userdbd: add new service that can merge userdb queries from multiple clients

meson.build

@@ -243,6 +243,7 @@ conf.set_quoted('SYSTEMD_EXPORT_PATH',                        join_paths(rootlib
 conf.set_quoted('VENDOR_KEYRING_PATH',                        join_paths(rootlibexecdir, 'import-pubring.gpg'))
 conf.set_quoted('USER_KEYRING_PATH',                          join_paths(pkgsysconfdir, 'import-pubring.gpg'))
 conf.set_quoted('DOCUMENT_ROOT',                              join_paths(pkgdatadir, 'gatewayd'))
+conf.set_quoted('SYSTEMD_USERWORK_PATH',                      join_paths(rootlibexecdir, 'systemd-userwork'))
 conf.set10('MEMORY_ACCOUNTING_DEFAULT',                       memory_accounting_default)
 conf.set_quoted('MEMORY_ACCOUNTING_DEFAULT_YES_NO',           memory_accounting_default ? 'yes' : 'no')
 conf.set('STATUS_UNIT_FORMAT_DEFAULT',                        'STATUS_UNIT_FORMAT_' + status_unit_format_default.to_upper())
@@ -1322,6 +1323,7 @@ foreach term : ['utmp',
                 'localed',
                 'machined',
                 'portabled',
+                'userdb',
                 'networkd',
                 'timedated',
                 'timesyncd',
@@ -1538,6 +1540,7 @@ subdir('src/kernel-install')
 subdir('src/locale')
 subdir('src/machine')
 subdir('src/portable')
+subdir('src/userdb')
 subdir('src/nspawn')
 subdir('src/resolve')
 subdir('src/timedate')
@@ -1974,6 +1977,26 @@ if conf.get('ENABLE_PORTABLED') == 1
         public_programs += exe
 endif
 
+if conf.get('ENABLE_USERDB') == 1
+        executable('systemd-userwork',
+                   systemd_userwork_sources,
+                   include_directories : includes,
+                   link_with : [libshared],
+                   dependencies : [threads],
+                   install_rpath : rootlibexecdir,
+                   install : true,
+                   install_dir : rootlibexecdir)
+
+        executable('systemd-userdbd',
+                   systemd_userdbd_sources,
+                   include_directories : includes,
+                   link_with : [libshared],
+                   dependencies : [threads],
+                   install_rpath : rootlibexecdir,
+                   install : true,
+                   install_dir : rootlibexecdir)
+endif
+
 foreach alias : ['halt', 'poweroff', 'reboot', 'runlevel', 'shutdown', 'telinit']
         meson.add_install_script(meson_make_symlink,
                                  join_paths(rootbindir, 'systemctl'),
@@ -3252,6 +3275,7 @@ foreach tuple : [
         ['logind'],
         ['machined'],
         ['portabled'],
+        ['userdb'],
         ['importd'],
         ['hostnamed'],
         ['timedated'],

meson_options.txt

@@ -94,6 +94,8 @@ option('machined', type : 'boolean',
        description : 'install the systemd-machined stack')
 option('portabled', type : 'boolean',
        description : 'install the systemd-portabled stack')
+option('userdb', type : 'boolean',
+       description : 'install the systemd-userdbd stack')
 option('networkd', type : 'boolean',
        description : 'install the systemd-networkd stack')
 option('timedated', type : 'boolean',

src/userdb/meson.build

@@ -0,0 +1,11 @@
+# SPDX-License-Identifier: LGPL-2.1+
+
+systemd_userwork_sources = files('''
+        userwork.c
+'''.split())
+
+systemd_userdbd_sources = files('''
+        userdbd-manager.c
+        userdbd-manager.h
+        userdbd.c
+'''.split())

src/userdb/userdbd-manager.c

@@ -0,0 +1,302 @@
+/* SPDX-License-Identifier: LGPL-2.1+ */
+
+#include <sys/wait.h>
+
+#include "sd-daemon.h"
+
+#include "fd-util.h"
+#include "fs-util.h"
+#include "mkdir.h"
+#include "process-util.h"
+#include "set.h"
+#include "signal-util.h"
+#include "socket-util.h"
+#include "stdio-util.h"
+#include "umask-util.h"
+#include "userdbd-manager.h"
+
+#define LISTEN_TIMEOUT_USEC (25 * USEC_PER_SEC)
+
+static int start_workers(Manager *m, bool explicit_request);
+
+static int on_sigchld(sd_event_source *s, const struct signalfd_siginfo *si, void *userdata) {
+        Manager *m = userdata;
+
+        assert(s);
+        assert(m);
+
+        for (;;) {
+                siginfo_t siginfo = {};
+                bool removed = false;
+
+                if (waitid(P_ALL, 0, &siginfo, WNOHANG|WEXITED) < 0) {
+                        if (errno == ECHILD)
+                                break;
+
+                        log_warning_errno(errno, "Failed to invoke waitid(): %m");
+                        break;
+                }
+                if (siginfo.si_pid == 0)
+                        break;
+
+                if (set_remove(m->workers_dynamic, PID_TO_PTR(siginfo.si_pid)))
+                        removed = true;
+                if (set_remove(m->workers_fixed, PID_TO_PTR(siginfo.si_pid)))
+                        removed = true;
+
+                if (!removed) {
+                        log_warning("Weird, got SIGCHLD for unknown child " PID_FMT ", ignoring.", siginfo.si_pid);
+                        continue;
+                }
+
+                if (siginfo.si_code == CLD_EXITED) {
+                        if (siginfo.si_status == EXIT_SUCCESS)
+                                log_debug("Worker " PID_FMT " exited successfully.", siginfo.si_pid);
+                        else
+                                log_warning("Worker " PID_FMT " died with a failure exit status %i, ignoring.", siginfo.si_pid, siginfo.si_status);
+                } else if (siginfo.si_code == CLD_KILLED)
+                        log_warning("Worker " PID_FMT " was killed by signal %s, ignoring.", siginfo.si_pid, signal_to_string(siginfo.si_status));
+                else if (siginfo.si_code == CLD_DUMPED)
+                        log_warning("Worker " PID_FMT " dumped core by signal %s, ignoring.", siginfo.si_pid, signal_to_string(siginfo.si_status));
+                else
+                        log_warning("Can't handle SIGCHLD of this type");
+        }
+
+        (void) start_workers(m, false); /* Fill up workers again if we fell below the low watermark */
+        return 0;
+}
+
+static int on_sigusr2(sd_event_source *s, const struct signalfd_siginfo *si, void *userdata) {
+        Manager *m = userdata;
+
+        assert(s);
+        assert(m);
+
+        (void) start_workers(m, true); /* Workers told us there's more work, let's add one more worker as long as we are below the high watermark */
+        return 0;
+}
+
+int manager_new(Manager **ret) {
+        Manager *m;
+        int r;
+
+        m = new(Manager, 1);
+        if (!m)
+                return -ENOMEM;
+
+        *m = (Manager) {
+                .listen_fd = -1,
+                .worker_ratelimit = {
+                        .interval = 5 * USEC_PER_SEC,
+                        .burst = 50,
+                },
+        };
+
+        r = sd_event_new(&m->event);
+        if (r < 0)
+                return r;
+
+        r = sd_event_add_signal(m->event, NULL, SIGINT, NULL, NULL);
+        if (r < 0)
+                return r;
+
+        r = sd_event_add_signal(m->event, NULL, SIGTERM, NULL, NULL);
+        if (r < 0)
+                return r;
+
+        (void) sd_event_set_watchdog(m->event, true);
+
+        m->workers_fixed = set_new(NULL);
+        m->workers_dynamic = set_new(NULL);
+
+        if (!m->workers_fixed || !m->workers_dynamic)
+                return -ENOMEM;
+
+        r = sd_event_add_signal(m->event, &m->sigusr2_event_source, SIGUSR2, on_sigusr2, m);
+        if (r < 0)
+                return r;
+
+        r = sd_event_add_signal(m->event, &m->sigchld_event_source, SIGCHLD, on_sigchld, m);
+        if (r < 0)
+                return r;
+
+        *ret = TAKE_PTR(m);
+        return 0;
+}
+
+Manager* manager_free(Manager *m) {
+        if (!m)
+                return NULL;
+
+        set_free(m->workers_fixed);
+        set_free(m->workers_dynamic);
+
+        sd_event_source_disable_unref(m->sigusr2_event_source);
+        sd_event_source_disable_unref(m->sigchld_event_source);
+
+        sd_event_unref(m->event);
+
+        return mfree(m);
+}
+
+static size_t manager_current_workers(Manager *m) {
+        assert(m);
+
+        return set_size(m->workers_fixed) + set_size(m->workers_dynamic);
+}
+
+static int start_one_worker(Manager *m) {
+        bool fixed;
+        pid_t pid;
+        int r;
+
+        assert(m);
+
+        fixed = set_size(m->workers_fixed) < USERDB_WORKERS_MIN;
+
+        r = safe_fork("(sd-worker)", FORK_RESET_SIGNALS|FORK_DEATHSIG|FORK_LOG, &pid);
+        if (r < 0)
+                return log_error_errno(r, "Failed to fork new worker child: %m");
+        if (r == 0) {
+                char pids[DECIMAL_STR_MAX(pid_t)];
+                /* Child */
+
+                log_close();
+
+                r = close_all_fds(&m->listen_fd, 1);
+                if (r < 0) {
+                        log_error_errno(r, "Failed to close fds in child: %m");
+                        _exit(EXIT_FAILURE);
+                }
+
+                log_open();
+
+                if (m->listen_fd == 3) {
+                        r = fd_cloexec(3, false);
+                        if (r < 0) {
+                                log_error_errno(r, "Failed to turn off O_CLOEXEC for fd 3: %m");
+                                _exit(EXIT_FAILURE);
+                        }
+                } else {
+                        if (dup2(m->listen_fd, 3) < 0) { /* dup2() creates with O_CLOEXEC off */
+                                log_error_errno(errno, "Failed to move listen fd to 3: %m");
+                                _exit(EXIT_FAILURE);
+                        }
+
+                        safe_close(m->listen_fd);
+                }
+
+                xsprintf(pids, PID_FMT, pid);
+                if (setenv("LISTEN_PID", pids, 1) < 0) {
+                        log_error_errno(errno, "Failed to set $LISTEN_PID: %m");
+                        _exit(EXIT_FAILURE);
+                }
+
+                if (setenv("LISTEN_FDS", "1", 1) < 0) {
+                        log_error_errno(errno, "Failed to set $LISTEN_FDS: %m");
+                        _exit(EXIT_FAILURE);
+                }
+
+
+                if (setenv("USERDB_FIXED_WORKER", one_zero(fixed), 1) < 0) {
+                        log_error_errno(errno, "Failed to set $USERDB_FIXED_WORKER: %m");
+                        _exit(EXIT_FAILURE);
+                }
+
+                /* execl("/home/lennart/projects/systemd/build/systemd-userwork", "systemd-userwork", "xxxxxxxxxxxxxxxx", NULL); /\* With some extra space rename_process() can make use of *\/ */
+                /* execl("/usr/bin/valgrind", "valgrind", "/home/lennart/projects/systemd/build/systemd-userwork", "systemd-userwork", "xxxxxxxxxxxxxxxx", NULL); /\* With some extra space rename_process() can make use of *\/ */
+
+                execl(SYSTEMD_USERWORK_PATH, "systemd-userwork", "xxxxxxxxxxxxxxxx", NULL); /* With some extra space rename_process() can make use of */
+                log_error_errno(errno, "Failed start worker process: %m");
+                _exit(EXIT_FAILURE);
+        }
+
+        if (fixed)
+                r = set_put(m->workers_fixed, PID_TO_PTR(pid));
+        else
+                r = set_put(m->workers_dynamic, PID_TO_PTR(pid));
+        if (r < 0)
+                return log_error_errno(r, "Failed to add child process to set: %m");
+
+        return 0;
+}
+
+static int start_workers(Manager *m, bool explicit_request) {
+        int r;
+
+        assert(m);
+
+        for (;;)  {
+                size_t n;
+
+                n = manager_current_workers(m);
+                if (n >= USERDB_WORKERS_MIN && (!explicit_request || n >= USERDB_WORKERS_MAX))
+                        break;
+
+                if (!ratelimit_below(&m->worker_ratelimit)) {
+                        /* If we keep starting workers too often, let's fail the whole daemon, something is wrong */
+                        sd_event_exit(m->event, EXIT_FAILURE);
+
+                        return log_error_errno(SYNTHETIC_ERRNO(EUCLEAN), "Worker threads requested too frequently, something is wrong.");
+                }
+
+                r = start_one_worker(m);
+                if (r < 0)
+                        return r;
+
+                explicit_request = false;
+        }
+
+        return 0;
+}
+
+int manager_startup(Manager *m) {
+        struct timeval ts;
+        int n, r;
+
+        assert(m);
+        assert(m->listen_fd < 0);
+
+        n = sd_listen_fds(false);
+        if (n < 0)
+                return log_error_errno(n, "Failed to determine number of passed file descriptors: %m");
+        if (n > 1)
+                return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Expected one listening fd, got %i.", n);
+        if (n == 1)
+                m->listen_fd = SD_LISTEN_FDS_START;
+        else {
+                union sockaddr_union sockaddr;
+
+                r = sockaddr_un_set_path(&sockaddr.un, "/run/systemd/userdb/io.systemd.NameServiceSwitch");
+                if (r < 0)
+                        return log_error_errno(r, "Cannot assign socket path to socket address: %m");
+
+                r = mkdir_p("/run/systemd/userdb", 0755);
+                if (r < 0)
+                        return log_error_errno(r, "Failed to create /run/systemd/userdb: %m");
+
+                m->listen_fd = socket(AF_UNIX, SOCK_STREAM|SOCK_CLOEXEC, 0);
+                if (m->listen_fd < 0)
+                        return log_error_errno(errno, "Failed to bind on socket: %m");
+
+                (void) sockaddr_un_unlink(&sockaddr.un);
+
+                RUN_WITH_UMASK(0000)
+                        if (bind(m->listen_fd, &sockaddr.sa, SOCKADDR_UN_LEN(sockaddr.un)) < 0)
+                                return log_error_errno(errno, "Failed to bind socket: %m");
+
+                r = symlink_idempotent("io.systemd.NameServiceSwitch", "/run/systemd/userdb/io.systemd.Multiplexer", false);
+                if (r < 0)
+                        return log_error_errno(r, "Failed to bind io.systemd.Multiplexer: %m");
+
+                if (listen(m->listen_fd, SOMAXCONN) < 0)
+                        return log_error_errno(errno, "Failed to listen on socket: %m");
+        }
+
+        /* Let's make sure every accept() call on this socket times out after 25s. This allows workers to be
+         * GC'ed on idle */
+        if (setsockopt(m->listen_fd, SOL_SOCKET, SO_RCVTIMEO, timeval_store(&ts, LISTEN_TIMEOUT_USEC), sizeof(ts)) < 0)
+                return log_error_errno(errno, "Failed to se SO_RCVTIMEO: %m");
+
+        return start_workers(m, false);
+}

src/userdb/userdbd-manager.h

@@ -0,0 +1,34 @@
+/* SPDX-License-Identifier: LGPL-2.1+ */
+#pragma once
+
+#include "sd-bus.h"
+#include "sd-event.h"
+
+typedef struct Manager Manager;
+
+#include "hashmap.h"
+#include "varlink.h"
+#include "ratelimit.h"
+
+#define USERDB_WORKERS_MIN 3
+#define USERDB_WORKERS_MAX 4096
+
+struct Manager {
+        sd_event *event;
+
+        Set *workers_fixed;    /* Workers 0…USERDB_WORKERS_MIN */
+        Set *workers_dynamic;  /* Workers USERD_WORKERS_MIN+1…USERDB_WORKERS_MAX */
+
+        sd_event_source *sigusr2_event_source;
+        sd_event_source *sigchld_event_source;
+
+        int listen_fd;
+
+        RateLimit worker_ratelimit;
+};
+
+int manager_new(Manager **ret);
+Manager* manager_free(Manager *m);
+DEFINE_TRIVIAL_CLEANUP_FUNC(Manager*, manager_free);
+
+int manager_startup(Manager *m);

src/userdb/userdbd.c

@@ -0,0 +1,56 @@
+/* SPDX-License-Identifier: LGPL-2.1+ */
+
+#include <sys/stat.h>
+#include <sys/types.h>
+
+#include "daemon-util.h"
+#include "userdbd-manager.h"
+#include "log.h"
+#include "main-func.h"
+#include "signal-util.h"
+
+/* This service offers two Varlink services, both implementing io.systemd.UserDatabase:
+ *
+ *         → io.systemd.NameServiceSwitch: this is a compatibility interface for glibc NSS: it response to
+ *           name lookups by checking the classic NSS interfaces and responding that.
+ *
+ *         → io.systemd.Multiplexer: this multiplexes lookup requests to all Varlink services that have a
+ *           socket in /run/systemd/userdb/. It's supposed to simplify clients that don't want to implement
+ *           the full iterative logic on their own.
+ */
+
+static int run(int argc, char *argv[]) {
+        _cleanup_(notify_on_cleanup) const char *notify_stop = NULL;
+        _cleanup_(manager_freep) Manager *m = NULL;
+        int r;
+
+        log_setup_service();
+
+        umask(0022);
+
+        if (argc != 1)
+                return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "This program takes no arguments.");
+
+        if (setenv("SYSTEMD_BYPASS_USERDB", "io.systemd.NameServiceSwitch:io.systemd.Multiplexer", 1) < 0)
+                return log_error_errno(errno, "Failed to se $SYSTEMD_BYPASS_USERDB: %m");
+
+        assert_se(sigprocmask_many(SIG_BLOCK, NULL, SIGCHLD, SIGTERM, SIGINT, SIGUSR2, -1) >= 0);
+
+        r = manager_new(&m);
+        if (r < 0)
+                return log_error_errno(r, "Could not create manager: %m");
+
+        r = manager_startup(m);
+        if (r < 0)
+                return log_error_errno(r, "Failed to start up daemon: %m");
+
+        notify_stop = notify_start(NOTIFY_READY, NOTIFY_STOPPING);
+
+        r = sd_event_loop(m->event);
+        if (r < 0)
+                return log_error_errno(r, "Event loop failed: %m");
+
+        return 0;
+}
+
+DEFINE_MAIN_FUNCTION(run);

units/meson.build

@@ -96,6 +96,8 @@ units = [
          'sockets.target.wants/'],
         ['systemd-journald.socket',             '',
          'sockets.target.wants/'],
+        ['systemd-userdbd.socket',              'ENABLE_USERDB',
+         'sockets.target.wants/'],
         ['systemd-networkd.socket',             'ENABLE_NETWORKD'],
         ['systemd-poweroff.service',             ''],
         ['systemd-reboot.service',               ''],
@@ -182,6 +184,7 @@ in_units = [
         ['systemd-nspawn@.service',              ''],
         ['systemd-portabled.service',            'ENABLE_PORTABLED',
          'dbus-org.freedesktop.portable1.service'],
+        ['systemd-userdbd.service',              'ENABLE_USERDB'],
         ['systemd-quotacheck.service',           'ENABLE_QUOTACHECK'],
         ['systemd-random-seed.service',          'ENABLE_RANDOMSEED',
          'sysinit.target.wants/'],

units/systemd-userdbd.service.in

@@ -0,0 +1,41 @@
+#  SPDX-License-Identifier: LGPL-2.1+
+#
+#  This file is part of systemd.
+#
+#  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.
+
+[Unit]
+Description=User Database Manager
+Documentation=man:systemd-userdbd.service(8)
+Requires=systemd-userdbd.socket
+After=systemd-userdbd.socket
+Before=sysinit.target
+DefaultDependencies=no
+
+[Service]
+CapabilityBoundingSet=CAP_DAC_READ_SEARCH
+ExecStart=@rootlibexecdir@/systemd-userdbd
+IPAddressDeny=any
+LimitNOFILE=@HIGH_RLIMIT_NOFILE@
+LockPersonality=yes
+MemoryDenyWriteExecute=yes
+NoNewPrivileges=yes
+PrivateDevices=yes
+ProtectControlGroups=yes
+ProtectHome=yes
+ProtectHostname=yes
+ProtectKernelLogs=yes
+ProtectKernelModules=yes
+ProtectSystem=strict
+RestrictAddressFamilies=AF_UNIX AF_NETLINK AF_INET AF_INET6
+RestrictNamespaces=yes
+RestrictRealtime=yes
+RestrictSUIDSGID=yes
+SystemCallArchitectures=native
+SystemCallErrorNumber=EPERM
+SystemCallFilter=@system-service
+Type=notify
+@SERVICE_WATCHDOG@

units/systemd-userdbd.socket

@@ -0,0 +1,19 @@
+#  SPDX-License-Identifier: LGPL-2.1+
+#
+#  This file is part of systemd.
+#
+#  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.
+
+[Unit]
+Description=User Database Manager Socket
+Documentation=man:systemd-userdbd.service(8)
+DefaultDependencies=no
+Before=sockets.target
+
+[Socket]
+ListenStream=/run/systemd/userdb/io.systemd.Multiplexer
+Symlinks=/run/systemd/userdb/io.systemd.NameServiceSwitch
+SocketMode=0666