chiark / gitweb /
logind: remove unused session->closing field
[elogind.git] / src / core / execute.c
index 4aed4f68497d3fa1dd277d43e51b88542b684f6f..a0d63a41f9a521b82b806537d4db1e19505da468 100644 (file)
@@ -38,6 +38,9 @@
 #include <linux/fs.h>
 #include <linux/oom.h>
 #include <sys/poll.h>
+#include <linux/seccomp-bpf.h>
+#include <glob.h>
+#include <libgen.h>
 
 #ifdef HAVE_PAM
 #include <security/pam_appl.h>
 #include "capability.h"
 #include "util.h"
 #include "log.h"
+#include "sd-messages.h"
 #include "ioprio.h"
 #include "securebits.h"
-#include "cgroup.h"
 #include "namespace.h"
 #include "tcpwrap.h"
 #include "exit-status.h"
 #include "missing.h"
 #include "utmp-wtmp.h"
 #include "def.h"
-#include "loopback-setup.h"
 #include "path-util.h"
+#include "syscall-list.h"
+#include "env-util.h"
+#include "fileio.h"
+#include "unit.h"
+#include "async.h"
 
-#define IDLE_TIMEOUT_USEC (1*USEC_PER_SEC)
+#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
@@ -137,7 +145,7 @@ static int flags_fds(const int fds[], unsigned n_fds, bool nonblock) {
         return 0;
 }
 
-static const char *tty_path(const ExecContext *context) {
+_pure_ static const char *tty_path(const ExecContext *context) {
         assert(context);
 
         if (context->tty_path)
@@ -146,7 +154,7 @@ static const char *tty_path(const ExecContext *context) {
         return "/dev/console";
 }
 
-void exec_context_tty_reset(const ExecContext *context) {
+static void exec_context_tty_reset(const ExecContext *context) {
         assert(context);
 
         if (context->tty_vhangup)
@@ -159,12 +167,21 @@ void exec_context_tty_reset(const ExecContext *context) {
                 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);
 
-        if ((fd = open("/dev/null", flags|O_NOCTTY)) < 0)
+        fd = open("/dev/null", flags|O_NOCTTY);
+        if (fd < 0)
                 return -errno;
 
         if (fd != nfd) {
@@ -176,9 +193,12 @@ static int open_null_as(int flags, int nfd) {
         return r;
 }
 
-static int connect_logger_as(const ExecContext *context, ExecOutput output, const char *ident, int nfd) {
+static int connect_logger_as(const ExecContext *context, ExecOutput output, const char *ident, const char *unit_id, int nfd) {
         int fd, r;
-        union sockaddr_union sa;
+        union sockaddr_union sa = {
+                .un.sun_family = AF_UNIX,
+                .un.sun_path = "/run/systemd/journal/stdout",
+        };
 
         assert(context);
         assert(output < _EXEC_OUTPUT_MAX);
@@ -189,10 +209,6 @@ static int connect_logger_as(const ExecContext *context, ExecOutput output, cons
         if (fd < 0)
                 return -errno;
 
-        zero(sa);
-        sa.un.sun_family = AF_UNIX;
-        strncpy(sa.un.sun_path, "/run/systemd/journal/stdout", sizeof(sa.un.sun_path));
-
         r = connect(fd, &sa.sa, offsetof(struct sockaddr_un, sun_path) + strlen(sa.un.sun_path));
         if (r < 0) {
                 close_nointr_nofail(fd);
@@ -205,6 +221,7 @@ static int connect_logger_as(const ExecContext *context, ExecOutput output, cons
         }
 
         dprintf(fd,
+                "%s\n"
                 "%s\n"
                 "%i\n"
                 "%i\n"
@@ -212,11 +229,12 @@ static int connect_logger_as(const ExecContext *context, ExecOutput output, cons
                 "%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,
-                output == EXEC_OUTPUT_SYSLOG_AND_CONSOLE || output == EXEC_OUTPUT_KMSG_AND_CONSOLE || output == EXEC_OUTPUT_JOURNAL_AND_CONSOLE);
+                is_terminal_output(output));
 
         if (fd != nfd) {
                 r = dup2(fd, nfd) < 0 ? -errno : nfd;
@@ -287,11 +305,12 @@ static int setup_input(const ExecContext *context, int socket_fd, bool apply_tty
         case EXEC_INPUT_TTY_FAIL: {
                 int fd, r;
 
-                if ((fd = acquire_terminal(
-                                     tty_path(context),
-                                     i == EXEC_INPUT_TTY_FAIL,
-                                     i == EXEC_INPUT_TTY_FORCE,
-                                     false)) < 0)
+                fd = acquire_terminal(tty_path(context),
+                                      i == EXEC_INPUT_TTY_FAIL,
+                                      i == EXEC_INPUT_TTY_FORCE,
+                                      false,
+                                      (usec_t) -1);
+                if (fd < 0)
                         return fd;
 
                 if (fd != STDIN_FILENO) {
@@ -311,9 +330,10 @@ static int setup_input(const ExecContext *context, int socket_fd, bool apply_tty
         }
 }
 
-static int setup_output(const ExecContext *context, int socket_fd, const char *ident, bool apply_tty_stdin) {
+static int setup_output(const ExecContext *context, int fileno, int socket_fd, const char *ident, const char *unit_id, bool apply_tty_stdin) {
         ExecOutput o;
         ExecInput i;
+        int r;
 
         assert(context);
         assert(ident);
@@ -321,91 +341,55 @@ static int setup_output(const ExecContext *context, int socket_fd, const char *i
         i = fixup_input(context->std_input, socket_fd, apply_tty_stdin);
         o = fixup_output(context->std_output, socket_fd);
 
-        /* This expects the input is already set up */
+        if (fileno == STDERR_FILENO) {
+                ExecOutput e;
+                e = fixup_output(context->std_error, socket_fd);
 
-        switch (o) {
+                /* 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;
 
-        case EXEC_OUTPUT_INHERIT:
+                /* 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, STDOUT_FILENO);
+                        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, STDOUT_FILENO) < 0 ? -errno : STDOUT_FILENO;
+                        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 STDOUT_FILENO;
-
-                /* We need to open /dev/null here anew, to get the
-                 * right access mode. So we fall through */
-
-        case EXEC_OUTPUT_NULL:
-                return open_null_as(O_WRONLY, STDOUT_FILENO);
-
-        case EXEC_OUTPUT_TTY:
-                if (is_terminal_input(i))
-                        return dup2(STDIN_FILENO, STDOUT_FILENO) < 0 ? -errno : STDOUT_FILENO;
-
-                /* We don't reset the terminal if this is just about output */
-                return open_terminal_as(tty_path(context), O_WRONLY, STDOUT_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:
-                return connect_logger_as(context, o, ident, STDOUT_FILENO);
+                        return fileno;
 
-        case EXEC_OUTPUT_SOCKET:
-                assert(socket_fd >= 0);
-                return dup2(socket_fd, STDOUT_FILENO) < 0 ? -errno : STDOUT_FILENO;
-
-        default:
-                assert_not_reached("Unknown output type");
+                /* We need to open /dev/null here anew, to get the right access mode. */
+                return open_null_as(O_WRONLY, fileno);
         }
-}
-
-static int setup_error(const ExecContext *context, int socket_fd, const char *ident, bool apply_tty_stdin) {
-        ExecOutput o, e;
-        ExecInput i;
-
-        assert(context);
-        assert(ident);
-
-        i = fixup_input(context->std_input, socket_fd, apply_tty_stdin);
-        o = fixup_output(context->std_output, socket_fd);
-        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 STDERR_FILENO;
-
-        /* Duplicate from stdout if possible */
-        if (e == o || e == EXEC_OUTPUT_INHERIT)
-                return dup2(STDOUT_FILENO, STDERR_FILENO) < 0 ? -errno : STDERR_FILENO;
-
-        switch (e) {
+        switch (o) {
 
         case EXEC_OUTPUT_NULL:
-                return open_null_as(O_WRONLY, STDERR_FILENO);
+                return open_null_as(O_WRONLY, fileno);
 
         case EXEC_OUTPUT_TTY:
                 if (is_terminal_input(i))
-                        return dup2(STDIN_FILENO, STDERR_FILENO) < 0 ? -errno : STDERR_FILENO;
+                        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, STDERR_FILENO);
+                return open_terminal_as(tty_path(context), O_WRONLY, fileno);
 
         case EXEC_OUTPUT_SYSLOG:
         case EXEC_OUTPUT_SYSLOG_AND_CONSOLE:
@@ -413,11 +397,21 @@ static int setup_error(const ExecContext *context, int socket_fd, const char *id
         case EXEC_OUTPUT_KMSG_AND_CONSOLE:
         case EXEC_OUTPUT_JOURNAL:
         case EXEC_OUTPUT_JOURNAL_AND_CONSOLE:
-                return connect_logger_as(context, e, ident, STDERR_FILENO);
+                r = connect_logger_as(context, o, ident, unit_id, fileno);
+                if (r < 0) {
+                        log_struct_unit(LOG_CRIT, unit_id,
+                                "MESSAGE=Failed to connect std%s of %s to the journal socket: %s",
+                                fileno == STDOUT_FILENO ? "out" : "err",
+                                unit_id, strerror(-r),
+                                "ERRNO=%d", -r,
+                                NULL);
+                        r = open_null_as(O_WRONLY, fileno);
+                }
+                return r;
 
         case EXEC_OUTPUT_SOCKET:
                 assert(socket_fd >= 0);
-                return dup2(socket_fd, STDERR_FILENO) < 0 ? -errno : STDERR_FILENO;
+                return dup2(socket_fd, fileno) < 0 ? -errno : fileno;
 
         default:
                 assert_not_reached("Unknown error type");
@@ -442,47 +436,45 @@ static int chown_terminal(int fd, uid_t uid) {
         return 0;
 }
 
-static int setup_confirm_stdio(const ExecContext *context,
-                               int *_saved_stdin,
+static int setup_confirm_stdio(int *_saved_stdin,
                                int *_saved_stdout) {
         int fd = -1, saved_stdin, saved_stdout = -1, r;
 
-        assert(context);
         assert(_saved_stdin);
         assert(_saved_stdout);
 
-        /* This returns positive EXIT_xxx return values instead of
-         * negative errno style values! */
-
-        if ((saved_stdin = fcntl(STDIN_FILENO, F_DUPFD, 3)) < 0)
-                return EXIT_STDIN;
+        saved_stdin = fcntl(STDIN_FILENO, F_DUPFD, 3);
+        if (saved_stdin < 0)
+                return -errno;
 
-        if ((saved_stdout = fcntl(STDOUT_FILENO, F_DUPFD, 3)) < 0) {
-                r = EXIT_STDOUT;
+        saved_stdout = fcntl(STDOUT_FILENO, F_DUPFD, 3);
+        if (saved_stdout < 0) {
+                r = errno;
                 goto fail;
         }
 
-        if ((fd = acquire_terminal(
-                             tty_path(context),
-                             context->std_input == EXEC_INPUT_TTY_FAIL,
-                             context->std_input == EXEC_INPUT_TTY_FORCE,
-                             false)) < 0) {
-                r = EXIT_STDIN;
+        fd = acquire_terminal(
+                        "/dev/console",
+                        false,
+                        false,
+                        false,
+                        DEFAULT_CONFIRM_USEC);
+        if (fd < 0) {
+                r = fd;
                 goto fail;
         }
 
-        if (chown_terminal(fd, getuid()) < 0) {
-                r = EXIT_STDIN;
+        r = chown_terminal(fd, getuid());
+        if (r < 0)
                 goto fail;
-        }
 
         if (dup2(fd, STDIN_FILENO) < 0) {
-                r = EXIT_STDIN;
+                r = -errno;
                 goto fail;
         }
 
         if (dup2(fd, STDOUT_FILENO) < 0) {
-                r = EXIT_STDOUT;
+                r = -errno;
                 goto fail;
         }
 
@@ -507,48 +499,70 @@ fail:
         return r;
 }
 
-static int restore_confirm_stdio(const ExecContext *context,
-                                 int *saved_stdin,
-                                 int *saved_stdout,
-                                 bool *keep_stdin,
-                                 bool *keep_stdout) {
+_printf_(1, 2) static int write_confirm_message(const char *format, ...) {
+        int fd;
+        va_list ap;
 
-        assert(context);
-        assert(saved_stdin);
-        assert(*saved_stdin >= 0);
-        assert(saved_stdout);
-        assert(*saved_stdout >= 0);
+        assert(format);
 
-        /* This returns positive EXIT_xxx return values instead of
-         * negative errno style values! */
+        fd = open_terminal("/dev/console", O_WRONLY|O_NOCTTY|O_CLOEXEC);
+        if (fd < 0)
+                return fd;
 
-        if (is_terminal_input(context->std_input)) {
+        va_start(ap, format);
+        vdprintf(fd, format, ap);
+        va_end(ap);
 
-                /* The service wants terminal input. */
+        close_nointr_nofail(fd);
 
-                *keep_stdin = true;
-                *keep_stdout =
-                        context->std_output == EXEC_OUTPUT_INHERIT ||
-                        context->std_output == EXEC_OUTPUT_TTY;
+        return 0;
+}
 
-        } else {
-                /* If the service doesn't want a controlling terminal,
-                 * then we need to get rid entirely of what we have
-                 * already. */
+static int restore_confirm_stdio(int *saved_stdin,
+                                 int *saved_stdout) {
+
+        int r = 0;
+
+        assert(saved_stdin);
+        assert(saved_stdout);
 
-                if (release_terminal() < 0)
-                        return EXIT_STDIN;
+        release_terminal();
 
+        if (*saved_stdin >= 0)
                 if (dup2(*saved_stdin, STDIN_FILENO) < 0)
-                        return EXIT_STDIN;
+                        r = -errno;
 
+        if (*saved_stdout >= 0)
                 if (dup2(*saved_stdout, STDOUT_FILENO) < 0)
-                        return EXIT_STDOUT;
+                        r = -errno;
 
-                *keep_stdout = *keep_stdin = false;
-        }
+        if (*saved_stdin >= 0)
+                close_nointr_nofail(*saved_stdin);
 
-        return 0;
+        if (*saved_stdout >= 0)
+                close_nointr_nofail(*saved_stdout);
+
+        return r;
+}
+
+static int ask_for_confirmation(char *response, char **argv) {
+        int saved_stdout = -1, saved_stdin = -1, r;
+        char *line;
+
+        r = setup_confirm_stdio(&saved_stdin, &saved_stdout);
+        if (r < 0)
+                return r;
+
+        line = exec_command_line(argv);
+        if (!line)
+                return -ENOMEM;
+
+        r = ask(response, "yns", "Execute %s? [Yes, No, Skip] ", line);
+        free(line);
+
+        restore_confirm_stdio(&saved_stdin, &saved_stdout);
+
+        return r;
 }
 
 static int enforce_groups(const ExecContext *context, const char *username, gid_t gid) {
@@ -646,9 +660,9 @@ static int enforce_user(const ExecContext *context, uid_t uid) {
 
                 /* First step: If we need to keep capabilities but
                  * drop privileges we need to make sure we keep our
-                 * caps, whiel we drop privileges. */
+                 * caps, while we drop privileges. */
                 if (uid != 0) {
-                        int sb = context->secure_bits|SECURE_KEEP_CAPS;
+                        int sb = context->secure_bits | 1<<SECURE_KEEP_CAPS;
 
                         if (prctl(PR_GET_SECUREBITS) != sb)
                                 if (prctl(PR_SET_SECUREBITS, sb) < 0)
@@ -722,6 +736,7 @@ static int setup_pam(
         char **e = NULL;
         bool close_session = false;
         pid_t pam_pid = 0, parent_pid;
+        int flags = 0;
 
         assert(name);
         assert(user);
@@ -734,24 +749,33 @@ static int setup_pam(
          * daemon. We do things this way to ensure that the main PID
          * of the daemon is the one we initially fork()ed. */
 
-        if ((pam_code = pam_start(name, user, &conv, &handle)) != PAM_SUCCESS) {
+        if (log_get_max_level() < LOG_PRI(LOG_DEBUG))
+                flags |= PAM_SILENT;
+
+        pam_code = pam_start(name, user, &conv, &handle);
+        if (pam_code != PAM_SUCCESS) {
                 handle = NULL;
                 goto fail;
         }
 
-        if (tty)
-                if ((pam_code = pam_set_item(handle, PAM_TTY, tty)) != PAM_SUCCESS)
+        if (tty) {
+                pam_code = pam_set_item(handle, PAM_TTY, tty);
+                if (pam_code != PAM_SUCCESS)
                         goto fail;
+        }
 
-        if ((pam_code = pam_acct_mgmt(handle, PAM_SILENT)) != PAM_SUCCESS)
+        pam_code = pam_acct_mgmt(handle, flags);
+        if (pam_code != PAM_SUCCESS)
                 goto fail;
 
-        if ((pam_code = pam_open_session(handle, PAM_SILENT)) != PAM_SUCCESS)
+        pam_code = pam_open_session(handle, flags);
+        if (pam_code != PAM_SUCCESS)
                 goto fail;
 
         close_session = true;
 
-        if ((!(e = pam_getenvlist(handle)))) {
+        e = pam_getenvlist(handle);
+        if (!e) {
                 pam_code = PAM_BUF_ERR;
                 goto fail;
         }
@@ -765,7 +789,8 @@ static int setup_pam(
 
         parent_pid = getpid();
 
-        if ((pam_pid = fork()) < 0)
+        pam_pid = fork();
+        if (pam_pid < 0)
                 goto fail;
 
         if (pam_pid == 0) {
@@ -816,14 +841,16 @@ static int setup_pam(
                 }
 
                 /* If our parent died we'll end the session */
-                if (getppid() != parent_pid)
-                        if ((pam_code = pam_close_session(handle, PAM_DATA_SILENT)) != PAM_SUCCESS)
+                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 | PAM_DATA_SILENT);
+                pam_end(handle, pam_code | flags);
                 _exit(r);
         }
 
@@ -845,16 +872,19 @@ static int setup_pam(
         return 0;
 
 fail:
-        if (pam_code != PAM_SUCCESS)
+        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
+        } else {
+                log_error("PAM failed: %m");
                 err = -errno;
+        }
 
         if (handle) {
                 if (close_session)
-                        pam_code = pam_close_session(handle, PAM_DATA_SILENT);
+                        pam_code = pam_close_session(handle, flags);
 
-                pam_end(handle, pam_code | PAM_DATA_SILENT);
+                pam_end(handle, pam_code | flags);
         }
 
         strv_free(e);
@@ -870,68 +900,6 @@ fail:
 }
 #endif
 
-static int do_capability_bounding_set_drop(uint64_t drop) {
-        unsigned long i;
-        cap_t old_cap = NULL, new_cap = NULL;
-        cap_flag_value_t fv;
-        int r;
-
-        /* If we are run as PID 1 we will lack CAP_SETPCAP by default
-         * in the effective set (yes, the kernel drops that when
-         * executing init!), so get it back temporarily so that we can
-         * call PR_CAPBSET_DROP. */
-
-        old_cap = cap_get_proc();
-        if (!old_cap)
-                return -errno;
-
-        if (cap_get_flag(old_cap, CAP_SETPCAP, CAP_EFFECTIVE, &fv) < 0) {
-                r = -errno;
-                goto finish;
-        }
-
-        if (fv != CAP_SET) {
-                static const cap_value_t v = CAP_SETPCAP;
-
-                new_cap = cap_dup(old_cap);
-                if (!new_cap) {
-                        r = -errno;
-                        goto finish;
-                }
-
-                if (cap_set_flag(new_cap, CAP_EFFECTIVE, 1, &v, CAP_SET) < 0) {
-                        r = -errno;
-                        goto finish;
-                }
-
-                if (cap_set_proc(new_cap) < 0) {
-                        r = -errno;
-                        goto finish;
-                }
-        }
-
-        for (i = 0; i <= cap_last_cap(); i++)
-                if (drop & ((uint64_t) 1ULL << (uint64_t) i)) {
-                        if (prctl(PR_CAPBSET_DROP, i) < 0) {
-                                r = -errno;
-                                goto finish;
-                        }
-                }
-
-        r = 0;
-
-finish:
-        if (new_cap)
-                cap_free(new_cap);
-
-        if (old_cap) {
-                cap_set_proc(old_cap);
-                cap_free(old_cap);
-        }
-
-        return r;
-}
-
 static void rename_process_from_path(const char *path) {
         char process_name[11];
         const char *p;
@@ -963,26 +931,108 @@ static void rename_process_from_path(const char *path) {
         rename_process(process_name);
 }
 
+static int apply_seccomp(uint32_t *syscall_filter) {
+        static const struct sock_filter header[] = {
+                VALIDATE_ARCHITECTURE,
+                EXAMINE_SYSCALL
+        };
+        static const struct sock_filter footer[] = {
+                _KILL_PROCESS
+        };
+
+        int i;
+        unsigned n;
+        struct sock_filter *f;
+        struct sock_fprog prog = {};
+
+        assert(syscall_filter);
+
+        /* First: count the syscalls to check for */
+        for (i = 0, n = 0; i < syscall_max(); i++)
+                if (syscall_filter[i >> 4] & (1 << (i & 31)))
+                        n++;
+
+        /* Second: build the filter program from a header the syscall
+         * matches and the footer */
+        f = alloca(sizeof(struct sock_filter) * (ELEMENTSOF(header) + 2*n + ELEMENTSOF(footer)));
+        memcpy(f, header, sizeof(header));
+
+        for (i = 0, n = 0; i < syscall_max(); i++)
+                if (syscall_filter[i >> 4] & (1 << (i & 31))) {
+                        struct sock_filter item[] = {
+                                BPF_JUMP(BPF_JMP+BPF_JEQ+BPF_K, INDEX_TO_SYSCALL(i), 0, 1),
+                                BPF_STMT(BPF_RET+BPF_K, SECCOMP_RET_ALLOW)
+                        };
+
+                        assert_cc(ELEMENTSOF(item) == 2);
+
+                        f[ELEMENTSOF(header) + 2*n]  = item[0];
+                        f[ELEMENTSOF(header) + 2*n+1] = item[1];
+
+                        n++;
+                }
+
+        memcpy(f + (ELEMENTSOF(header) + 2*n), footer, sizeof(footer));
+
+        /* Third: install the filter */
+        prog.len = ELEMENTSOF(header) + ELEMENTSOF(footer) + 2*n;
+        prog.filter = f;
+        if (prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog) < 0)
+                return -errno;
+
+        return 0;
+}
+
+static void do_idle_pipe_dance(int idle_pipe[4]) {
+        assert(idle_pipe);
+
+        if (idle_pipe[1] >= 0)
+                close_nointr_nofail(idle_pipe[1]);
+        if (idle_pipe[2] >= 0)
+                close_nointr_nofail(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. */
+                        write(idle_pipe[3], "x", 1);
+
+                        /* Wait for systemd to react to the signal above. */
+                        fd_wait_for_event(idle_pipe[0], POLLHUP, IDLE_TIMEOUT2_USEC);
+                }
+
+                close_nointr_nofail(idle_pipe[0]);
+
+        }
+
+        if (idle_pipe[3] >= 0)
+                close_nointr_nofail(idle_pipe[3]);
+}
+
 int exec_spawn(ExecCommand *command,
                char **argv,
-               const ExecContext *context,
+               ExecContext *context,
                int fds[], unsigned n_fds,
                char **environment,
                bool apply_permissions,
                bool apply_chroot,
                bool apply_tty_stdin,
                bool confirm_spawn,
-               CGroupBonding *cgroup_bondings,
-               CGroupAttribute *cgroup_attributes,
-               const char *cgroup_suffix,
-               int idle_pipe[2],
+               CGroupControllerMask cgroup_supported,
+               const char *cgroup_path,
+               const char *unit_id,
+               int idle_pipe[4],
+               ExecRuntime *runtime,
                pid_t *ret) {
 
+        _cleanup_strv_free_ char **files_env = NULL;
+        int socket_fd;
+        char *line;
         pid_t pid;
         int r;
-        char *line;
-        int socket_fd;
-        char **files_env = NULL;
 
         assert(command);
         assert(context);
@@ -1003,43 +1053,43 @@ int exec_spawn(ExecCommand *command,
         } else
                 socket_fd = -1;
 
-        if ((r = exec_context_load_environment(context, &files_env)) < 0) {
-                log_error("Failed to load environment files: %s", strerror(-r));
+        r = exec_context_load_environment(context, &files_env);
+        if (r < 0) {
+                log_struct_unit(LOG_ERR,
+                           unit_id,
+                           "MESSAGE=Failed to load environment files: %s", strerror(-r),
+                           "ERRNO=%d", -r,
+                           NULL);
                 return r;
         }
 
         if (!argv)
                 argv = command->argv;
 
-        if (!(line = exec_command_line(argv))) {
-                r = -ENOMEM;
-                goto fail_parent;
-        }
+        line = exec_command_line(argv);
+        if (!line)
+                return log_oom();
 
-        log_debug("About to execute: %s", line);
+        log_struct_unit(LOG_DEBUG,
+                        unit_id,
+                        "EXECUTABLE=%s", command->path,
+                        "MESSAGE=About to execute: %s", line,
+                        NULL);
         free(line);
 
-        r = cgroup_bonding_realize_list(cgroup_bondings);
-        if (r < 0)
-                goto fail_parent;
-
-        cgroup_attribute_apply_list(cgroup_attributes, cgroup_bondings);
-
-        if ((pid = fork()) < 0) {
-                r = -errno;
-                goto fail_parent;
-        }
+        pid = fork();
+        if (pid < 0)
+                return -errno;
 
         if (pid == 0) {
-                int i, err;
-                sigset_t ss;
-                const char *username = NULL, *home = NULL;
+                _cleanup_strv_free_ char **our_env = NULL, **pam_env = NULL, **final_env = NULL, **final_argv = NULL;
+                const char *username = NULL, *home = NULL, *shell = NULL;
+                unsigned n_dont_close = 0, n_env = 0;
+                int dont_close[n_fds + 3];
                 uid_t uid = (uid_t) -1;
                 gid_t gid = (gid_t) -1;
-                char **our_env = NULL, **pam_env = NULL, **final_env = NULL, **final_argv = NULL;
-                unsigned n_env = 0;
-                int saved_stdout = -1, saved_stdin = -1;
-                bool keep_stdout = false, keep_stdin = false, set_access = false;
+                sigset_t ss;
+                int i, err;
 
                 /* child */
 
@@ -1063,21 +1113,28 @@ int exec_spawn(ExecCommand *command,
                         goto fail_child;
                 }
 
-                if (idle_pipe) {
-                        if (idle_pipe[1] >= 0)
-                                close_nointr_nofail(idle_pipe[1]);
-                        if (idle_pipe[0] >= 0) {
-                                fd_wait_for_event(idle_pipe[0], POLLHUP, IDLE_TIMEOUT_USEC);
-                                close_nointr_nofail(idle_pipe[0]);
-                        }
-                }
+                if (idle_pipe)
+                        do_idle_pipe_dance(idle_pipe);
 
                 /* Close sockets very early to make sure we don't
                  * block init reexecution because it cannot bind its
                  * sockets */
                 log_forget_fds();
-                err = close_all_fds(socket_fd >= 0 ? &socket_fd : fds,
-                                           socket_fd >= 0 ? 1 : n_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 (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];
+                }
+
+                err = close_all_fds(dont_close, n_dont_close);
                 if (err < 0) {
                         r = EXIT_FDS;
                         goto fail_child;
@@ -1109,44 +1166,24 @@ int exec_spawn(ExecCommand *command,
 
                 exec_context_tty_reset(context);
 
-                /* We skip the confirmation step if we shall not apply the TTY */
-                if (confirm_spawn &&
-                    (!is_terminal_input(context->std_input) || apply_tty_stdin)) {
+                if (confirm_spawn) {
                         char response;
 
-                        /* Set up terminal for the question */
-                        if ((r = setup_confirm_stdio(context,
-                                                     &saved_stdin, &saved_stdout))) {
-                                err = -errno;
-                                goto fail_child;
-                        }
-
-                        /* Now ask the question. */
-                        if (!(line = exec_command_line(argv))) {
-                                err = -ENOMEM;
-                                r = EXIT_MEMORY;
-                                goto fail_child;
-                        }
-
-                        r = ask(&response, "yns", "Execute %s? [Yes, No, Skip] ", line);
-                        free(line);
-
-                        if (r < 0 || response == 'n') {
+                        err = ask_for_confirmation(&response, argv);
+                        if (err == -ETIMEDOUT)
+                                write_confirm_message("Confirmation question timed out, assuming positive response.\n");
+                        else if (err < 0)
+                                write_confirm_message("Couldn't ask confirmation question, assuming positive response: %s\n", strerror(-err));
+                        else if (response == 's') {
+                                write_confirm_message("Skipping execution.\n");
                                 err = -ECANCELED;
                                 r = EXIT_CONFIRM;
                                 goto fail_child;
-                        } else if (response == 's') {
+                        } else if (response == 'n') {
+                                write_confirm_message("Failing execution.\n");
                                 err = r = 0;
                                 goto fail_child;
                         }
-
-                        /* Release terminal for the question */
-                        if ((r = restore_confirm_stdio(context,
-                                                       &saved_stdin, &saved_stdout,
-                                                       &keep_stdin, &keep_stdout))) {
-                                err = -errno;
-                                goto fail_child;
-                        }
                 }
 
                 /* If a socket is connected to STDIN/STDOUT/STDERR, we
@@ -1154,30 +1191,26 @@ int exec_spawn(ExecCommand *command,
                 if (socket_fd >= 0)
                         fd_nonblock(socket_fd, false);
 
-                if (!keep_stdin) {
-                        err = setup_input(context, socket_fd, apply_tty_stdin);
-                        if (err < 0) {
-                                r = EXIT_STDIN;
-                                goto fail_child;
-                        }
+                err = setup_input(context, socket_fd, apply_tty_stdin);
+                if (err < 0) {
+                        r = EXIT_STDIN;
+                        goto fail_child;
                 }
 
-                if (!keep_stdout) {
-                        err = setup_output(context, socket_fd, path_get_file_name(command->path), apply_tty_stdin);
-                        if (err < 0) {
-                                r = EXIT_STDOUT;
-                                goto fail_child;
-                        }
+                err = setup_output(context, STDOUT_FILENO, socket_fd, path_get_file_name(command->path), unit_id, apply_tty_stdin);
+                if (err < 0) {
+                        r = EXIT_STDOUT;
+                        goto fail_child;
                 }
 
-                err = setup_error(context, socket_fd, path_get_file_name(command->path), apply_tty_stdin);
+                err = setup_output(context, STDERR_FILENO, socket_fd, path_get_file_name(command->path), unit_id, apply_tty_stdin);
                 if (err < 0) {
                         r = EXIT_STDERR;
                         goto fail_child;
                 }
 
-                if (cgroup_bondings) {
-                        err = cgroup_bonding_install_list(cgroup_bondings, 0, cgroup_suffix);
+                if (cgroup_path) {
+                        err = cg_attach_everywhere(cgroup_supported, cgroup_path, 0);
                         if (err < 0) {
                                 r = EXIT_CGROUP;
                                 goto fail_child;
@@ -1190,23 +1223,10 @@ int exec_spawn(ExecCommand *command,
                         snprintf(t, sizeof(t), "%i", context->oom_score_adjust);
                         char_array_0(t);
 
-                        if (write_one_line_file("/proc/self/oom_score_adj", t) < 0) {
-                                /* Compatibility with Linux <= 2.6.35 */
-
-                                int adj;
-
-                                adj = (context->oom_score_adjust * -OOM_DISABLE) / OOM_SCORE_ADJ_MAX;
-                                adj = CLAMP(adj, OOM_DISABLE, OOM_ADJUST_MAX);
-
-                                snprintf(t, sizeof(t), "%i", adj);
-                                char_array_0(t);
-
-                                if (write_one_line_file("/proc/self/oom_adj", t) < 0
-                                    && errno != EACCES) {
-                                        err = -errno;
-                                        r = EXIT_OOM_ADJUST;
-                                        goto fail_child;
-                                }
+                        if (write_string_file("/proc/self/oom_score_adj", t) < 0) {
+                                err = -errno;
+                                r = EXIT_OOM_ADJUST;
+                                goto fail_child;
                         }
                 }
 
@@ -1218,13 +1238,16 @@ int exec_spawn(ExecCommand *command,
                         }
 
                 if (context->cpu_sched_set) {
-                        struct sched_param param;
-
-                        zero(param);
-                        param.sched_priority = context->cpu_sched_priority;
-
-                        if (sched_setscheduler(0, context->cpu_sched_policy |
-                                               (context->cpu_sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0), &param) < 0) {
+                        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) {
                                 err = -errno;
                                 r = EXIT_SETSCHEDULER;
                                 goto fail_child;
@@ -1245,7 +1268,7 @@ int exec_spawn(ExecCommand *command,
                                 goto fail_child;
                         }
 
-                if (context->timer_slack_nsec_set)
+                if (context->timer_slack_nsec != (nsec_t) -1)
                         if (prctl(PR_SET_TIMERSLACK, context->timer_slack_nsec) < 0) {
                                 err = -errno;
                                 r = EXIT_TIMERSLACK;
@@ -1257,7 +1280,7 @@ int exec_spawn(ExecCommand *command,
 
                 if (context->user) {
                         username = context->user;
-                        err = get_user_creds(&username, &uid, &gid, &home);
+                        err = get_user_creds(&username, &uid, &gid, &home, &shell);
                         if (err < 0) {
                                 r = EXIT_USER;
                                 goto fail_child;
@@ -1270,27 +1293,24 @@ int exec_spawn(ExecCommand *command,
                                         goto fail_child;
                                 }
                         }
+                }
 
-                        if (cgroup_bondings && context->control_group_modify) {
-                                err = cgroup_bonding_set_group_access_list(cgroup_bondings, 0755, uid, gid);
-                                if (err >= 0)
-                                        err = cgroup_bonding_set_task_access_list(cgroup_bondings, 0644, uid, gid, context->control_group_persistent);
-                                if (err < 0) {
-                                        r = EXIT_CGROUP;
-                                        goto fail_child;
-                                }
-
-                                set_access = true;
+#ifdef HAVE_PAM
+                if (cgroup_path && context->user && context->pam_name) {
+                        err = cg_set_task_access(SYSTEMD_CGROUP_CONTROLLER, cgroup_path, 0644, uid, gid);
+                        if (err < 0) {
+                                r = EXIT_CGROUP;
+                                goto fail_child;
                         }
-                }
 
-                if (cgroup_bondings && !set_access && context->control_group_persistent >= 0)  {
-                        err = cgroup_bonding_set_task_access_list(cgroup_bondings, (mode_t) -1, (uid_t) -1, (uid_t) -1, context->control_group_persistent);
+
+                        err = cg_set_group_access(SYSTEMD_CGROUP_CONTROLLER, cgroup_path, 0755, uid, gid);
                         if (err < 0) {
                                 r = EXIT_CGROUP;
                                 goto fail_child;
                         }
                 }
+#endif
 
                 if (apply_permissions) {
                         err = enforce_groups(context, username, gid);
@@ -1303,7 +1323,7 @@ int exec_spawn(ExecCommand *command,
                 umask(context->umask);
 
 #ifdef HAVE_PAM
-                if (context->pam_name && username) {
+                if (apply_permissions && context->pam_name && username) {
                         err = setup_pam(context->pam_name, username, uid, context->tty_path, &pam_env, fds, n_fds);
                         if (err < 0) {
                                 r = EXIT_PAM;
@@ -1311,26 +1331,43 @@ int exec_spawn(ExecCommand *command,
                         }
                 }
 #endif
-                if (context->private_network) {
-                        if (unshare(CLONE_NEWNET) < 0) {
-                                err = -errno;
+                if (context->private_network && runtime && runtime->netns_storage_socket[0] >= 0) {
+                        err = setup_netns(runtime->netns_storage_socket);
+                        if (err < 0) {
                                 r = EXIT_NETWORK;
                                 goto fail_child;
                         }
-
-                        loopback_setup();
                 }
 
-                if (strv_length(context->read_write_dirs) > 0 ||
-                    strv_length(context->read_only_dirs) > 0 ||
-                    strv_length(context->inaccessible_dirs) > 0 ||
-                    context->mount_flags != MS_SHARED ||
-                    context->private_tmp) {
-                        err = setup_namespace(context->read_write_dirs,
-                                              context->read_only_dirs,
-                                              context->inaccessible_dirs,
-                                              context->private_tmp,
-                                              context->mount_flags);
+                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))) {
+
+                        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 = strappenda(runtime->tmp_dir, "/tmp");
+                                if (runtime->var_tmp_dir)
+                                        var = strappenda(runtime->var_tmp_dir, "/tmp");
+                        }
+
+                        err = setup_namespace(
+                                        context->read_write_dirs,
+                                        context->read_only_dirs,
+                                        context->inaccessible_dirs,
+                                        tmp,
+                                        var,
+                                        context->mount_flags);
+
                         if (err < 0) {
                                 r = EXIT_NAMESPACE;
                                 goto fail_child;
@@ -1351,8 +1388,7 @@ int exec_spawn(ExecCommand *command,
                                 goto fail_child;
                         }
                 } else {
-
-                        char *d;
+                        _cleanup_free_ char *d = NULL;
 
                         if (asprintf(&d, "%s/%s",
                                      context->root_directory ? context->root_directory : "",
@@ -1364,12 +1400,9 @@ int exec_spawn(ExecCommand *command,
 
                         if (chdir(d) < 0) {
                                 err = -errno;
-                                free(d);
                                 r = EXIT_CHDIR;
                                 goto fail_child;
                         }
-
-                        free(d);
                 }
 
                 /* We repeat the fd closing here, to make sure that
@@ -1398,7 +1431,7 @@ int exec_spawn(ExecCommand *command,
                         }
 
                         if (context->capability_bounding_set_drop) {
-                                err = do_capability_bounding_set_drop(context->capability_bounding_set_drop);
+                                err = capability_bounding_set_drop(context->capability_bounding_set_drop, false);
                                 if (err < 0) {
                                         r = EXIT_CAPABILITIES;
                                         goto fail_child;
@@ -1430,62 +1463,61 @@ int exec_spawn(ExecCommand *command,
                                         r = EXIT_CAPABILITIES;
                                         goto fail_child;
                                 }
+
+                        if (context->no_new_privileges)
+                                if (prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0) < 0) {
+                                        err = -errno;
+                                        r = EXIT_NO_NEW_PRIVILEGES;
+                                        goto fail_child;
+                                }
+
+                        if (context->syscall_filter) {
+                                err = apply_seccomp(context->syscall_filter);
+                                if (err < 0) {
+                                        r = EXIT_SECCOMP;
+                                        goto fail_child;
+                                }
+                        }
                 }
 
-                if (!(our_env = new0(char*, 7))) {
+                our_env = new(char*, 8);
+                if (!our_env ||
+                    (n_fds > 0 && (
+                            asprintf(our_env + n_env++, "LISTEN_PID=%lu", (unsigned long) getpid()) < 0 ||
+                            asprintf(our_env + n_env++, "LISTEN_FDS=%u", n_fds) < 0)) ||
+                    (home && asprintf(our_env + n_env++, "HOME=%s", home) < 0) ||
+                    (username && (
+                            asprintf(our_env + n_env++, "LOGNAME=%s", username) < 0 ||
+                            asprintf(our_env + n_env++, "USER=%s", username) < 0)) ||
+                    (shell && asprintf(our_env + n_env++, "SHELL=%s", shell) < 0) ||
+                    ((is_terminal_input(context->std_input) ||
+                      context->std_output == EXEC_OUTPUT_TTY ||
+                      context->std_error == EXEC_OUTPUT_TTY) && (
+                              !(our_env[n_env++] = strdup(default_term_for_tty(tty_path(context))))))) {
+
                         err = -ENOMEM;
                         r = EXIT_MEMORY;
                         goto fail_child;
                 }
 
-                if (n_fds > 0)
-                        if (asprintf(our_env + n_env++, "LISTEN_PID=%lu", (unsigned long) getpid()) < 0 ||
-                            asprintf(our_env + n_env++, "LISTEN_FDS=%u", n_fds) < 0) {
-                                err = -ENOMEM;
-                                r = EXIT_MEMORY;
-                                goto fail_child;
-                        }
-
-                if (home)
-                        if (asprintf(our_env + n_env++, "HOME=%s", home) < 0) {
-                                err = -ENOMEM;
-                                r = EXIT_MEMORY;
-                                goto fail_child;
-                        }
-
-                if (username)
-                        if (asprintf(our_env + n_env++, "LOGNAME=%s", username) < 0 ||
-                            asprintf(our_env + n_env++, "USER=%s", username) < 0) {
-                                err = -ENOMEM;
-                                r = EXIT_MEMORY;
-                                goto fail_child;
-                        }
-
-                if (is_terminal_input(context->std_input) ||
-                    context->std_output == EXEC_OUTPUT_TTY ||
-                    context->std_error == EXEC_OUTPUT_TTY)
-                        if (!(our_env[n_env++] = strdup(default_term_for_tty(tty_path(context))))) {
-                                err = -ENOMEM;
-                                r = EXIT_MEMORY;
-                                goto fail_child;
-                        }
-
-                assert(n_env <= 7);
-
-                if (!(final_env = strv_env_merge(
-                                      5,
-                                      environment,
-                                      our_env,
-                                      context->environment,
-                                      files_env,
-                                      pam_env,
-                                      NULL))) {
+                our_env[n_env++] = NULL;
+                assert(n_env <= 8);
+
+                final_env = strv_env_merge(5,
+                                           environment,
+                                           our_env,
+                                           context->environment,
+                                           files_env,
+                                           pam_env,
+                                           NULL);
+                if (!final_env) {
                         err = -ENOMEM;
                         r = EXIT_MEMORY;
                         goto fail_child;
                 }
 
-                if (!(final_argv = replace_env_argv(argv, final_env))) {
+                final_argv = replace_env_argv(argv, final_env);
+                if (!final_argv) {
                         err = -ENOMEM;
                         r = EXIT_MEMORY;
                         goto fail_child;
@@ -1493,6 +1525,20 @@ int exec_spawn(ExecCommand *command,
 
                 final_env = strv_env_clean(final_env);
 
+                if (_unlikely_(log_get_max_level() >= LOG_PRI(LOG_DEBUG))) {
+                        line = exec_command_line(final_argv);
+                        if (line) {
+                                log_open();
+                                log_struct_unit(LOG_DEBUG,
+                                                unit_id,
+                                                "EXECUTABLE=%s", command->path,
+                                                "MESSAGE=Executing: %s", line,
+                                                NULL);
+                                log_close();
+                                free(line);
+                                line = NULL;
+                        }
+                }
                 execve(command->path, final_argv, final_env);
                 err = -errno;
                 r = EXIT_EXEC;
@@ -1500,47 +1546,37 @@ int exec_spawn(ExecCommand *command,
         fail_child:
                 if (r != 0) {
                         log_open();
-                        log_warning("Failed at step %s spawning %s: %s",
-                                    exit_status_to_string(r, EXIT_STATUS_SYSTEMD),
-                                    command->path, strerror(-err));
+                        log_struct(LOG_ERR, MESSAGE_ID(SD_MESSAGE_SPAWN_FAILED),
+                                   "EXECUTABLE=%s", command->path,
+                                   "MESSAGE=Failed at step %s spawning %s: %s",
+                                          exit_status_to_string(r, EXIT_STATUS_SYSTEMD),
+                                          command->path, strerror(-err),
+                                   "ERRNO=%d", -err,
+                                   NULL);
+                        log_close();
                 }
 
-                strv_free(our_env);
-                strv_free(final_env);
-                strv_free(pam_env);
-                strv_free(files_env);
-                strv_free(final_argv);
-
-                if (saved_stdin >= 0)
-                        close_nointr_nofail(saved_stdin);
-
-                if (saved_stdout >= 0)
-                        close_nointr_nofail(saved_stdout);
-
                 _exit(r);
         }
 
-        strv_free(files_env);
+        log_struct_unit(LOG_DEBUG,
+                        unit_id,
+                        "MESSAGE=Forked %s as %lu",
+                        command->path, (unsigned long) pid,
+                        NULL);
 
         /* 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 (cgroup_bondings)
-                cgroup_bonding_install_list(cgroup_bondings, pid, cgroup_suffix);
-
-        log_debug("Forked %s as %lu", command->path, (unsigned long) pid);
+        if (cgroup_path)
+                cg_attach(SYSTEMD_CGROUP_CONTROLLER, cgroup_path, pid);
 
         exec_status_start(&command->exec_status, pid);
 
         *ret = pid;
         return 0;
-
-fail_parent:
-        strv_free(files_env);
-
-        return r;
 }
 
 void exec_context_init(ExecContext *c) {
@@ -1551,11 +1587,8 @@ void exec_context_init(ExecContext *c) {
         c->cpu_sched_policy = SCHED_OTHER;
         c->syslog_priority = LOG_DAEMON|LOG_INFO;
         c->syslog_level_prefix = true;
-        c->mount_flags = MS_SHARED;
-        c->kill_signal = SIGTERM;
-        c->send_sigkill = true;
-        c->control_group_persistent = -1;
         c->ignore_sigpipe = true;
+        c->timer_slack_nsec = (nsec_t) -1;
 }
 
 void exec_context_done(ExecContext *c) {
@@ -1619,6 +1652,9 @@ void exec_context_done(ExecContext *c) {
 
         free(c->utmp_id);
         c->utmp_id = NULL;
+
+        free(c->syscall_filter);
+        c->syscall_filter = NULL;
 }
 
 void exec_command_done(ExecCommand *c) {
@@ -1642,7 +1678,7 @@ void exec_command_free_list(ExecCommand *c) {
         ExecCommand *i;
 
         while ((i = c)) {
-                LIST_REMOVE(ExecCommand, command, c, i);
+                LIST_REMOVE(command, c, i);
                 exec_command_done(i);
                 free(i);
         }
@@ -1668,6 +1704,8 @@ int exec_context_load_environment(const ExecContext *c, char ***l) {
                 int k;
                 bool ignore = false;
                 char **p;
+                _cleanup_globfree_ glob_t pglob = {};
+                int count, n;
 
                 fn = *i;
 
@@ -1677,7 +1715,6 @@ int exec_context_load_environment(const ExecContext *c, char ***l) {
                 }
 
                 if (!path_is_absolute(fn)) {
-
                         if (ignore)
                                 continue;
 
@@ -1685,28 +1722,49 @@ int exec_context_load_environment(const ExecContext *c, char ***l) {
                         return -EINVAL;
                 }
 
-                if ((k = load_env_file(fn, &p)) < 0) {
-
+                /* Filename supports globbing, take all matching files */
+                errno = 0;
+                if (glob(fn, 0, NULL, &pglob) != 0) {
                         if (ignore)
                                 continue;
 
                         strv_free(r);
-                        return k;
+                        return errno ? -errno : -EINVAL;
                 }
+                count = pglob.gl_pathc;
+                if (count == 0) {
+                        if (ignore)
+                                continue;
 
-                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;
+                        return -EINVAL;
+                }
+                for (n = 0; n < count; n++) {
+                        k = load_env_file(pglob.gl_pathv[n], NULL, &p);
+                        if (k < 0) {
+                                if (ignore)
+                                        continue;
 
-                        r = m;
+                                strv_free(r);
+                                return k;
+                        }
+                        /* Log invalid environment variables with filename */
+                        if (p)
+                                p = strv_env_clean_log(p, pglob.gl_pathv[n]);
+
+                        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;
+                        }
                 }
         }
 
@@ -1715,6 +1773,37 @@ int exec_context_load_environment(const ExecContext *c, char ***l) {
         return 0;
 }
 
+static bool tty_may_match_dev_console(const char *tty) {
+        char *active = NULL, *console;
+        bool b;
+
+        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 */
+        b = streq(console, tty) || (streq(console, "tty0") && tty_is_vc(tty));
+        free(active);
+
+        return b;
+}
+
+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;
 
@@ -1725,14 +1814,13 @@ static void strv_fprintf(FILE *f, char **l) {
 }
 
 void exec_context_dump(ExecContext *c, FILE* f, const char *prefix) {
-        char ** e;
+        char **e;
         unsigned i;
 
         assert(c);
         assert(f);
 
-        if (!prefix)
-                prefix = "";
+        prefix = strempty(prefix);
 
         fprintf(f,
                 "%sUMask: %04o\n"
@@ -1740,17 +1828,15 @@ void exec_context_dump(ExecContext *c, FILE* f, const char *prefix) {
                 "%sRootDirectory: %s\n"
                 "%sNonBlocking: %s\n"
                 "%sPrivateTmp: %s\n"
-                "%sControlGroupModify: %s\n"
-                "%sControlGroupPersistent: %s\n"
-                "%sPrivateNetwork: %s\n",
+                "%sPrivateNetwork: %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->control_group_modify),
-                prefix, yes_no(c->control_group_persistent),
-                prefix, yes_no(c->private_network));
+                prefix, yes_no(c->private_network),
+                prefix, yes_no(c->ignore_sigpipe));
 
         STRV_FOREACH(e, c->environment)
                 fprintf(f, "%sEnvironment: %s\n", prefix, *e);
@@ -1777,21 +1863,37 @@ void exec_context_dump(ExecContext *c, FILE* f, const char *prefix) {
                 if (c->rlimit[i])
                         fprintf(f, "%s%s: %llu\n", prefix, rlimit_to_string(i), (unsigned long long) c->rlimit[i]->rlim_max);
 
-        if (c->ioprio_set)
+        if (c->ioprio_set) {
+                char *class_str;
+                int r;
+
+                r = ioprio_class_to_string_alloc(IOPRIO_PRIO_CLASS(c->ioprio), &class_str);
+                if (r < 0)
+                        class_str = NULL;
                 fprintf(f,
                         "%sIOSchedulingClass: %s\n"
                         "%sIOPriority: %i\n",
-                        prefix, ioprio_class_to_string(IOPRIO_PRIO_CLASS(c->ioprio)),
+                        prefix, strna(class_str),
                         prefix, (int) IOPRIO_PRIO_DATA(c->ioprio));
+                free(class_str);
+        }
 
-        if (c->cpu_sched_set)
+        if (c->cpu_sched_set) {
+                char *policy_str;
+                int r;
+
+                r = sched_policy_to_string_alloc(c->cpu_sched_policy, &policy_str);
+                if (r < 0)
+                        policy_str = NULL;
                 fprintf(f,
                         "%sCPUSchedulingPolicy: %s\n"
                         "%sCPUSchedulingPriority: %i\n"
                         "%sCPUSchedulingResetOnFork: %s\n",
-                        prefix, sched_policy_to_string(c->cpu_sched_policy),
+                        prefix, strna(policy_str),
                         prefix, c->cpu_sched_priority,
                         prefix, yes_no(c->cpu_sched_reset_on_fork));
+                free(policy_str);
+        }
 
         if (c->cpuset) {
                 fprintf(f, "%sCPUAffinity:", prefix);
@@ -1801,8 +1903,8 @@ void exec_context_dump(ExecContext *c, FILE* f, const char *prefix) {
                 fputs("\n", f);
         }
 
-        if (c->timer_slack_nsec_set)
-                fprintf(f, "%sTimerSlackNSec: %lu\n", prefix, c->timer_slack_nsec);
+        if (c->timer_slack_nsec != (nsec_t) -1)
+                fprintf(f, "%sTimerSlackNSec: %lu\n", prefix, (unsigned long)c->timer_slack_nsec);
 
         fprintf(f,
                 "%sStandardInput: %s\n"
@@ -1826,12 +1928,26 @@ void exec_context_dump(ExecContext *c, FILE* f, const char *prefix) {
         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)
+            c->std_error == EXEC_OUTPUT_SYSLOG_AND_CONSOLE || c->std_error == EXEC_OUTPUT_KMSG_AND_CONSOLE || c->std_error == EXEC_OUTPUT_JOURNAL_AND_CONSOLE) {
+                char *fac_str, *lvl_str;
+                int r;
+
+                r = log_facility_unshifted_to_string_alloc(c->syslog_priority >> 3, &fac_str);
+                if (r < 0)
+                        fac_str = NULL;
+
+                r = log_level_to_string_alloc(LOG_PRI(c->syslog_priority), &lvl_str);
+                if (r < 0)
+                        lvl_str = NULL;
+
                 fprintf(f,
                         "%sSyslogFacility: %s\n"
                         "%sSyslogLevel: %s\n",
-                        prefix, log_facility_unshifted_to_string(c->syslog_priority >> 3),
-                        prefix, log_level_to_string(LOG_PRI(c->syslog_priority)));
+                        prefix, strna(fac_str),
+                        prefix, strna(lvl_str));
+                free(lvl_str);
+                free(fac_str);
+        }
 
         if (c->capabilities) {
                 char *t;
@@ -1845,12 +1961,12 @@ void exec_context_dump(ExecContext *c, FILE* f, const char *prefix) {
         if (c->secure_bits)
                 fprintf(f, "%sSecure Bits:%s%s%s%s%s%s\n",
                         prefix,
-                        (c->secure_bits & SECURE_KEEP_CAPS) ? " keep-caps" : "",
-                        (c->secure_bits & SECURE_KEEP_CAPS_LOCKED) ? " keep-caps-locked" : "",
-                        (c->secure_bits & SECURE_NO_SETUID_FIXUP) ? " no-setuid-fixup" : "",
-                        (c->secure_bits & SECURE_NO_SETUID_FIXUP_LOCKED) ? " no-setuid-fixup-locked" : "",
-                        (c->secure_bits & SECURE_NOROOT) ? " noroot" : "",
-                        (c->secure_bits & SECURE_NOROOT_LOCKED) ? "noroot-locked" : "");
+                        (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;
@@ -1901,16 +2017,6 @@ void exec_context_dump(ExecContext *c, FILE* f, const char *prefix) {
                 fputs("\n", f);
         }
 
-        fprintf(f,
-                "%sKillMode: %s\n"
-                "%sKillSignal: SIG%s\n"
-                "%sSendSIGKILL: %s\n"
-                "%sIgnoreSIGPIPE: %s\n",
-                prefix, kill_mode_to_string(c->kill_mode),
-                prefix, signal_to_string(c->kill_signal),
-                prefix, yes_no(c->send_sigkill),
-                prefix, yes_no(c->ignore_sigpipe));
-
         if (c->utmp_id)
                 fprintf(f,
                         "%sUtmpIdentifier: %s\n",
@@ -2060,8 +2166,8 @@ void exec_command_append_list(ExecCommand **l, ExecCommand *e) {
 
         if (*l) {
                 /* It's kind of important, that we keep the order here */
-                LIST_FIND_TAIL(ExecCommand, command, *l, end);
-                LIST_INSERT_AFTER(ExecCommand, command, *l, end, e);
+                LIST_FIND_TAIL(command, *l, end);
+                LIST_INSERT_AFTER(command, *l, end, e);
         } else
               *l = e;
 }
@@ -2080,7 +2186,8 @@ int exec_command_set(ExecCommand *c, const char *path, ...) {
         if (!l)
                 return -ENOMEM;
 
-        if (!(p = strdup(path))) {
+        p = strdup(path);
+        if (!p) {
                 strv_free(l);
                 return -ENOMEM;
         }
@@ -2094,6 +2201,216 @@ int exec_command_set(ExecCommand *c, const char *path, ...) {
         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);
+                close_pipe(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_debug_unit(u->id, "Failed to parse netns socket value %s", value);
+                else {
+                        if ((*rt)->netns_storage_socket[0] >= 0)
+                                close_nointr_nofail((*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_debug_unit(u->id, "Failed to parse netns socket value %s", value);
+                else {
+                        if ((*rt)->netns_storage_socket[1] >= 0)
+                                close_nointr_nofail((*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) {
+        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);
+                asynchronous_job(remove_tmpdir_thread, rt->tmp_dir);
+                rt->tmp_dir = NULL;
+        }
+
+        if (rt->var_tmp_dir) {
+                log_debug("Spawning thread to nuke %s", rt->var_tmp_dir);
+                asynchronous_job(remove_tmpdir_thread, rt->var_tmp_dir);
+                rt->var_tmp_dir = NULL;
+        }
+
+        close_pipe(rt->netns_storage_socket);
+}
+
 static const char* const exec_input_table[_EXEC_INPUT_MAX] = {
         [EXEC_INPUT_NULL] = "null",
         [EXEC_INPUT_TTY] = "tty",
@@ -2118,19 +2435,3 @@ static const char* const exec_output_table[_EXEC_OUTPUT_MAX] = {
 };
 
 DEFINE_STRING_TABLE_LOOKUP(exec_output, ExecOutput);
-
-static const char* const kill_mode_table[_KILL_MODE_MAX] = {
-        [KILL_CONTROL_GROUP] = "control-group",
-        [KILL_PROCESS] = "process",
-        [KILL_NONE] = "none"
-};
-
-DEFINE_STRING_TABLE_LOOKUP(kill_mode, KillMode);
-
-static const char* const kill_who_table[_KILL_WHO_MAX] = {
-        [KILL_MAIN] = "main",
-        [KILL_CONTROL] = "control",
-        [KILL_ALL] = "all"
-};
-
-DEFINE_STRING_TABLE_LOOKUP(kill_who, KillWho);