;
-}
-
-_printf_(1, 2) static int write_confirm_message(const char *format, ...) {
-        _cleanup_close_ int fd = -1;
-        va_list ap;
-
-        assert(format);
-
-        fd = open_terminal("/dev/console", O_WRONLY|O_NOCTTY|O_CLOEXEC);
-        if (fd < 0)
-                return fd;
-
-        va_start(ap, format);
-        vdprintf(fd, format, ap);
-        va_end(ap);
-
-        return 0;
-}
-
-static int restore_confirm_stdio(int *saved_stdin,
-                                 int *saved_stdout) {
-
-        int r = 0;
-
-        assert(saved_stdin);
-        assert(saved_stdout);
-
-        release_terminal();
-
-        if (*saved_stdin >= 0)
-                if (dup2(*saved_stdin, STDIN_FILENO) < 0)
-                        r = -errno;
-
-        if (*saved_stdout >= 0)
-                if (dup2(*saved_stdout, STDOUT_FILENO) < 0)
-                        r = -errno;
-
-        safe_close(*saved_stdin);
-        safe_close(*saved_stdout);
-
-        return r;
-}
-
-static int ask_for_confirmation(char *response, char **argv) {
-        int saved_stdout = -1, saved_stdin = -1, r;
-        _cleanup_free_ char *line = NULL;
-
-        r = setup_confirm_stdio(&saved_stdin, &saved_stdout);
-        if (r < 0)
-                return r;
-
-        line = exec_command_line(argv);
-        if (!line)
-                return -ENOMEM;
-
-        r = ask_char(response, "yns", "Execute %s? [Yes, No, Skip] ", line);
-
-        restore_confirm_stdio(&saved_stdin, &saved_stdout);
-
-        return r;
-}
-
-static int enforce_groups(const ExecContext *context, const char *username, gid_t gid) {
-        bool keep_groups = false;
-        int r;
-
-        assert(context);
-
-        /* Lookup and set GID and supplementary group list. Here too
-         * we avoid NSS lookups for gid=0. */
-
-        if (context->group || username) {
-
-                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) {
-                        if (initgroups(username, gid) < 0)
-                                return -errno;
-
-                        keep_groups = true;
-                }
-
-                /* Second step, set our gids */
-                if (setresgid(gid, gid, gid) < 0)
-                        return -errno;
-        }
-
-        if (context->supplementary_groups) {
-                int ngroups_max, k;
-                gid_t *gids;
-                char **i;
-
-                /* Final step, initialize any manually set supplementary groups */
-                assert_se((ngroups_max = (int) sysconf(_SC_NGROUPS_MAX)) > 0);
-
-                if (!(gids = new(gid_t, ngroups_max)))
-                        return -ENOMEM;
-
-                if (keep_groups) {
-                        if ((k = getgroups(ngroups_max, gids)) < 0) {
-                                free(gids);
-                                return -errno;
-                        }
-                } else
-                        k = 0;
-
-                STRV_FOREACH(i, context->supplementary_groups) {
-                        const char *g;
-
-                        if (k >= ngroups_max) {
-                                free(gids);
-                                return -E2BIG;
-                        }
-
-                        g = *i;
-                        r = get_group_creds(&g, gids+k);
-                        if (r < 0) {
-                                free(gids);
-                                return r;
-                        }
-
-                        k++;
-                }
-
-                if (setgroups(k, gids) < 0) {
-                        free(gids);
-                        return -errno;
-                }
-
-                free(gids);
-        }
-
-        return 0;
-}
-
-static int enforce_user(const ExecContext *context, uid_t uid) {
-        assert(context);
-
-        /* 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 */
-                };
-
-                /* First step: If we need to keep capabilities but
-                 * drop privileges we need to make sure we keep our
-                 * caps, while we drop privileges. */
-                if (uid != 0) {
-                        int sb = context->secure_bits | 1<<SECURE_KEEP_CAPS;
-
-                        if (prctl(PR_GET_SECUREBITS) != sb)
-                                if (prctl(PR_SET_SECUREBITS, sb) < 0)
-                                        return -errno;
-                }
-
-                /* 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 (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 */
-        if (setresuid(uid, uid, uid) < 0)
-                return -errno;
-
-        /* At this point we should have all necessary capabilities but
-           are otherwise a normal user. However, the caps might got
-           corrupted due to the setresuid() so we need clean them up
-           later. This is done outside of this call. */
-
-        return 0;
-}
-
-#ifdef HAVE_PAM
-
-static int null_conv(
-                int num_msg,
-                const struct pam_message **msg,
-                struct pam_response **resp,
-                void *appdata_ptr) {
-
-        /* We don't support conversations */
-
-        return PAM_CONV_ERR;
-}
-
-static int setup_pam(
-                const char *name,
-                const char *user,
-                uid_t uid,
-                const char *tty,
-                char ***pam_env,
-                int fds[], unsigned n_fds) {
-
-        static const struct pam_conv conv = {
-                .conv = null_conv,
-                .appdata_ptr = NULL
-        };
-
-        pam_handle_t *handle = NULL;
-        sigset_t ss, old_ss;
-        int pam_code = PAM_SUCCESS;
-        int err;
-        char **e = NULL;
-        bool close_session = false;
-        pid_t pam_pid = 0, parent_pid;
-        int flags = 0;
-
-        assert(name);
-        assert(user);
-        assert(pam_env);
-
-        /* We set up PAM in the parent process, then fork. The child
-         * will then stay around until killed via PR_GET_PDEATHSIG or
-         * systemd via the cgroup logic. It will then remove the PAM
-         * session again. The parent process will exec() the actual
-         * daemon. We do things this way to ensure that the main PID
-         * of the daemon is the one we initially fork()ed. */
-
-        if (log_get_max_level() < LOG_DEBUG)
-                flags |= PAM_SILENT;
-
-        pam_code = pam_start(name, user, &conv, &handle);
-        if (pam_code != PAM_SUCCESS) {
-                handle = NULL;
-                goto fail;
-        }
-
-        if (tty) {
-                pam_code = pam_set_item(handle, PAM_TTY, tty);
-                if (pam_code != PAM_SUCCESS)
-                        goto fail;
-        }
-
-        pam_code = pam_acct_mgmt(handle, flags);
-        if (pam_code != PAM_SUCCESS)
-                goto fail;
-
-        pam_code = pam_open_session(handle, flags);
-        if (pam_code != PAM_SUCCESS)
-                goto fail;
-
-        close_session = true;
-
-        e = pam_getenvlist(handle);
-        if (!e) {
-                pam_code = PAM_BUF_ERR;
-                goto fail;
-        }
-
-        /* Block SIGTERM, so that we know that it won't get lost in
-         * the child */
-        if (sigemptyset(&ss) < 0 ||
-            sigaddset(&ss, SIGTERM) < 0 ||
-            sigprocmask(SIG_BLOCK, &ss, &old_ss) < 0)
-                goto fail;
-
-        parent_pid = getpid();
-
-        pam_pid = fork();
-        if (pam_pid < 0)
-                goto fail;
-
-        if (pam_pid == 0) {
-                int sig;
-                int r = EXIT_PAM;
-
-                /* The child's job is to reset the PAM session on
-                 * termination */
-
-                /* This string must fit in 10 chars (i.e. the length
-                 * of "/sbin/init"), to look pretty in /bin/ps */
-                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
-                open here that have been opened by PAM. */
-                close_many(fds, n_fds);
-
-                /* Drop privileges - we don't need any to pam_close_session
-                 * and this will make PR_SET_PDEATHSIG work in most cases.
-                 * If this fails, ignore the error - but expect sd-pam threads
-                 * to fail to exit normally */
-                if (setresuid(uid, uid, uid) < 0)
-                        log_error_errno(r, "Error: Failed to setresuid() in sd-pam: %m");
-
-                /* Wait until our parent died. This will only work if
-                 * the above setresuid() succeeds, otherwise the kernel
-                 * will not allow unprivileged parents kill their privileged
-                 * children this way. We rely on the control groups kill logic
-                 * to do the rest for us. */
-                if (prctl(PR_SET_PDEATHSIG, SIGTERM) < 0)
-                        goto child_finish;
-
-                /* Check if our parent process might already have
-                 * died? */
-                if (getppid() == parent_pid) {
-                        for (;;) {
-                                if (sigwait(&ss, &sig) < 0) {
-                                        if (errno == EINTR)
-                                                continue;
-
-                                        goto child_finish;
-                                }
-
-                                assert(sig == SIGTERM);
-                                break;
-                        }
-                }
-
-                /* If our parent died we'll end the session */
-                if (getppid() != parent_pid) {
-                        pam_code = pam_close_session(handle, flags);
-                        if (pam_code != PAM_SUCCESS)
-                                goto child_finish;
-                }
-
-                r = 0;
-
-        child_finish:
-                pam_end(handle, pam_code | flags);
-                _exit(r);
-        }
-
-        /* If the child was forked off successfully it will do all the
-         * cleanups, so forget about the handle here. */
-        handle = NULL;
-
-        /* Unblock SIGTERM again in the parent */
-        if (sigprocmask(SIG_SETMASK, &old_ss, NULL) < 0)
-                goto fail;
-
-        /* We close the log explicitly here, since the PAM modules
-         * might have opened it, but we don't want this fd around. */
-        closelog();
-
-        *pam_env = e;
-        e = NULL;
-
-        return 0;
-
-fail:
-        if (pam_code != PAM_SUCCESS) {
-                log_error("PAM failed: %s", pam_strerror(handle, pam_code));
-                err = -EPERM;  /* PAM errors do not map to errno */
-        } else {
-                log_error_errno(errno, "PAM failed: %m");
-                err = -errno;
-        }
-
-        if (handle) {
-                if (close_session)
-                        pam_code = pam_close_session(handle, flags);
-
-                pam_end(handle, pam_code | flags);
-        }
-
-        strv_free(e);
-
-        closelog();
-
-        if (pam_pid > 1) {
-                kill(pam_pid, SIGTERM);
-                kill(pam_pid, SIGCONT);
-        }
-
-        return err;
-}
-#endif
-
-static void rename_process_from_path(const char *path) {
-        char process_name[11];
-        const char *p;
-        size_t l;
-
-        /* This resulting string must fit in 10 chars (i.e. the length
-         * of "/sbin/init") to look pretty in /bin/ps */
-
-        p = basename(path);
-        if (isempty(p)) {
-                rename_process("(...)");
-                return;
-        }
-
-        l = strlen(p);
-        if (l > 8) {
-                /* The end of the process name is usually more
-                 * interesting, since the first bit might just be
-                 * "systemd-" */
-                p = p + l - 8;
-                l = 8;
-        }
-
-        process_name[0] = '(';
-        memcpy(process_name+1, p, l);
-        process_name[1+l] = ')';
-        process_name[1+l+1] = 0;
-
-        rename_process(process_name);
-}
-
-#ifdef HAVE_SECCOMP
-
-static int apply_seccomp(const ExecContext *c) {
-        uint32_t negative_action, action;
-        scmp_filter_ctx *seccomp;
-        Iterator i;
-        void *id;
-        int r;
-
-        assert(c);
-
-        negative_action = c->syscall_errno == 0 ? SCMP_ACT_KILL : SCMP_ACT_ERRNO(c->syscall_errno);
-
-        seccomp = seccomp_init(c->syscall_whitelist ? negative_action : SCMP_ACT_ALLOW);
-        if (!seccomp)
-                return -ENOMEM;
-
-        if (c->syscall_archs) {
-
-                SET_FOREACH(id, c->syscall_archs, i) {
-                        r = seccomp_arch_add(seccomp, PTR_TO_UINT32(id) - 1);
-                        if (r == -EEXIST)
-                                continue;
-                        if (r < 0)
-                                goto finish;
-                }
-
-        } else {
-                r = seccomp_add_secondary_archs(seccomp);
-                if (r < 0)
-                        goto finish;
-        }
-
-        action = c->syscall_whitelist ? SCMP_ACT_ALLOW : negative_action;
-        SET_FOREACH(id, c->syscall_filter, i) {
-                r = seccomp_rule_add(seccomp, action, PTR_TO_INT(id) - 1, 0);
-                if (r < 0)
-                        goto finish;
-        }
-
-        r = seccomp_attr_set(seccomp, SCMP_FLTATR_CTL_NNP, 0);
-        if (r < 0)
-                goto finish;
-
-        r = seccomp_load(seccomp);
-
-finish:
-        seccomp_release(seccomp);
-        return r;
-}
-
-static int apply_address_families(const ExecContext *c) {
-        scmp_filter_ctx *seccomp;
-        Iterator i;
-        int r;
-
-        assert(c);
-
-        seccomp = seccomp_init(SCMP_ACT_ALLOW);
-        if (!seccomp)
-                return -ENOMEM;
-
-        r = seccomp_add_secondary_archs(seccomp);
-        if (r < 0)
-                goto finish;
-
-        if (c->address_families_whitelist) {
-                int af, first = 0, last = 0;
-                void *afp;
-
-                /* If this is a whitelist, we first block the address
-                 * families that are out of range and then everything
-                 * that is not in the set. First, we find the lowest
-                 * and highest address family in the set. */
-
-                SET_FOREACH(afp, c->address_families, i) {
-                        af = PTR_TO_INT(afp);
-
-                        if (af <= 0 || af >= af_max())
-                                continue;
-
-                        if (first == 0 || af < first)
-                                first = af;
-
-                        if (last == 0 || af > last)
-                                last = af;
-                }
-
-                assert((first == 0) == (last == 0));
-
-                if (first == 0) {
-
-                        /* No entries in the valid range, block everything */
-                        r = seccomp_rule_add(
-                                        seccomp,
-                                        SCMP_ACT_ERRNO(EPROTONOSUPPORT),
-                                        SCMP_SYS(socket),
-                                        0);
-                        if (r < 0)
-                                goto finish;
-
-                } else {
-
-                        /* Block everything below the first entry */
-                        r = seccomp_rule_add(
-                                        seccomp,
-                                        SCMP_ACT_ERRNO(EPROTONOSUPPORT),
-                                        SCMP_SYS(socket),
-                                        1,
-                                        SCMP_A0(SCMP_CMP_LT, first));
-                        if (r < 0)
-                                goto finish;
-
-                        /* Block everything above the last entry */
-                        r = seccomp_rule_add(
-                                        seccomp,
-                                        SCMP_ACT_ERRNO(EPROTONOSUPPORT),
-                                        SCMP_SYS(socket),
-                                        1,
-                                        SCMP_A0(SCMP_CMP_GT, last));
-                        if (r < 0)
-                                goto finish;
-
-                        /* Block everything between the first and last
-                         * entry */
-                        for (af = 1; af < af_max(); af++) {
-
-                                if (set_contains(c->address_families, INT_TO_PTR(af)))
-                                        continue;
-
-                                r = seccomp_rule_add(
-                                                seccomp,
-                                                SCMP_ACT_ERRNO(EPROTONOSUPPORT),
-                                                SCMP_SYS(socket),
-                                                1,
-                                                SCMP_A0(SCMP_CMP_EQ, af));
-                                if (r < 0)
-                                        goto finish;
-                        }
-                }
-
-        } else {
-                void *af;
-
-                /* If this is a blacklist, then generate one rule for
-                 * each address family that are then combined in OR
-                 * checks. */
-
-                SET_FOREACH(af, c->address_families, i) {
-
-                        r = seccomp_rule_add(
-                                        seccomp,
-                                        SCMP_ACT_ERRNO(EPROTONOSUPPORT),
-                                        SCMP_SYS(socket),
-                                        1,
-                                        SCMP_A0(SCMP_CMP_EQ, PTR_TO_INT(af)));
-                        if (r < 0)
-                                goto finish;
-                }
-        }
-
-        r = seccomp_attr_set(seccomp, SCMP_FLTATR_CTL_NNP, 0);
-        if (r < 0)
-                goto finish;
-
-        r = seccomp_load(seccomp);
-
-finish:
-        seccomp_release(seccomp);
-        return r;
-}
-
-#endif
-
-static void do_idle_pipe_dance(int idle_pipe[4]) {
-        assert(idle_pipe);
-
-
-        safe_close(idle_pipe[1]);
-        safe_close(idle_pipe[2]);
-
-        if (idle_pipe[0] >= 0) {
-                int r;
-
-                r = fd_wait_for_event(idle_pipe[0], POLLHUP, IDLE_TIMEOUT_USEC);
-
-                if (idle_pipe[3] >= 0 && r == 0 /* timeout */) {
-                        /* Signal systemd that we are bored and want to continue. */
-                        r = write(idle_pipe[3], "x", 1);
-                        if (r > 0)
-                                /* Wait for systemd to react to the signal above. */
-                                fd_wait_for_event(idle_pipe[0], POLLHUP, IDLE_TIMEOUT2_USEC);
-                }
-
-                safe_close(idle_pipe[0]);
-
-        }
-
-        safe_close(idle_pipe[3]);
-}
-
-static int build_environment(
-                const ExecContext *c,
-                unsigned n_fds,
-                usec_t watchdog_usec,
-                const char *home,
-                const char *username,
-                const char *shell,
-                char ***ret) {
-
-        _cleanup_strv_free_ char **our_env = NULL;
-        unsigned n_env = 0;
-        char *x;
-
-        assert(c);
-        assert(ret);
-
-        our_env = new0(char*, 10);
-        if (!our_env)
-                return -ENOMEM;
-
-        if (n_fds > 0) {
-                if (asprintf(&x, "LISTEN_PID="PID_FMT, getpid()) < 0)
-                        return -ENOMEM;
-                our_env[n_env++] = x;
-
-                if (asprintf(&x, "LISTEN_FDS=%u", n_fds) < 0)
-                        return -ENOMEM;
-                our_env[n_env++] = x;
-        }
-
-        if (watchdog_usec > 0) {
-                if (asprintf(&x, "WATCHDOG_PID="PID_FMT, getpid()) < 0)
-                        return -ENOMEM;
-                our_env[n_env++] = x;
-
-                if (asprintf(&x, "WATCHDOG_USEC="USEC_FMT, watchdog_usec) < 0)
-                        return -ENOMEM;
-                our_env[n_env++] = x;
-        }
-
-        if (home) {
-                x = strappend("HOME=", home);
-                if (!x)
-                        return -ENOMEM;
-                our_env[n_env++] = x;
-        }
-
-        if (username) {
-                x = strappend("LOGNAME=", username);
-                if (!x)
-                        return -ENOMEM;
-                our_env[n_env++] = x;
-
-                x = strappend("USER=", username);
-                if (!x)
-                        return -ENOMEM;
-                our_env[n_env++] = x;
-        }
-
-        if (shell) {
-                x = strappend("SHELL=", shell);
-                if (!x)
-                        return -ENOMEM;
-                our_env[n_env++] = x;
-        }
-
-        if (is_terminal_input(c->std_input) ||
-            c->std_output == EXEC_OUTPUT_TTY ||
-            c->std_error == EXEC_OUTPUT_TTY ||
-            c->tty_path) {
-
-                x = strdup(default_term_for_tty(tty_path(c)));
-                if (!x)
-                        return -ENOMEM;
-                our_env[n_env++] = x;
-        }
-
-        our_env[n_env++] = NULL;
-        assert(n_env <= 10);
-
-        *ret = our_env;
-        our_env = NULL;
-
-        return 0;
-}
-
-static int exec_child(
-                ExecCommand *command,
-                const ExecContext *context,
-                const ExecParameters *params,
-                ExecRuntime *runtime,
-                char **argv,
-                int socket_fd,
-                int *fds, unsigned n_fds,
-                char **files_env,
-                int *exit_status) {
-
-        _cleanup_strv_free_ char **our_env = NULL, **pam_env = NULL, **final_env = NULL, **final_argv = NULL;
-        _cleanup_free_ char *mac_selinux_context_net = NULL;
-        const char *username = NULL, *home = NULL, *shell = NULL;
-        unsigned n_dont_close = 0;
-        int dont_close[n_fds + 4];
-        uid_t uid = UID_INVALID;
-        gid_t gid = GID_INVALID;
-        int i, r;
-
-        assert(command);
-        assert(context);
-        assert(params);
-        assert(exit_status);
-
-        rename_process_from_path(command->path);
-
-        /* We reset exactly these signals, since they are the
-         * only ones we set to SIG_IGN in the main daemon. All
-         * others we leave untouched because we set them to
-         * SIG_DFL or a valid handler initially, both of which
-         * will be demoted to SIG_DFL. */
-        default_signals(SIGNALS_CRASH_HANDLER,
-                        SIGNALS_IGNORE, -1);
-
-        if (context->ignore_sigpipe)
-                ignore_signals(SIGPIPE, -1);
-
-        r = reset_signal_mask();
-        if (r < 0) {
-                *exit_status = EXIT_SIGNAL_MASK;
-                return r;
-        }
-
-        if (params->idle_pipe)
-                do_idle_pipe_dance(params->idle_pipe);
-
-        /* Close sockets very early to make sure we don't
-         * block init reexecution because it cannot bind its
-         * sockets */
-
-        log_forget_fds();
-
-        if (socket_fd >= 0)
-                dont_close[n_dont_close++] = socket_fd;
-        if (n_fds > 0) {
-                memcpy(dont_close + n_dont_close, fds, sizeof(int) * n_fds);
-                n_dont_close += n_fds;
-        }
-        if (params->bus_endpoint_fd >= 0)
-                dont_close[n_dont_close++] = params->bus_endpoint_fd;
-        if (runtime) {
-                if (runtime->netns_storage_socket[0] >= 0)
-                        dont_close[n_dont_close++] = runtime->netns_storage_socket[0];
-                if (runtime->netns_storage_socket[1] >= 0)
-                        dont_close[n_dont_close++] = runtime->netns_storage_socket[1];
-        }
-
-        r = close_all_fds(dont_close, n_dont_close);
-        if (r < 0) {
-                *exit_status = EXIT_FDS;
-                return r;
-        }
-
-        if (!context->same_pgrp)
-                if (setsid() < 0) {
-                        *exit_status = EXIT_SETSID;
-                        return -errno;
-                }
-
-        exec_context_tty_reset(context);
-
-        if (params->confirm_spawn) {
-                char response;
-
-                r = ask_for_confirmation(&response, argv);
-                if (r == -ETIMEDOUT)
-                        write_confirm_message("Confirmation question timed out, assuming positive response.\n");
-                else if (r < 0)
-                        write_confirm_message("Couldn't ask confirmation question, assuming positive response: %s\n", strerror(-r));
-                else if (response == 's') {
-                        write_confirm_message("Skipping execution.\n");
-                        *exit_status = EXIT_CONFIRM;
-                        return -ECANCELED;
-                } else if (response == 'n') {
-                        write_confirm_message("Failing execution.\n");
-                        *exit_status = 0;
-                        return 0;
-                }
-        }
-
-        if (context->user) {
-                username = context->user;
-                r = get_user_creds(&username, &uid, &gid, &home, &shell);
-                if (r < 0) {
-                        *exit_status = EXIT_USER;
-                        return r;
-                }
-        }
-
-        /* If a socket is connected to STDIN/STDOUT/STDERR, we
-         * must sure to drop O_NONBLOCK */
-        if (socket_fd >= 0)
-                fd_nonblock(socket_fd, false);
-
-        r = setup_input(context, socket_fd, params->apply_tty_stdin);
-        if (r < 0) {
-                *exit_status = EXIT_STDIN;
-                return r;
-        }
-
-        r = setup_output(context, STDOUT_FILENO, socket_fd, basename(command->path), params->unit_id, params->apply_tty_stdin, uid, gid);
-        if (r < 0) {
-                *exit_status = EXIT_STDOUT;
-                return r;
-        }
-
-        r = setup_output(context, STDERR_FILENO, socket_fd, basename(command->path), params->unit_id, params->apply_tty_stdin, uid, gid);
-        if (r < 0) {
-                *exit_status = EXIT_STDERR;
-                return r;
-        }
-
-        if (params->cgroup_path) {
-                r = cg_attach_everywhere(params->cgroup_supported, params->cgroup_path, 0, NULL, NULL);
-                if (r < 0) {
-                        *exit_status = EXIT_CGROUP;
-                        return r;
-                }
-        }
-
-        if (context->oom_score_adjust_set) {
-                char t[DECIMAL_STR_MAX(context->oom_score_adjust)];
-
-                /* When we can't make this change due to EPERM, then
-                 * let's silently skip over it. User namespaces
-                 * prohibit write access to this file, and we
-                 * shouldn't trip up over that. */
-
-                sprintf(t, "%i", context->oom_score_adjust);
-                r = write_string_file("/proc/self/oom_score_adj", t);
-                if (r == -EPERM || r == -EACCES) {
-                        log_open();
-                        log_unit_debug_errno(params->unit_id, r, "Failed to adjust OOM setting, assuming containerized execution, ignoring: %m");
-                        log_close();
-                } else if (r < 0) {
-                        *exit_status = EXIT_OOM_ADJUST;
-                        return -errno;
-                }
-        }
-
-        if (context->nice_set)
-                if (setpriority(PRIO_PROCESS, 0, context->nice) < 0) {
-                        *exit_status = EXIT_NICE;
-                        return -errno;
-                }
-
-        if (context->cpu_sched_set) {
-                struct sched_param param = {
-                        .sched_priority = context->cpu_sched_priority,
-                };
-
-                r = sched_setscheduler(0,
-                                       context->cpu_sched_policy |
-                                       (context->cpu_sched_reset_on_fork ?
-                                        SCHED_RESET_ON_FORK : 0),
-                                       &param);
-                if (r < 0) {
-                        *exit_status = EXIT_SETSCHEDULER;
-                        return -errno;
-                }
-        }
-
-        if (context->cpuset)
-                if (sched_setaffinity(0, CPU_ALLOC_SIZE(context->cpuset_ncpus), context->cpuset) < 0) {
-                        *exit_status = EXIT_CPUAFFINITY;
-                        return -errno;
-                }
-
-        if (context->ioprio_set)
-                if (ioprio_set(IOPRIO_WHO_PROCESS, 0, context->ioprio) < 0) {
-                        *exit_status = EXIT_IOPRIO;
-                        return -errno;
-                }
-
-        if (context->timer_slack_nsec != NSEC_INFINITY)
-                if (prctl(PR_SET_TIMERSLACK, context->timer_slack_nsec) < 0) {
-                        *exit_status = EXIT_TIMERSLACK;
-                        return -errno;
-                }
-
-        if (context->personality != 0xffffffffUL)
-                if (personality(context->personality) < 0) {
-                        *exit_status = EXIT_PERSONALITY;
-                        return -errno;
-                }
-
-        if (context->utmp_id)
-                utmp_put_init_process(context->utmp_id, getpid(), getsid(0), context->tty_path);
-
-        if (context->user && is_terminal_input(context->std_input)) {
-                r = chown_terminal(STDIN_FILENO, uid);
-                if (r < 0) {
-                        *exit_status = EXIT_STDIN;
-                        return r;
-                }
-        }
-
-#ifdef ENABLE_KDBUS
-        if (params->bus_endpoint_fd >= 0 && context->bus_endpoint) {
-                uid_t ep_uid = (uid == UID_INVALID) ? 0 : uid;
-
-                r = bus_kernel_set_endpoint_policy(params->bus_endpoint_fd, ep_uid, context->bus_endpoint);
-                if (r < 0) {
-                        *exit_status = EXIT_BUS_ENDPOINT;
-                        return r;
-                }
-        }
-#endif
-
-        /* If delegation is enabled we'll pass ownership of the cgroup
-         * (but only in systemd's own controller hierarchy!) to the
-         * user of the new process. */
-        if (params->cgroup_path && context->user && params->cgroup_delegate) {
-                r = cg_set_task_access(SYSTEMD_CGROUP_CONTROLLER, params->cgroup_path, 0644, uid, gid);
-                if (r < 0) {
-                        *exit_status = EXIT_CGROUP;
-                        return r;
-                }
-
-
-                r = cg_set_group_access(SYSTEMD_CGROUP_CONTROLLER, params->cgroup_path, 0755, uid, gid);
-                if (r < 0) {
-                        *exit_status = EXIT_CGROUP;
-                        return r;
-                }
-        }
-
-        if (!strv_isempty(context->runtime_directory) && params->runtime_prefix) {
-                char **rt;
-
-                STRV_FOREACH(rt, context->runtime_directory) {
-                        _cleanup_free_ char *p;
-
-                        p = strjoin(params->runtime_prefix, "/", *rt, NULL);
-                        if (!p) {
-                                *exit_status = EXIT_RUNTIME_DIRECTORY;
-                                return -ENOMEM;
-                        }
-
-                        r = mkdir_safe_label(p, context->runtime_directory_mode, uid, gid);
-                        if (r < 0) {
-                                *exit_status = EXIT_RUNTIME_DIRECTORY;
-                                return r;
-                        }
-                }
-        }
-
-        if (params->apply_permissions) {
-                r = enforce_groups(context, username, gid);
-                if (r < 0) {
-                        *exit_status = EXIT_GROUP;
-                        return r;
-                }
-        }
-
-        umask(context->umask);
-
-#ifdef HAVE_PAM
-        if (params->apply_permissions && context->pam_name && username) {
-                r = setup_pam(context->pam_name, username, uid, context->tty_path, &pam_env, fds, n_fds);
-                if (r < 0) {
-                        *exit_status = EXIT_PAM;
-                        return r;
-                }
-        }
-#endif
-
-        if (context->private_network && runtime && runtime->netns_storage_socket[0] >= 0) {
-                r = setup_netns(runtime->netns_storage_socket);
-                if (r < 0) {
-                        *exit_status = EXIT_NETWORK;
-                        return r;
-                }
-        }
-
-        if (!strv_isempty(context->read_write_dirs) ||
-            !strv_isempty(context->read_only_dirs) ||
-            !strv_isempty(context->inaccessible_dirs) ||
-            context->mount_flags != 0 ||
-            (context->private_tmp && runtime && (runtime->tmp_dir || runtime->var_tmp_dir)) ||
-            params->bus_endpoint_path ||
-            context->private_devices ||
-            context->protect_system != PROTECT_SYSTEM_NO ||
-            context->protect_home != PROTECT_HOME_NO) {
-
-                char *tmp = NULL, *var = NULL;
-
-                /* The runtime struct only contains the parent
-                 * of the private /tmp, which is
-                 * non-accessible to world users. Inside of it
-                 * there's a /tmp that is sticky, and that's
-                 * the one we want to use here. */
-
-                if (context->private_tmp && runtime) {
-                        if (runtime->tmp_dir)
-                                tmp = strjoina(runtime->tmp_dir, "/tmp");
-                        if (runtime->var_tmp_dir)
-                                var = strjoina(runtime->var_tmp_dir, "/tmp");
-                }
-
-                r = setup_namespace(
-                                context->read_write_dirs,
-                                context->read_only_dirs,
-                                context->inaccessible_dirs,
-                                tmp,
-                                var,
-                                params->bus_endpoint_path,
-                                context->private_devices,
-                                context->protect_home,
-                                context->protect_system,
-                                context->mount_flags);
-
-                /* If we couldn't set up the namespace this is
-                 * probably due to a missing capability. In this case,
-                 * silently proceeed. */
-                if (r == -EPERM || r == -EACCES) {
-                        log_open();
-                        log_unit_debug_errno(params->unit_id, r, "Failed to set up namespace, assuming containerized execution, ignoring: %m");
-                        log_close();
-                } else if (r < 0) {
-                        *exit_status = EXIT_NAMESPACE;
-                        return r;
-                }
-        }
-
-        if (params->apply_chroot) {
-                if (context->root_directory)
-                        if (chroot(context->root_directory) < 0) {
-                                *exit_status = EXIT_CHROOT;
-                                return -errno;
-                        }
-
-                if (chdir(context->working_directory ?: "/") < 0 &&
-                    !context->working_directory_missing_ok) {
-                        *exit_status = EXIT_CHDIR;
-                        return -errno;
-                }
-        } else {
-                _cleanup_free_ char *d = NULL;
-
-                if (asprintf(&d, "%s/%s",
-                             context->root_directory ?: "",
-                             context->working_directory ?: "") < 0) {
-                        *exit_status = EXIT_MEMORY;
-                        return -ENOMEM;
-                }
-
-                if (chdir(d) < 0 &&
-                    !context->working_directory_missing_ok) {
-                        *exit_status = EXIT_CHDIR;
-                        return -errno;
-                }
-        }
-
-#ifdef HAVE_SELINUX
-        if (params->apply_permissions && mac_selinux_use() && params->selinux_context_net && socket_fd >= 0) {
-                r = mac_selinux_get_child_mls_label(socket_fd, command->path, context->selinux_context, &mac_selinux_context_net);
-                if (r < 0) {
-                        *exit_status = EXIT_SELINUX_CONTEXT;
-                        return r;
-                }
-        }
-#endif
-
-        /* We repeat the fd closing here, to make sure that
-         * nothing is leaked from the PAM modules. Note that
-         * we are more aggressive this time since socket_fd
-         * and the netns fds we don't need anymore. The custom
-         * endpoint fd was needed to upload the policy and can
-         * now be closed as well. */
-        r = close_all_fds(fds, n_fds);
-        if (r >= 0)
-                r = shift_fds(fds, n_fds);
-        if (r >= 0)
-                r = flags_fds(fds, n_fds, context->non_blocking);
-        if (r < 0) {
-                *exit_status = EXIT_FDS;
-                return r;
-        }
-
-        if (params->apply_permissions) {
-
-                for (i = 0; i < _RLIMIT_MAX; i++) {
-                        if (!context->rlimit[i])
-                                continue;
-
-                        if (setrlimit_closest(i, context->rlimit[i]) < 0) {
-                                *exit_status = EXIT_LIMITS;
-                                return -errno;
-                        }
-                }
-
-                if (context->capability_bounding_set_drop) {
-                        r = capability_bounding_set_drop(context->capability_bounding_set_drop, false);
-                        if (r < 0) {
-                                *exit_status = EXIT_CAPABILITIES;
-                                return r;
-                        }
-                }
-
-#ifdef HAVE_SMACK
-                if (context->smack_process_label) {
-                        r = mac_smack_apply_pid(0, context->smack_process_label);
-                        if (r < 0) {
-                                *exit_status = EXIT_SMACK_PROCESS_LABEL;
-                                return r;
-                        }
-                }
-#endif
-
-                if (context->user) {
-                        r = enforce_user(context, uid);
-                        if (r < 0) {
-                                *exit_status = EXIT_USER;
-                                return r;
-                        }
-                }
-
-                /* 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) {
-                                *exit_status = EXIT_SECUREBITS;
-                                return -errno;
-                        }
-
-                if (context->capabilities)
-                        if (cap_set_proc(context->capabilities) < 0) {
-                                *exit_status = EXIT_CAPABILITIES;
-                                return -errno;
-                        }
-
-                if (context->no_new_privileges)
-                        if (prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0) < 0) {
-                                *exit_status = EXIT_NO_NEW_PRIVILEGES;
-                                return -errno;
-                        }
-
-#ifdef HAVE_SECCOMP
-                if (context->address_families_whitelist ||
-                    !set_isempty(context->address_families)) {
-                        r = apply_address_families(context);
-                        if (r < 0) {
-                                *exit_status = EXIT_ADDRESS_FAMILIES;
-                                return r;
-                        }
-                }
-
-                if (context->syscall_whitelist ||
-                    !set_isempty(context->syscall_filter) ||
-                    !set_isempty(context->syscall_archs)) {
-                        r = apply_seccomp(context);
-                        if (r < 0) {
-                                *exit_status = EXIT_SECCOMP;
-                                return r;
-                        }
-                }
-#endif
-
-#ifdef HAVE_SELINUX
-                if (mac_selinux_use()) {
-                        char *exec_context = mac_selinux_context_net ?: context->selinux_context;
-
-                        if (exec_context) {
-                                r = setexeccon(exec_context);
-                                if (r < 0) {
-                                        *exit_status = EXIT_SELINUX_CONTEXT;
-                                        return r;
-                                }
-                        }
-                }
-#endif
-
-#ifdef HAVE_APPARMOR
-                if (context->apparmor_profile && mac_apparmor_use()) {
-                        r = aa_change_onexec(context->apparmor_profile);
-                        if (r < 0 && !context->apparmor_profile_ignore) {
-                                *exit_status = EXIT_APPARMOR_PROFILE;
-                                return -errno;
-                        }
-                }
-#endif
-        }
-
-        r = build_environment(context, n_fds, params->watchdog_usec, home, username, shell, &our_env);
-        if (r < 0) {
-                *exit_status = EXIT_MEMORY;
-                return r;
-        }
-
-        final_env = strv_env_merge(5,
-                                   params->environment,
-                                   our_env,
-                                   context->environment,
-                                   files_env,
-                                   pam_env,
-                                   NULL);
-        if (!final_env) {
-                *exit_status = EXIT_MEMORY;
-                return -ENOMEM;
-        }
-
-        final_argv = replace_env_argv(argv, final_env);
-        if (!final_argv) {
-                *exit_status = EXIT_MEMORY;
-                return -ENOMEM;
-        }
-
-        final_env = strv_env_clean(final_env);
-
-        if (_unlikely_(log_get_max_level() >= LOG_DEBUG)) {
-                _cleanup_free_ char *line;
-
-                line = exec_command_line(final_argv);
-                if (line) {
-                        log_open();
-                        log_unit_struct(params->unit_id,
-                                        LOG_DEBUG,
-                                        "EXECUTABLE=%s", command->path,
-                                        LOG_MESSAGE("Executing: %s", line),
-                                        NULL);
-                        log_close();
-                }
-        }
-        execve(command->path, final_argv, final_env);
-        *exit_status = EXIT_EXEC;
-        return -errno;
-}
-
-int exec_spawn(ExecCommand *command,
-               const ExecContext *context,
-               const ExecParameters *params,
-               ExecRuntime *runtime,
-               pid_t *ret) {
-
-        _cleanup_strv_free_ char **files_env = NULL;
-        int *fds = NULL; unsigned n_fds = 0;
-        _cleanup_free_ char *line = NULL;
-        int socket_fd, r;
-        char **argv;
-        pid_t pid;
-
-        assert(command);
-        assert(context);
-        assert(ret);
-        assert(params);
-        assert(params->fds || params->n_fds <= 0);
-
-        if (context->std_input == EXEC_INPUT_SOCKET ||
-            context->std_output == EXEC_OUTPUT_SOCKET ||
-            context->std_error == EXEC_OUTPUT_SOCKET) {
-
-                if (params->n_fds != 1) {
-                        log_unit_error(params->unit_id, "Got more than one socket.");
-                        return -EINVAL;
-                }
-
-                socket_fd = params->fds[0];
-        } else {
-                socket_fd = -1;
-                fds = params->fds;
-                n_fds = params->n_fds;
-        }
-
-        r = exec_context_load_environment(context, params->unit_id, &files_env);
-        if (r < 0)
-                return log_unit_error_errno(params->unit_id, r, "Failed to load environment files: %m");
-
-        argv = params->argv ?: command->argv;
-        line = exec_command_line(argv);
-        if (!line)
-                return log_oom();
-
-        log_unit_struct(params->unit_id,
-                        LOG_DEBUG,
-                        "EXECUTABLE=%s", command->path,
-                        LOG_MESSAGE("About to execute: %s", line),
-                        NULL);
-        pid = fork();
-        if (pid < 0)
-                return log_unit_error_errno(params->unit_id, r, "Failed to fork: %m");
-
-        if (pid == 0) {
-                int exit_status;
-
-                r = exec_child(command,
-                               context,
-                               params,
-                               runtime,
-                               argv,
-                               socket_fd,
-                               fds, n_fds,
-                               files_env,
-                               &exit_status);
-                if (r < 0) {
-                        log_open();
-                        log_unit_struct(params->unit_id,
-                                        LOG_ERR,
-                                        LOG_MESSAGE_ID(SD_MESSAGE_SPAWN_FAILED),
-                                        "EXECUTABLE=%s", command->path,
-                                        LOG_MESSAGE("Failed at step %s spawning %s: %s",
-                                                    exit_status_to_string(exit_status, EXIT_STATUS_SYSTEMD),
-                                                    command->path, strerror(-r)),
-                                        LOG_ERRNO(r),
-                                        NULL);
-                }
-
-                _exit(exit_status);
-        }
-
-        log_unit_debug(params->unit_id, "Forked %s as "PID_FMT, command->path, pid);
-
-        /* We add the new process to the cgroup both in the child (so
-         * that we can be sure that no user code is ever executed
-         * outside of the cgroup) and in the parent (so that we can be
-         * sure that when we kill the cgroup the process will be
-         * killed too). */
-        if (params->cgroup_path)
-                cg_attach(SYSTEMD_CGROUP_CONTROLLER, params->cgroup_path, pid);
-
-        exec_status_start(&command->exec_status, pid);
-
-        *ret = pid;
-        return 0;
-}
-
-void exec_context_init(ExecContext *c) {
-        assert(c);
-
-        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;
-        c->syslog_level_prefix = true;
-        c->ignore_sigpipe = true;
-        c->timer_slack_nsec = NSEC_INFINITY;
-        c->personality = 0xffffffffUL;
-        c->runtime_directory_mode = 0755;
-}
-
-void exec_context_done(ExecContext *c) {
-        unsigned l;
-
-        assert(c);
-
-        strv_free(c->environment);
-        c->environment = NULL;
-
-        strv_free(c->environment_files);
-        c->environment_files = NULL;
-
-        for (l = 0; l < ELEMENTSOF(c->rlimit); l++) {
-                free(c->rlimit[l]);
-                c->rlimit[l] = NULL;
-        }
-
-        free(c->working_directory);
-        c->working_directory = NULL;
-        free(c->root_directory);
-        c->root_directory = NULL;
-
-        free(c->tty_path);
-        c->tty_path = NULL;
-
-        free(c->syslog_identifier);
-        c->syslog_identifier = NULL;
-
-        free(c->user);
-        c->user = NULL;
-
-        free(c->group);
-        c->group = NULL;
-
-        strv_free(c->supplementary_groups);
-        c->supplementary_groups = NULL;
-
-        free(c->pam_name);
-        c->pam_name = NULL;
-
-        if (c->capabilities) {
-                cap_free(c->capabilities);
-                c->capabilities = NULL;
-        }
-
-        strv_free(c->read_only_dirs);
-        c->read_only_dirs = NULL;
-
-        strv_free(c->read_write_dirs);
-        c->read_write_dirs = NULL;
-
-        strv_free(c->inaccessible_dirs);
-        c->inaccessible_dirs = NULL;
-
-        if (c->cpuset)
-                CPU_FREE(c->cpuset);
-
-        free(c->utmp_id);
-        c->utmp_id = NULL;
-
-        free(c->selinux_context);
-        c->selinux_context = NULL;
-
-        free(c->apparmor_profile);
-        c->apparmor_profile = NULL;
-
-        set_free(c->syscall_filter);
-        c->syscall_filter = NULL;
-
-        set_free(c->syscall_archs);
-        c->syscall_archs = NULL;
-
-        set_free(c->address_families);
-        c->address_families = NULL;
-
-        strv_free(c->runtime_directory);
-        c->runtime_directory = NULL;
-
-        bus_endpoint_free(c->bus_endpoint);
-        c->bus_endpoint = NULL;
-}
-
-int exec_context_destroy_runtime_directory(ExecContext *c, const char *runtime_prefix) {
-        char **i;
-
-        assert(c);
-
-        if (!runtime_prefix)
-                return 0;
-
-        STRV_FOREACH(i, c->runtime_directory) {
-                _cleanup_free_ char *p;
-
-                p = strjoin(runtime_prefix, "/", *i, NULL);
-                if (!p)
-                        return -ENOMEM;
-
-                /* We execute this synchronously, since we need to be
-                 * sure this is gone when we start the service
-                 * next. */
-                rm_rf(p, false, true, false);
-        }
-
-        return 0;
-}
-
-void exec_command_done(ExecCommand *c) {
-        assert(c);
-
-        free(c->path);
-        c->path = NULL;
-
-        strv_free(c->argv);
-        c->argv = NULL;
-}
-
-void exec_command_done_array(ExecCommand *c, unsigned n) {
-        unsigned i;
-
-        for (i = 0; i < n; i++)
-                exec_command_done(c+i);
-}
-
-ExecCommand* exec_command_free_list(ExecCommand *c) {
-        ExecCommand *i;
-
-        while ((i = c)) {
-                LIST_REMOVE(command, c, i);
-                exec_command_done(i);
-                free(i);
-        }
-
-        return NULL;
-}
-
-void exec_command_free_array(ExecCommand **c, unsigned n) {
-        unsigned i;
-
-        for (i = 0; i < n; i++)
-                c[i] = exec_command_free_list(c[i]);
-}
-
-typedef struct InvalidEnvInfo {
-        const char *unit_id;
-        const char *path;
-} InvalidEnvInfo;
-
-static void invalid_env(const char *p, void *userdata) {
-        InvalidEnvInfo *info = userdata;
-
-        log_unit_error(info->unit_id, "Ignoring invalid environment assignment '%s': %s", p, info->path);
-}
-
-int exec_context_load_environment(const ExecContext *c, const char *unit_id, char ***l) {
-        char **i, **r = NULL;
-
-        assert(c);
-        assert(l);
-
-        STRV_FOREACH(i, c->environment_files) {
-                char *fn;
-                int k;
-                bool ignore = false;
-                char **p;
-                _cleanup_globfree_ glob_t pglob = {};
-                int count, n;
-
-                fn = *i;
-
-                if (fn[0] == '-') {
-                        ignore = true;
-                        fn ++;
-                }
-
-                if (!path_is_absolute(fn)) {
-                        if (ignore)
-                                continue;
-
-                        strv_free(r);
-                        return -EINVAL;
-                }
-
-                /* Filename supports globbing, take all matching files */
-                errno = 0;
-                if (glob(fn, 0, NULL, &pglob) != 0) {
-                        if (ignore)
-                                continue;
-
-                        strv_free(r);
-                        return errno ? -errno : -EINVAL;
-                }
-                count = pglob.gl_pathc;
-                if (count == 0) {
-                        if (ignore)
-                                continue;
-
-                        strv_free(r);
-                        return -EINVAL;
-                }
-                for (n = 0; n < count; n++) {
-                        k = load_env_file(NULL, pglob.gl_pathv[n], NULL, &p);
-                        if (k < 0) {
-                                if (ignore)
-                                        continue;
-
-                                strv_free(r);
-                                return k;
-                        }
-                        /* Log invalid environment variables with filename */
-                        if (p) {
-                                InvalidEnvInfo info = {
-                                        .unit_id = unit_id,
-                                        .path = pglob.gl_pathv[n]
-                                };
-
-                                p = strv_env_clean_with_callback(p, invalid_env, &info);
-                        }
-
-                        if (r == NULL)
-                                r = p;
-                        else {
-                                char **m;
-
-                                m = strv_env_merge(2, r, p);
-                                strv_free(r);
-                                strv_free(p);
-                                if (!m)
-                                        return -ENOMEM;
-
-                                r = m;
-                        }
-                }
-        }
-
-        *l = r;
-
-        return 0;
-}
-
-static bool tty_may_match_dev_console(const char *tty) {
-        _cleanup_free_ char *active = NULL;
-       char *console;
-
-        if (startswith(tty, "/dev/"))
-                tty += 5;
-
-        /* trivial identity? */
-        if (streq(tty, "console"))
-                return true;
-
-        console = resolve_dev_console(&active);
-        /* if we could not resolve, assume it may */
-        if (!console)
-                return true;
-
-        /* "tty0" means the active VC, so it may be the same sometimes */
-        return streq(console, tty) || (streq(console, "tty0") && tty_is_vc(tty));
-}
-
-bool exec_context_may_touch_console(ExecContext *ec) {
-        return (ec->tty_reset || ec->tty_vhangup || ec->tty_vt_disallocate ||
-                is_terminal_input(ec->std_input) ||
-                is_terminal_output(ec->std_output) ||
-                is_terminal_output(ec->std_error)) &&
-               tty_may_match_dev_console(tty_path(ec));
-}
-
-static void strv_fprintf(FILE *f, char **l) {
-        char **g;
-
-        assert(f);
-
-        STRV_FOREACH(g, l)
-                fprintf(f, " %s", *g);
-}
-
-void exec_context_dump(ExecContext *c, FILE* f, const char *prefix) {
-        char **e;
-        unsigned i;
-
-        assert(c);
-        assert(f);
-
-        prefix = strempty(prefix);
-
-        fprintf(f,
-                "%sUMask: %04o\n"
-                "%sWorkingDirectory: %s\n"
-                "%sRootDirectory: %s\n"
-                "%sNonBlocking: %s\n"
-                "%sPrivateTmp: %s\n"
-                "%sPrivateNetwork: %s\n"
-                "%sPrivateDevices: %s\n"
-                "%sProtectHome: %s\n"
-                "%sProtectSystem: %s\n"
-                "%sIgnoreSIGPIPE: %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_network),
-                prefix, yes_no(c->private_devices),
-                prefix, protect_home_to_string(c->protect_home),
-                prefix, protect_system_to_string(c->protect_system),
-                prefix, yes_no(c->ignore_sigpipe));
-
-        STRV_FOREACH(e, c->environment)
-                fprintf(f, "%sEnvironment: %s\n", prefix, *e);
-
-        STRV_FOREACH(e, c->environment_files)
-                fprintf(f, "%sEnvironmentFile: %s\n", prefix, *e);
-
-        if (c->nice_set)
-                fprintf(f,
-                        "%sNice: %i\n",
-                        prefix, c->nice);
-
-        if (c->oom_score_adjust_set)
-                fprintf(f,
-                        "%sOOMScoreAdjust: %i\n",
-                        prefix, c->oom_score_adjust);
-
-        for (i = 0; i < RLIM_NLIMITS; i++)
-                if (c->rlimit[i])
-                        fprintf(f, "%s%s: "RLIM_FMT"\n",
-                                prefix, rlimit_to_string(i), c->rlimit[i]->rlim_max);
-
-        if (c->ioprio_set) {
-                _cleanup_free_ char *class_str = NULL;
-
-                ioprio_class_to_string_alloc(IOPRIO_PRIO_CLASS(c->ioprio), &class_str);
-                fprintf(f,
-                        "%sIOSchedulingClass: %s\n"
-                        "%sIOPriority: %i\n",
-                        prefix, strna(class_str),
-                        prefix, (int) IOPRIO_PRIO_DATA(c->ioprio));
-        }
-
-        if (c->cpu_sched_set) {
-                _cleanup_free_ char *policy_str = NULL;
-
-                sched_policy_to_string_alloc(c->cpu_sched_policy, &policy_str);
-                fprintf(f,
-                        "%sCPUSchedulingPolicy: %s\n"
-                        "%sCPUSchedulingPriority: %i\n"
-                        "%sCPUSchedulingResetOnFork: %s\n",
-                        prefix, strna(policy_str),
-                        prefix, c->cpu_sched_priority,
-                        prefix, yes_no(c->cpu_sched_reset_on_fork));
-        }
-
-        if (c->cpuset) {
-                fprintf(f, "%sCPUAffinity:", prefix);
-                for (i = 0; i < c->cpuset_ncpus; i++)
-                        if (CPU_ISSET_S(i, CPU_ALLOC_SIZE(c->cpuset_ncpus), c->cpuset))
-                                fprintf(f, " %u", i);
-                fputs("\n", f);
-        }
-
-        if (c->timer_slack_nsec != NSEC_INFINITY)
-                fprintf(f, "%sTimerSlackNSec: "NSEC_FMT "\n", prefix, c->timer_slack_nsec);
-
-        fprintf(f,
-                "%sStandardInput: %s\n"
-                "%sStandardOutput: %s\n"
-                "%sStandardError: %s\n",
-                prefix, exec_input_to_string(c->std_input),
-                prefix, exec_output_to_string(c->std_output),
-                prefix, exec_output_to_string(c->std_error));
-
-        if (c->tty_path)
-                fprintf(f,
-                        "%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_JOURNAL ||
-            c->std_output == EXEC_OUTPUT_SYSLOG_AND_CONSOLE ||
-            c->std_output == EXEC_OUTPUT_KMSG_AND_CONSOLE ||
-            c->std_output == EXEC_OUTPUT_JOURNAL_AND_CONSOLE ||
-            c->std_error == EXEC_OUTPUT_SYSLOG ||
-            c->std_error == EXEC_OUTPUT_KMSG ||
-            c->std_error == EXEC_OUTPUT_JOURNAL ||
-            c->std_error == EXEC_OUTPUT_SYSLOG_AND_CONSOLE ||
-            c->std_error == EXEC_OUTPUT_KMSG_AND_CONSOLE ||
-            c->std_error == EXEC_OUTPUT_JOURNAL_AND_CONSOLE) {
-
-                _cleanup_free_ char *fac_str = NULL, *lvl_str = NULL;
-
-                log_facility_unshifted_to_string_alloc(c->syslog_priority >> 3, &fac_str);
-                log_level_to_string_alloc(LOG_PRI(c->syslog_priority), &lvl_str);
-
-                fprintf(f,
-                        "%sSyslogFacility: %s\n"
-                        "%sSyslogLevel: %s\n",
-                        prefix, strna(fac_str),
-                        prefix, strna(lvl_str));
-        }
-
-        if (c->capabilities) {
-                _cleanup_cap_free_charp_ char *t;
-
-                t = cap_to_text(c->capabilities, NULL);
-                if (t)
-                        fprintf(f, "%sCapabilities: %s\n", prefix, t);
-        }
-
-        if (c->secure_bits)
-                fprintf(f, "%sSecure Bits:%s%s%s%s%s%s\n",
-                        prefix,
-                        (c->secure_bits & 1<<SECURE_KEEP_CAPS) ? " keep-caps" : "",
-                        (c->secure_bits & 1<<SECURE_KEEP_CAPS_LOCKED) ? " keep-caps-locked" : "",
-                        (c->secure_bits & 1<<SECURE_NO_SETUID_FIXUP) ? " no-setuid-fixup" : "",
-                        (c->secure_bits & 1<<SECURE_NO_SETUID_FIXUP_LOCKED) ? " no-setuid-fixup-locked" : "",
-                        (c->secure_bits & 1<<SECURE_NOROOT) ? " noroot" : "",
-                        (c->secure_bits & 1<<SECURE_NOROOT_LOCKED) ? "noroot-locked" : "");
-
-        if (c->capability_bounding_set_drop) {
-                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)))
-                                fprintf(f, " %s", strna(capability_to_name(l)));
-
-                fputs("\n", f);
-        }
-
-        if (c->user)
-                fprintf(f, "%sUser: %s\n", prefix, c->user);
-        if (c->group)
-                fprintf(f, "%sGroup: %s\n", prefix, c->group);
-
-        if (strv_length(c->supplementary_groups) > 0) {
-                fprintf(f, "%sSupplementaryGroups:", prefix);
-                strv_fprintf(f, c->supplementary_groups);
-                fputs("\n", f);
-        }
-
-        if (c->pam_name)
-                fprintf(f, "%sPAMName: %s\n", prefix, c->pam_name);
-
-        if (strv_length(c->read_write_dirs) > 0) {
-                fprintf(f, "%sReadWriteDirs:", prefix);
-                strv_fprintf(f, c->read_write_dirs);
-                fputs("\n", f);
-        }
-
-        if (strv_length(c->read_only_dirs) > 0) {
-                fprintf(f, "%sReadOnlyDirs:", prefix);
-                strv_fprintf(f, c->read_only_dirs);
-                fputs("\n", f);
-        }
-
-        if (strv_length(c->inaccessible_dirs) > 0) {
-                fprintf(f, "%sInaccessibleDirs:", prefix);
-                strv_fprintf(f, c->inaccessible_dirs);
-                fputs("\n", f);
-        }
-
-        if (c->utmp_id)
-                fprintf(f,
-                        "%sUtmpIdentifier: %s\n",
-                        prefix, c->utmp_id);
-
-        if (c->selinux_context)
-                fprintf(f,
-                        "%sSELinuxContext: %s%s\n",
-                        prefix, c->selinux_context_ignore ? "-" : "", c->selinux_context);
-
-        if (c->personality != 0xffffffffUL)
-                fprintf(f,
-                        "%sPersonality: %s\n",
-                        prefix, strna(personality_to_string(c->personality)));
-
-        if (c->syscall_filter) {
-#ifdef HAVE_SECCOMP
-                Iterator j;
-                void *id;
-                bool first = true;
-#endif
-
-                fprintf(f,
-                        "%sSystemCallFilter: ",
-                        prefix);
-
-                if (!c->syscall_whitelist)
-                        fputc('~', f);
-
-#ifdef HAVE_SECCOMP
-                SET_FOREACH(id, c->syscall_filter, j) {
-                        _cleanup_free_ char *name = NULL;
-
-                        if (first)
-                                first = false;
-                        else
-                                fputc(' ', f);
-
-                        name = seccomp_syscall_resolve_num_arch(SCMP_ARCH_NATIVE, PTR_TO_INT(id) - 1);
-                        fputs(strna(name), f);
-                }
-#endif
-
-                fputc('\n', f);
-        }
-
-        if (c->syscall_archs) {
-#ifdef HAVE_SECCOMP
-                Iterator j;
-                void *id;
-#endif
-
-                fprintf(f,
-                        "%sSystemCallArchitectures:",
-                        prefix);
-
-#ifdef HAVE_SECCOMP
-                SET_FOREACH(id, c->syscall_archs, j)
-                        fprintf(f, " %s", strna(seccomp_arch_to_string(PTR_TO_UINT32(id) - 1)));
-#endif
-                fputc('\n', f);
-        }
-
-        if (c->syscall_errno != 0)
-                fprintf(f,
-                        "%sSystemCallErrorNumber: %s\n",
-                        prefix, strna(errno_to_name(c->syscall_errno)));
-
-        if (c->apparmor_profile)
-                fprintf(f,
-                        "%sAppArmorProfile: %s%s\n",
-                        prefix, c->apparmor_profile_ignore ? "-" : "", c->apparmor_profile);
-}
-
-bool exec_context_maintains_privileges(ExecContext *c) {
-        assert(c);
-
-        /* Returns true if the process forked off would run run under
-         * an unchanged UID or as root. */
-
-        if (!c->user)
-                return true;
-
-        if (streq(c->user, "root") || streq(c->user, "0"))
-                return true;
-
-        return false;
-}
-
-void exec_status_start(ExecStatus *s, pid_t pid) {
-        assert(s);
-
-        zero(*s);
-        s->pid = pid;
-        dual_timestamp_get(&s->start_timestamp);
-}
-
-void exec_status_exit(ExecStatus *s, ExecContext *context, pid_t pid, int code, int status) {
-        assert(s);
-
-        if (s->pid && s->pid != pid)
-                zero(*s);
-
-        s->pid = pid;
-        dual_timestamp_get(&s->exit_timestamp);
-
-        s->code = code;
-        s->status = 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) {
-        char buf[FORMAT_TIMESTAMP_MAX];
-
-        assert(s);
-        assert(f);
-
-        if (s->pid <= 0)
-                return;
-
-        prefix = strempty(prefix);
-
-        fprintf(f,
-                "%sPID: "PID_FMT"\n",
-                prefix, s->pid);
-
-        if (s->start_timestamp.realtime > 0)
-                fprintf(f,
-                        "%sStart Timestamp: %s\n",
-                        prefix, format_timestamp(buf, sizeof(buf), s->start_timestamp.realtime));
-
-        if (s->exit_timestamp.realtime > 0)
-                fprintf(f,
-                        "%sExit Timestamp: %s\n"
-                        "%sExit Code: %s\n"
-                        "%sExit Status: %i\n",
-                        prefix, format_timestamp(buf, sizeof(buf), s->exit_timestamp.realtime),
-                        prefix, sigchld_code_to_string(s->code),
-                        prefix, s->status);
-}
-
-char *exec_command_line(char **argv) {
-        size_t k;
-        char *n, *p, **a;
-        bool first = true;
-
-        assert(argv);
-
-        k = 1;
-        STRV_FOREACH(a, argv)
-                k += strlen(*a)+3;
-
-        if (!(n = new(char, k)))
-                return NULL;
-
-        p = n;
-        STRV_FOREACH(a, argv) {
-
-                if (!first)
-                        *(p++) = ' ';
-                else
-                        first = false;
-
-                if (strpbrk(*a, WHITESPACE)) {
-                        *(p++) = '\'';
-                        p = stpcpy(p, *a);
-                        *(p++) = '\'';
-                } else
-                        p = stpcpy(p, *a);
-
-        }
-
-        *p = 0;
-
-        /* FIXME: this doesn't really handle arguments that have
-         * spaces and ticks in them */
-
-        return n;
-}
-
-void exec_command_dump(ExecCommand *c, FILE *f, const char *prefix) {
-        _cleanup_free_ char *cmd = NULL;
-        const char *prefix2;
-
-        assert(c);
-        assert(f);
-
-        prefix = strempty(prefix);
-        prefix2 = strjoina(prefix, "\t");
-
-        cmd = exec_command_line(c->argv);
-        fprintf(f,
-                "%sCommand Line: %s\n",
-                prefix, cmd ? cmd : strerror(ENOMEM));
-
-        exec_status_dump(&c->exec_status, f, prefix2);
-}
-
-void exec_command_dump_list(ExecCommand *c, FILE *f, const char *prefix) {
-        assert(f);
-
-        prefix = strempty(prefix);
-
-        LIST_FOREACH(command, c, c)
-                exec_command_dump(c, f, prefix);
-}
-
-void exec_command_append_list(ExecCommand **l, ExecCommand *e) {
-        ExecCommand *end;
-
-        assert(l);
-        assert(e);
-
-        if (*l) {
-                /* It's kind of important, that we keep the order here */
-                LIST_FIND_TAIL(command, *l, end);
-                LIST_INSERT_AFTER(command, *l, end, e);
-        } else
-              *l = e;
-}
-
-int exec_command_set(ExecCommand *c, const char *path, ...) {
-        va_list ap;
-        char **l, *p;
-
-        assert(c);
-        assert(path);
-
-        va_start(ap, path);
-        l = strv_new_ap(path, ap);
-        va_end(ap);
-
-        if (!l)
-                return -ENOMEM;
-
-        p = strdup(path);
-        if (!p) {
-                strv_free(l);
-                return -ENOMEM;
-        }
-
-        free(c->path);
-        c->path = p;
-
-        strv_free(c->argv);
-        c->argv = l;
-
-        return 0;
-}
-
-int exec_command_append(ExecCommand *c, const char *path, ...) {
-        _cleanup_strv_free_ char **l = NULL;
-        va_list ap;
-        int r;
-
-        assert(c);
-        assert(path);
-
-        va_start(ap, path);
-        l = strv_new_ap(path, ap);
-        va_end(ap);
-
-        if (!l)
-                return -ENOMEM;
-
-        r = strv_extend_strv(&c->argv, l);
-        if (r < 0)
-                return r;
-
-        return 0;
-}
-
-
-static int exec_runtime_allocate(ExecRuntime **rt) {
-
-        if (*rt)
-                return 0;
-
-        *rt = new0(ExecRuntime, 1);
-        if (!*rt)
-                return -ENOMEM;
-
-        (*rt)->n_ref = 1;
-        (*rt)->netns_storage_socket[0] = (*rt)->netns_storage_socket[1] = -1;
-
-        return 0;
-}
-
-int exec_runtime_make(ExecRuntime **rt, ExecContext *c, const char *id) {
-        int r;
-
-        assert(rt);
-        assert(c);
-        assert(id);
-
-        if (*rt)
-                return 1;
-
-        if (!c->private_network && !c->private_tmp)
-                return 0;
-
-        r = exec_runtime_allocate(rt);
-        if (r < 0)
-                return r;
-
-        if (c->private_network && (*rt)->netns_storage_socket[0] < 0) {
-                if (socketpair(AF_UNIX, SOCK_DGRAM, 0, (*rt)->netns_storage_socket) < 0)
-                        return -errno;
-        }
-
-        if (c->private_tmp && !(*rt)->tmp_dir) {
-                r = setup_tmp_dirs(id, &(*rt)->tmp_dir, &(*rt)->var_tmp_dir);
-                if (r < 0)
-                        return r;
-        }
-
-        return 1;
-}
-
-ExecRuntime *exec_runtime_ref(ExecRuntime *r) {
-        assert(r);
-        assert(r->n_ref > 0);
-
-        r->n_ref++;
-        return r;
-}
-
-ExecRuntime *exec_runtime_unref(ExecRuntime *r) {
-
-        if (!r)
-                return NULL;
-
-        assert(r->n_ref > 0);
-
-        r->n_ref--;
-        if (r->n_ref <= 0) {
-                free(r->tmp_dir);
-                free(r->var_tmp_dir);
-                safe_close_pair(r->netns_storage_socket);
-                free(r);
-        }
-
-        return NULL;
-}
-
-int exec_runtime_serialize(ExecRuntime *rt, Unit *u, FILE *f, FDSet *fds) {
-        assert(u);
-        assert(f);
-        assert(fds);
-
-        if (!rt)
-                return 0;
-
-        if (rt->tmp_dir)
-                unit_serialize_item(u, f, "tmp-dir", rt->tmp_dir);
-
-        if (rt->var_tmp_dir)
-                unit_serialize_item(u, f, "var-tmp-dir", rt->var_tmp_dir);
-
-        if (rt->netns_storage_socket[0] >= 0) {
-                int copy;
-
-                copy = fdset_put_dup(fds, rt->netns_storage_socket[0]);
-                if (copy < 0)
-                        return copy;
-
-                unit_serialize_item_format(u, f, "netns-socket-0", "%i", copy);
-        }
-
-        if (rt->netns_storage_socket[1] >= 0) {
-                int copy;
-
-                copy = fdset_put_dup(fds, rt->netns_storage_socket[1]);
-                if (copy < 0)
-                        return copy;
-
-                unit_serialize_item_format(u, f, "netns-socket-1", "%i", copy);
-        }
-
-        return 0;
-}
-
-int exec_runtime_deserialize_item(ExecRuntime **rt, Unit *u, const char *key, const char *value, FDSet *fds) {
-        int r;
-
-        assert(rt);
-        assert(key);
-        assert(value);
-
-        if (streq(key, "tmp-dir")) {
-                char *copy;
-
-                r = exec_runtime_allocate(rt);
-                if (r < 0)
-                        return r;
-
-                copy = strdup(value);
-                if (!copy)
-                        return log_oom();
-
-                free((*rt)->tmp_dir);
-                (*rt)->tmp_dir = copy;
-
-        } else if (streq(key, "var-tmp-dir")) {
-                char *copy;
-
-                r = exec_runtime_allocate(rt);
-                if (r < 0)
-                        return r;
-
-                copy = strdup(value);
-                if (!copy)
-                        return log_oom();
-
-                free((*rt)->var_tmp_dir);
-                (*rt)->var_tmp_dir = copy;
-
-        } else if (streq(key, "netns-socket-0")) {
-                int fd;
-
-                r = exec_runtime_allocate(rt);
-                if (r < 0)
-                        return r;
-
-                if (safe_atoi(value, &fd) < 0 || !fdset_contains(fds, fd))
-                        log_unit_debug(u->id, "Failed to parse netns socket value %s", value);
-                else {
-                        safe_close((*rt)->netns_storage_socket[0]);
-                        (*rt)->netns_storage_socket[0] = fdset_remove(fds, fd);
-                }
-        } else if (streq(key, "netns-socket-1")) {
-                int fd;
-
-                r = exec_runtime_allocate(rt);
-                if (r < 0)
-                        return r;
-
-                if (safe_atoi(value, &fd) < 0 || !fdset_contains(fds, fd))
-                        log_unit_debug(u->id, "Failed to parse netns socket value %s", value);
-                else {
-                        safe_close((*rt)->netns_storage_socket[1]);
-                        (*rt)->netns_storage_socket[1] = fdset_remove(fds, fd);
-                }
-        } else
-                return 0;
-
-        return 1;
-}
-
-static void *remove_tmpdir_thread(void *p) {
-        _cleanup_free_ char *path = p;
-
-        rm_rf_dangerous(path, false, true, false);
-        return NULL;
-}
-
-void exec_runtime_destroy(ExecRuntime *rt) {
-        int r;
-
-        if (!rt)
-                return;
-
-        /* If there are multiple users of this, let's leave the stuff around */
-        if (rt->n_ref > 1)
-                return;
-
-        if (rt->tmp_dir) {
-                log_debug("Spawning thread to nuke %s", rt->tmp_dir);
-
-                r = asynchronous_job(remove_tmpdir_thread, rt->tmp_dir);
-                if (r < 0) {
-                        log_warning_errno(r, "Failed to nuke %s: %m", rt->tmp_dir);
-                        free(rt->tmp_dir);
-                }
-
-                rt->tmp_dir = NULL;
-        }
-
-        if (rt->var_tmp_dir) {
-                log_debug("Spawning thread to nuke %s", rt->var_tmp_dir);
-
-                r = asynchronous_job(remove_tmpdir_thread, rt->var_tmp_dir);
-                if (r < 0) {
-                        log_warning_errno(r, "Failed to nuke %s: %m", rt->var_tmp_dir);
-                        free(rt->var_tmp_dir);
-                }
-
-                rt->var_tmp_dir = NULL;
-        }
-
-        safe_close_pair(rt->netns_storage_socket);
-}
-
-static const char* const exec_input_table[_EXEC_INPUT_MAX] = {
-        [EXEC_INPUT_NULL] = "null",
-        [EXEC_INPUT_TTY] = "tty",
-        [EXEC_INPUT_TTY_FORCE] = "tty-force",
-        [EXEC_INPUT_TTY_FAIL] = "tty-fail",
-        [EXEC_INPUT_SOCKET] = "socket"
-};
-
-DEFINE_STRING_TABLE_LOOKUP(exec_input, ExecInput);
-
-static const char* const exec_output_table[_EXEC_OUTPUT_MAX] = {
-        [EXEC_OUTPUT_INHERIT] = "inherit",
-        [EXEC_OUTPUT_NULL] = "null",
-        [EXEC_OUTPUT_TTY] = "tty",
-        [EXEC_OUTPUT_SYSLOG] = "syslog",
-        [EXEC_OUTPUT_SYSLOG_AND_CONSOLE] = "syslog+console",
-        [EXEC_OUTPUT_KMSG] = "kmsg",
-        [EXEC_OUTPUT_KMSG_AND_CONSOLE] = "kmsg+console",
-        [EXEC_OUTPUT_JOURNAL] = "journal",
-        [EXEC_OUTPUT_JOURNAL_AND_CONSOLE] = "journal+console",
-        [EXEC_OUTPUT_SOCKET] = "socket"
-};
-
-DEFINE_STRING_TABLE_LOOKUP(exec_output, ExecOutput);