libvirtd.c 96.0 KB
Newer Older
D
Daniel P. Berrange 已提交
1
/*
D
Daniel P. Berrange 已提交
2
 * libvirtd.c: daemon start of day, guest process & i/o management
D
Daniel P. Berrange 已提交
3
 *
4
 * Copyright (C) 2006-2010 Red Hat, Inc.
D
Daniel P. Berrange 已提交
5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23
 * Copyright (C) 2006 Daniel P. Berrange
 *
 * This library 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.
 *
 * This library is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * Lesser General Public License for more details.
 *
 * You should have received a copy of the GNU Lesser General Public
 * License along with this library; if not, write to the Free Software
 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307  USA
 *
 * Author: Daniel P. Berrange <berrange@redhat.com>
 */

24
#include <config.h>
25

D
Daniel P. Berrange 已提交
26 27 28 29 30 31 32 33 34 35
#include <sys/types.h>
#include <sys/wait.h>
#include <sys/stat.h>
#include <unistd.h>
#include <fcntl.h>
#include <limits.h>
#include <sys/socket.h>
#include <sys/un.h>
#include <sys/poll.h>
#include <netinet/in.h>
36
#include <netinet/tcp.h>
D
Daniel P. Berrange 已提交
37 38 39 40
#include <netdb.h>
#include <stdlib.h>
#include <pwd.h>
#include <stdio.h>
41 42
#include <stdarg.h>
#include <syslog.h>
D
Daniel P. Berrange 已提交
43 44 45
#include <string.h>
#include <errno.h>
#include <getopt.h>
46
#include <fnmatch.h>
47
#include <grp.h>
48
#include <signal.h>
J
Jim Meyering 已提交
49
#include <netdb.h>
E
Eric Blake 已提交
50
#include <locale.h>
51

52
#include "libvirt_internal.h"
53
#include "virterror_internal.h"
54
#include "files.h"
55

56 57
#define VIR_FROM_THIS VIR_FROM_QEMU

D
Daniel P. Berrange 已提交
58
#include "libvirtd.h"
59 60
#include "dispatch.h"

61
#include "util.h"
62
#include "uuid.h"
63
#include "remote_driver.h"
64
#include "conf.h"
D
Daniel P. Berrange 已提交
65
#include "event.h"
66
#include "memory.h"
67
#include "stream.h"
68
#include "hooks.h"
69
#include "virtaudit.h"
70
#ifdef HAVE_AVAHI
71
# include "mdns.h"
72
#endif
D
Daniel P. Berrange 已提交
73

74
#ifdef WITH_DRIVER_MODULES
75
# include "driver.h"
76
#else
77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103
# ifdef WITH_QEMU
#  include "qemu/qemu_driver.h"
# endif
# ifdef WITH_LXC
#  include "lxc/lxc_driver.h"
# endif
# ifdef WITH_UML
#  include "uml/uml_driver.h"
# endif
# ifdef WITH_ONE
#  include "opennebula/one_driver.h"
# endif
# ifdef WITH_NETWORK
#  include "network/bridge_driver.h"
# endif
# ifdef WITH_NETCF
#  include "interface/netcf_driver.h"
# endif
# ifdef WITH_STORAGE_DIR
#  include "storage/storage_driver.h"
# endif
# ifdef WITH_NODE_DEVICES
#  include "node_device/node_device_driver.h"
# endif
# ifdef WITH_SECRETS
#  include "secret/secret_driver.h"
# endif
104 105 106
# ifdef WITH_NWFILTER
#  include "nwfilter/nwfilter_driver.h"
# endif
107
#endif
108 109


J
John Levon 已提交
110
#ifdef __sun
111 112
# include <ucred.h>
# include <priv.h>
J
John Levon 已提交
113

114 115 116
# ifndef PRIV_VIRT_MANAGE
#  define PRIV_VIRT_MANAGE ((const char *)"virt_manage")
# endif
J
John Levon 已提交
117

118 119 120
# ifndef PRIV_XVM_CONTROL
#  define PRIV_XVM_CONTROL ((const char *)"xvm_control")
# endif
J
John Levon 已提交
121

122 123
# define PU_RESETGROUPS          0x0001  /* Remove supplemental groups */
# define PU_CLEARLIMITSET        0x0008  /* L=0 */
J
John Levon 已提交
124 125 126

extern int __init_daemon_priv(int, uid_t, gid_t, ...);

127
# define SYSTEM_UID 60
J
John Levon 已提交
128 129 130 131 132 133 134 135 136 137 138 139 140

static gid_t unix_sock_gid = 60; /* Not used */
static int unix_sock_rw_mask = 0666;
static int unix_sock_ro_mask = 0666;

#else

static gid_t unix_sock_gid = 0; /* Only root by default */
static int unix_sock_rw_mask = 0700; /* Allow user only */
static int unix_sock_ro_mask = 0777; /* Allow world */

#endif /* __sun */

141 142
#include "configmake.h"

143 144
static int godaemon = 0;        /* -d: Be a daemon */
static int verbose = 0;         /* -v: Verbose mode */
145
static int timeout = -1;        /* -t: Shutdown timeout */
146
static int sigwrite = -1;       /* Signal handler pipe */
147
static int ipsock = 0;          /* -l  Listen for TCP/IP */
148

149
/* Defaults for configuration file elements */
150 151
static int listen_tls = 1;
static int listen_tcp = 0;
152
static char *listen_addr  = (char *) LIBVIRTD_LISTEN_ADDR;
153 154
static char *tls_port = (char *) LIBVIRTD_TLS_PORT;
static char *tcp_port = (char *) LIBVIRTD_TCP_PORT;
155

156 157
static char *unix_sock_dir = NULL;

158 159 160 161
#if HAVE_POLKIT
static int auth_unix_rw = REMOTE_AUTH_POLKIT;
static int auth_unix_ro = REMOTE_AUTH_POLKIT;
#else
162 163
static int auth_unix_rw = REMOTE_AUTH_NONE;
static int auth_unix_ro = REMOTE_AUTH_NONE;
164
#endif /* HAVE_POLKIT */
165 166 167 168 169 170 171
#if HAVE_SASL
static int auth_tcp = REMOTE_AUTH_SASL;
#else
static int auth_tcp = REMOTE_AUTH_NONE;
#endif
static int auth_tls = REMOTE_AUTH_NONE;

172
static int mdns_adv = 1;
173
static char *mdns_name = NULL;
174

175
static int tls_no_verify_certificate = 0;
176
static char **tls_allowed_dn_list = NULL;
177

178 179 180 181
static char *key_file = (char *) LIBVIRT_SERVERKEY;
static char *cert_file = (char *) LIBVIRT_SERVERCERT;
static char *ca_file = (char *) LIBVIRT_CACERT;
static char *crl_file = (char *) "";
182 183 184 185

static gnutls_certificate_credentials_t x509_cred;
static gnutls_dh_params_t dh_params;

186 187 188 189
static int min_workers = 5;
static int max_workers = 20;
static int max_clients = 20;

190 191 192 193 194
/* Total number of 'in-process' RPC calls allowed across all clients */
static int max_requests = 20;
/* Total number of 'in-process' RPC calls allowed by a single client*/
static int max_client_requests = 5;

195 196 197
static int audit_level = 1;
static int audit_logging = 0;

198
#define DH_BITS 1024
199

200 201
static sig_atomic_t sig_errors = 0;
static int sig_lasterrno = 0;
202
static const char *argv0;
203

204 205 206 207 208 209 210 211 212
enum {
    VIR_DAEMON_ERR_NONE = 0,
    VIR_DAEMON_ERR_PIDFILE,
    VIR_DAEMON_ERR_RUNDIR,
    VIR_DAEMON_ERR_INIT,
    VIR_DAEMON_ERR_SIGNAL,
    VIR_DAEMON_ERR_PRIVS,
    VIR_DAEMON_ERR_NETWORK,
    VIR_DAEMON_ERR_CONFIG,
213
    VIR_DAEMON_ERR_HOOKS,
214
    VIR_DAEMON_ERR_AUDIT,
215 216 217 218 219 220 221 222 223 224 225 226 227

    VIR_DAEMON_ERR_LAST
};

VIR_ENUM_DECL(virDaemonErr)
VIR_ENUM_IMPL(virDaemonErr, VIR_DAEMON_ERR_LAST,
              "Initialization successful",
              "Unable to obtain pidfile",
              "Unable to create rundir",
              "Unable to initialize libvirt",
              "Unable to setup signal handlers",
              "Unable to drop privileges",
              "Unable to initialize network sockets",
228
              "Unable to load configuration file",
229 230
              "Unable to look for hook scripts",
              "Unable to initialize audit system")
231

232 233
static void sig_handler(int sig, siginfo_t * siginfo,
                        void* context ATTRIBUTE_UNUSED) {
234
    int origerrno;
235
    int r;
236

237 238
    /* set the sig num in the struct */
    siginfo->si_signo = sig;
239 240

    origerrno = errno;
241
    r = safewrite(sigwrite, siginfo, sizeof(*siginfo));
242 243 244 245
    if (r == -1) {
        sig_errors++;
        sig_lasterrno = errno;
    }
246
    errno = origerrno;
D
Daniel P. Berrange 已提交
247
}
248

249 250
static void qemudDispatchClientEvent(int watch, int fd, int events, void *opaque);
static void qemudDispatchServerEvent(int watch, int fd, int events, void *opaque);
251
static int qemudStartWorker(struct qemud_server *server, struct qemud_worker *worker);
252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267

void
qemudClientMessageQueuePush(struct qemud_client_message **queue,
                            struct qemud_client_message *msg)
{
    struct qemud_client_message *tmp = *queue;

    if (tmp) {
        while (tmp->next)
            tmp = tmp->next;
        tmp->next = msg;
    } else {
        *queue = msg;
    }
}

268
struct qemud_client_message *
269 270 271 272 273 274 275 276 277 278 279
qemudClientMessageQueueServe(struct qemud_client_message **queue)
{
    struct qemud_client_message *tmp = *queue;

    if (tmp) {
        *queue = tmp->next;
        tmp->next = NULL;
    }

    return tmp;
}
D
Daniel P. Berrange 已提交
280

281 282 283 284 285
static int
remoteCheckCertFile(const char *type, const char *file)
{
    struct stat sb;
    if (stat(file, &sb) < 0) {
286 287 288
        char ebuf[1024];
        VIR_ERROR(_("Cannot access %s '%s': %s"),
                  type, file, virStrerror(errno, ebuf, sizeof ebuf));
289 290 291 292 293
        return -1;
    }
    return 0;
}

294 295 296 297 298 299 300 301 302 303
static int
remoteInitializeGnuTLS (void)
{
    int err;

    /* Initialise GnuTLS. */
    gnutls_global_init ();

    err = gnutls_certificate_allocate_credentials (&x509_cred);
    if (err) {
304
        VIR_ERROR(_("gnutls_certificate_allocate_credentials: %s"),
305 306 307 308 309
                  gnutls_strerror (err));
        return -1;
    }

    if (ca_file && ca_file[0] != '\0') {
310 311 312
        if (remoteCheckCertFile("CA certificate", ca_file) < 0)
            return -1;

313 314 315 316
        qemudDebug ("loading CA cert from %s", ca_file);
        err = gnutls_certificate_set_x509_trust_file (x509_cred, ca_file,
                                                      GNUTLS_X509_FMT_PEM);
        if (err < 0) {
317
            VIR_ERROR(_("gnutls_certificate_set_x509_trust_file: %s"),
318 319 320 321 322 323
                      gnutls_strerror (err));
            return -1;
        }
    }

    if (crl_file && crl_file[0] != '\0') {
J
Jim Meyering 已提交
324
        if (remoteCheckCertFile("CA revocation list", crl_file) < 0)
325 326
            return -1;

327
        DEBUG("loading CRL from %s", crl_file);
328 329 330
        err = gnutls_certificate_set_x509_crl_file (x509_cred, crl_file,
                                                    GNUTLS_X509_FMT_PEM);
        if (err < 0) {
331
            VIR_ERROR(_("gnutls_certificate_set_x509_crl_file: %s"),
332 333 334 335 336 337
                      gnutls_strerror (err));
            return -1;
        }
    }

    if (cert_file && cert_file[0] != '\0' && key_file && key_file[0] != '\0') {
338 339 340 341
        if (remoteCheckCertFile("server certificate", cert_file) < 0)
            return -1;
        if (remoteCheckCertFile("server key", key_file) < 0)
            return -1;
342
        DEBUG("loading cert and key from %s and %s", cert_file, key_file);
343 344 345 346 347
        err =
            gnutls_certificate_set_x509_key_file (x509_cred,
                                                  cert_file, key_file,
                                                  GNUTLS_X509_FMT_PEM);
        if (err < 0) {
348
            VIR_ERROR(_("gnutls_certificate_set_x509_key_file: %s"),
349 350 351 352 353 354 355 356 357 358 359 360
                      gnutls_strerror (err));
            return -1;
        }
    }

    /* Generate Diffie Hellman parameters - for use with DHE
     * kx algorithms. These should be discarded and regenerated
     * once a day, once a week or once a month. Depending on the
     * security requirements.
     */
    err = gnutls_dh_params_init (&dh_params);
    if (err < 0) {
361
        VIR_ERROR(_("gnutls_dh_params_init: %s"), gnutls_strerror (err));
362 363 364 365
        return -1;
    }
    err = gnutls_dh_params_generate2 (dh_params, DH_BITS);
    if (err < 0) {
366
        VIR_ERROR(_("gnutls_dh_params_generate2: %s"), gnutls_strerror (err));
367 368 369 370 371 372 373 374
        return -1;
    }

    gnutls_certificate_set_dh_params (x509_cred, dh_params);

    return 0;
}

375
static void
376 377
qemudDispatchSignalEvent(int watch ATTRIBUTE_UNUSED,
                         int fd ATTRIBUTE_UNUSED,
378 379
                         int events ATTRIBUTE_UNUSED,
                         void *opaque) {
D
Daniel P. Berrange 已提交
380
    struct qemud_server *server = (struct qemud_server *)opaque;
381
    siginfo_t siginfo;
382

383
    virMutexLock(&server->lock);
384

385
    if (saferead(server->sigread, &siginfo, sizeof(siginfo)) != sizeof(siginfo)) {
386 387 388
        char ebuf[1024];
        VIR_ERROR(_("Failed to read from signal pipe: %s"),
                  virStrerror(errno, ebuf, sizeof ebuf));
389
        virMutexUnlock(&server->lock);
D
Daniel P. Berrange 已提交
390
        return;
391 392
    }

393
    switch (siginfo.si_signo) {
394
    case SIGHUP:
395
        VIR_INFO0(_("Reloading configuration on SIGHUP"));
396 397
        virHookCall(VIR_HOOK_DRIVER_DAEMON, "-",
                    VIR_HOOK_DAEMON_OP_RELOAD, SIGHUP, "SIGHUP", NULL);
398
        if (virStateReload() < 0)
399
            VIR_WARN0("Error while reloading drivers");
400

401 402 403
        break;

    case SIGINT:
404
    case SIGQUIT:
405
    case SIGTERM:
406
        VIR_WARN("Shutting down on signal %d", siginfo.si_signo);
407
        server->quitEventThread = 1;
408 409 410
        break;

    default:
411
        VIR_INFO(_("Received unexpected signal %d"), siginfo.si_signo);
412 413 414
        break;
    }

415
    virMutexUnlock(&server->lock);
416 417
}

418

419 420 421 422 423
static int daemonForkIntoBackground(void) {
    int statuspipe[2];
    if (pipe(statuspipe) < 0)
        return -1;

D
Daniel P. Berrange 已提交
424 425 426 427 428 429
    int pid = fork();
    switch (pid) {
    case 0:
        {
            int stdinfd = -1;
            int stdoutfd = -1;
430
            int nextpid;
D
Daniel P. Berrange 已提交
431

432
            VIR_FORCE_CLOSE(statuspipe[0]);
433

434
            if ((stdinfd = open("/dev/null", O_RDONLY)) < 0)
D
Daniel P. Berrange 已提交
435
                goto cleanup;
436
            if ((stdoutfd = open("/dev/null", O_WRONLY)) < 0)
D
Daniel P. Berrange 已提交
437 438 439 440 441 442 443
                goto cleanup;
            if (dup2(stdinfd, STDIN_FILENO) != STDIN_FILENO)
                goto cleanup;
            if (dup2(stdoutfd, STDOUT_FILENO) != STDOUT_FILENO)
                goto cleanup;
            if (dup2(stdoutfd, STDERR_FILENO) != STDERR_FILENO)
                goto cleanup;
444
            if (VIR_CLOSE(stdinfd) < 0)
D
Daniel P. Berrange 已提交
445
                goto cleanup;
446
            if (VIR_CLOSE(stdoutfd) < 0)
D
Daniel P. Berrange 已提交
447 448 449 450 451 452 453 454
                goto cleanup;

            if (setsid() < 0)
                goto cleanup;

            nextpid = fork();
            switch (nextpid) {
            case 0:
455
                return statuspipe[1];
D
Daniel P. Berrange 已提交
456 457 458
            case -1:
                return -1;
            default:
459
                _exit(0);
D
Daniel P. Berrange 已提交
460 461 462
            }

        cleanup:
463 464
            VIR_FORCE_CLOSE(stdoutfd);
            VIR_FORCE_CLOSE(stdinfd);
D
Daniel P. Berrange 已提交
465 466 467 468 469 470 471 472 473
            return -1;

        }

    case -1:
        return -1;

    default:
        {
474 475 476 477
            int got, exitstatus = 0;
            int ret;
            char status;

478
            VIR_FORCE_CLOSE(statuspipe[1]);
479 480 481

            /* We wait to make sure the first child forked successfully */
            if ((got = waitpid(pid, &exitstatus, 0)) < 0 ||
D
Daniel P. Berrange 已提交
482
                got != pid ||
483
                exitstatus != 0) {
D
Daniel P. Berrange 已提交
484 485
                return -1;
            }
486 487 488 489 490 491 492 493

            /* Now block until the second child initializes successfully */
        again:
            ret = read(statuspipe[0], &status, 1);
            if (ret == -1 && errno == EINTR)
                goto again;

            if (ret == 1 && status != 0) {
494
                fprintf(stderr,
495 496
                        _("%s: error: %s. Check /var/log/messages or run without "
                          "--daemon for more info.\n"), argv0,
497
                        virDaemonErrTypeToString(status));
498 499
            }
            _exit(ret == 1 && status == 0 ? 0 : 1);
D
Daniel P. Berrange 已提交
500 501 502 503
        }
    }
}

504 505 506
static int qemudWritePidFile(const char *pidFile) {
    int fd;
    FILE *fh;
507
    char ebuf[1024];
508 509 510 511 512

    if (pidFile[0] == '\0')
        return 0;

    if ((fd = open(pidFile, O_WRONLY|O_CREAT|O_EXCL, 0644)) < 0) {
513
        VIR_ERROR(_("Failed to open pid file '%s' : %s"),
514
                  pidFile, virStrerror(errno, ebuf, sizeof ebuf));
515 516 517
        return -1;
    }

518
    if (!(fh = VIR_FDOPEN(fd, "w"))) {
519
        VIR_ERROR(_("Failed to fdopen pid file '%s' : %s"),
520
                  pidFile, virStrerror(errno, ebuf, sizeof ebuf));
521
        VIR_FORCE_CLOSE(fd);
522 523 524 525
        return -1;
    }

    if (fprintf(fh, "%lu\n", (unsigned long)getpid()) < 0) {
526 527
        VIR_ERROR(_("%s: Failed to write to pid file '%s' : %s"),
                  argv0, pidFile, virStrerror(errno, ebuf, sizeof ebuf));
528
        VIR_FORCE_FCLOSE(fh);
529 530 531
        return -1;
    }

532
    if (VIR_FCLOSE(fh) == EOF) {
533 534
        VIR_ERROR(_("%s: Failed to close pid file '%s' : %s"),
                  argv0, pidFile, virStrerror(errno, ebuf, sizeof ebuf));
535 536 537 538 539 540
        return -1;
    }

    return 0;
}

D
Daniel P. Berrange 已提交
541
static int qemudListenUnix(struct qemud_server *server,
542
                           char *path, int readonly, int auth) {
543
    struct qemud_socket *sock;
D
Daniel P. Berrange 已提交
544
    mode_t oldmask;
545
    gid_t oldgrp;
546
    char ebuf[1024];
D
Daniel P. Berrange 已提交
547

548
    if (VIR_ALLOC(sock) < 0) {
549
        VIR_ERROR0(_("Failed to allocate memory for struct qemud_socket"));
D
Daniel P. Berrange 已提交
550
        return -1;
551
    }
D
Daniel P. Berrange 已提交
552 553

    sock->readonly = readonly;
554
    sock->type = QEMUD_SOCK_TYPE_UNIX;
555
    sock->auth = auth;
556
    sock->path = path;
557 558 559 560 561 562
    sock->addr.len = sizeof(sock->addr.data.un);
    if (!(sock->addrstr = strdup(path))) {
        VIR_ERROR(_("Failed to copy socket address: %s"),
                  virStrerror(errno, ebuf, sizeof ebuf));
        goto cleanup;
    }
D
Daniel P. Berrange 已提交
563

564
    if ((sock->fd = socket(PF_UNIX, SOCK_STREAM, 0)) < 0) {
565
        VIR_ERROR(_("Failed to create socket: %s"),
566
                  virStrerror(errno, ebuf, sizeof ebuf));
D
Daniel P. Berrange 已提交
567
        goto cleanup;
568
    }
D
Daniel P. Berrange 已提交
569

570 571
    if (virSetCloseExec(sock->fd) < 0 ||
        virSetNonBlock(sock->fd) < 0)
D
Daniel P. Berrange 已提交
572
        goto cleanup;
D
Daniel P. Berrange 已提交
573

574 575
    sock->addr.data.un.sun_family = AF_UNIX;
    if (virStrcpyStatic(sock->addr.data.un.sun_path, path) == NULL) {
C
Chris Lalancette 已提交
576 577 578
        VIR_ERROR(_("Path %s too long for unix socket"), path);
        goto cleanup;
    }
579 580
    if (sock->addr.data.un.sun_path[0] == '@')
        sock->addr.data.un.sun_path[0] = '\0';
D
Daniel P. Berrange 已提交
581

582
    oldgrp = getgid();
583
    oldmask = umask(readonly ? ~unix_sock_ro_mask : ~unix_sock_rw_mask);
584 585 586 587
    if (server->privileged && setgid(unix_sock_gid)) {
        VIR_ERROR(_("Failed to set group ID to %d"), unix_sock_gid);
        goto cleanup;
    }
588

589
    if (bind(sock->fd, &sock->addr.data.sa, sock->addr.len) < 0) {
590
        VIR_ERROR(_("Failed to bind socket to '%s': %s"),
591
                  path, virStrerror(errno, ebuf, sizeof ebuf));
D
Daniel P. Berrange 已提交
592
        goto cleanup;
593
    }
D
Daniel P. Berrange 已提交
594
    umask(oldmask);
595 596 597 598
    if (server->privileged && setgid(oldgrp)) {
        VIR_ERROR(_("Failed to restore group ID to %d"), oldgrp);
        goto cleanup;
    }
D
Daniel P. Berrange 已提交
599

600
    if (listen(sock->fd, 30) < 0) {
601
        VIR_ERROR(_("Failed to listen for connections on '%s': %s"),
602
                  path, virStrerror(errno, ebuf, sizeof ebuf));
D
Daniel P. Berrange 已提交
603 604 605 606 607 608 609
        goto cleanup;
    }

    sock->next = server->sockets;
    server->sockets = sock;
    server->nsockets++;

D
Daniel P. Berrange 已提交
610
    return 0;
D
Daniel P. Berrange 已提交
611 612

 cleanup:
613
    VIR_FORCE_CLOSE(sock->fd);
614
    VIR_FREE(sock);
D
Daniel P. Berrange 已提交
615
    return -1;
D
Daniel P. Berrange 已提交
616 617
}

618 619
// See: http://people.redhat.com/drepper/userapi-ipv6.html
static int
620
remoteMakeSockets (int *fds, int max_fds, int *nfds_r, const char *node, const char *service)
621 622 623 624 625 626 627
{
    struct addrinfo *ai;
    struct addrinfo hints;
    memset (&hints, 0, sizeof hints);
    hints.ai_flags = AI_PASSIVE | AI_ADDRCONFIG;
    hints.ai_socktype = SOCK_STREAM;

628
    int e = getaddrinfo (node, service, &hints, &ai);
629
    if (e != 0) {
630
        VIR_ERROR(_("getaddrinfo: %s"), gai_strerror (e));
631 632 633 634 635
        return -1;
    }

    struct addrinfo *runp = ai;
    while (runp && *nfds_r < max_fds) {
636
        char ebuf[1024];
637 638 639
        fds[*nfds_r] = socket (runp->ai_family, runp->ai_socktype,
                               runp->ai_protocol);
        if (fds[*nfds_r] == -1) {
640
            VIR_ERROR(_("socket: %s"), virStrerror (errno, ebuf, sizeof ebuf));
641 642 643 644 645 646
            return -1;
        }

        int opt = 1;
        setsockopt (fds[*nfds_r], SOL_SOCKET, SO_REUSEADDR, &opt, sizeof opt);

647 648 649 650 651 652 653 654 655 656 657 658 659 660 661
#ifdef IPV6_V6ONLY
        if (runp->ai_family == PF_INET6) {
            int on = 1;
            /*
             * Normally on Linux an INET6 socket will bind to the INET4
             * address too. If getaddrinfo returns results with INET4
             * first though, this will result in INET6 binding failing.
             * We can trivially cope with multiple server sockets, so
             * we force it to only listen on IPv6
             */
            setsockopt(fds[*nfds_r], IPPROTO_IPV6,IPV6_V6ONLY,
                       (void*)&on, sizeof on);
        }
#endif

662 663
        if (bind (fds[*nfds_r], runp->ai_addr, runp->ai_addrlen) == -1) {
            if (errno != EADDRINUSE) {
664
                VIR_ERROR(_("bind: %s"), virStrerror (errno, ebuf, sizeof ebuf));
665 666
                return -1;
            }
667
            VIR_FORCE_CLOSE(fds[*nfds_r]);
668
        } else {
669 670 671 672
            ++*nfds_r;
        }
        runp = runp->ai_next;
    }
673

674 675 676 677 678 679 680 681 682
    freeaddrinfo (ai);
    return 0;
}

/* Listen on the named/numbered TCP port.  On a machine with IPv4 and
 * IPv6 interfaces this may generate several sockets.
 */
static int
remoteListenTCP (struct qemud_server *server,
683
                 const char *addr,
684
                 const char *port,
685
                 int type,
686
                 int auth)
687 688 689
{
    int fds[2];
    int nfds = 0;
690
    int i;
691 692
    struct qemud_socket *sock;

693
    if (remoteMakeSockets (fds, 2, &nfds, addr, port) == -1)
694 695 696
        return -1;

    for (i = 0; i < nfds; ++i) {
697
        char ebuf[1024];
698

699
        if (VIR_ALLOC(sock) < 0) {
700 701
            VIR_ERROR(_("remoteListenTCP: calloc: %s"),
                      virStrerror (errno, ebuf, sizeof ebuf));
702
            goto cleanup;
703 704
        }

705
        sock->addr.len = sizeof(sock->addr.data.stor);
706 707 708 709 710 711
        sock->readonly = 0;
        sock->next = server->sockets;
        server->sockets = sock;
        server->nsockets++;

        sock->fd = fds[i];
712
        sock->type = type;
713
        sock->auth = auth;
D
Daniel P. Berrange 已提交
714

715
        if (getsockname(sock->fd, &sock->addr.data.sa, &sock->addr.len) < 0)
716
            goto cleanup;
717

718 719
        if (!(sock->addrstr = virSocketFormatAddrFull(&sock->addr, true, ";")))
            goto cleanup;
720

721 722
        if (virSetCloseExec(sock->fd) < 0 ||
            virSetNonBlock(sock->fd) < 0)
723
            goto cleanup;
724 725

        if (listen (sock->fd, 30) < 0) {
726 727
            VIR_ERROR(_("remoteListenTCP: listen: %s"),
                      virStrerror (errno, ebuf, sizeof ebuf));
728
            goto cleanup;
729
        }
730 731 732
    }

    return 0;
733 734 735

cleanup:
    for (i = 0; i < nfds; ++i)
736
        VIR_FORCE_CLOSE(fds[i]);
737
    return -1;
738 739 740
}

static int qemudInitPaths(struct qemud_server *server,
741 742
                          char **sockname,
                          char **roSockname)
J
John Levon 已提交
743
{
744
    char *base_dir_prefix = NULL;
745
    char *sock_dir_prefix = NULL;
746 747 748 749 750
    int ret = -1;

    /* The base_dir_prefix is the base under which all libvirtd
     * files live */
    if (server->privileged) {
751
        if (!(base_dir_prefix = strdup (LOCALSTATEDIR)))
752 753 754 755 756 757
            goto no_memory;
    } else {
        uid_t uid = geteuid();
        if (!(base_dir_prefix = virGetUserDirectory(uid)))
            goto cleanup;
    }
758

759 760
    /* The unix_sock_dir is the location under which all
     * unix domain sockets live */
D
Daniel P. Berrange 已提交
761
    if (unix_sock_dir) {
762 763 764
        if (!(sock_dir_prefix = strdup(unix_sock_dir)))
            goto no_memory;

D
Daniel P. Berrange 已提交
765 766 767 768 769 770 771
        /* Change the group ownership of /var/run/libvirt to unix_sock_gid */
        if (server->privileged) {
            if (chown(unix_sock_dir, -1, unix_sock_gid) < 0)
                VIR_ERROR(_("Failed to change group ownership of %s"),
                          unix_sock_dir);
        }
    } else {
772
        if (server->privileged) {
773 774 775
            if (virAsprintf(&sock_dir_prefix, "%s/run/libvirt",
                            base_dir_prefix) < 0)
                goto no_memory;
776
        } else {
777 778 779
            if (virAsprintf(&sock_dir_prefix, "%s/.libvirt",
                            base_dir_prefix) < 0)
                goto no_memory;
780 781
        }
    }
782

783
    if (server->privileged) {
784 785 786 787 788 789 790 791
        if (virAsprintf(sockname, "%s/libvirt-sock",
                        sock_dir_prefix) < 0)
            goto no_memory;
        if (virAsprintf(roSockname, "%s/libvirt-sock-ro",
                        sock_dir_prefix) < 0)
            goto no_memory;
        unlink(*sockname);
        unlink(*roSockname);
D
Daniel P. Berrange 已提交
792
    } else {
793 794 795 796 797 798
        if (virAsprintf(sockname, "@%s/libvirt-sock",
                        sock_dir_prefix) < 0)
            goto no_memory;
        /* There is no RO socket in unprivileged mode,
         * since the user always has full RW access
         * to their private instance */
799
    }
800

801
    if (server->privileged) {
802 803 804
        if (virAsprintf(&server->logDir, "%s/log/libvirt",
                        base_dir_prefix) < 0)
            goto no_memory;
805
    } else {
806 807 808
        if (virAsprintf(&server->logDir, "%s/.libvirt/log",
                        base_dir_prefix) < 0)
            goto no_memory;
809
    }
810

811
    ret = 0;
812

813 814 815
no_memory:
    if (ret != 0)
        virReportOOMError();
816 817

 cleanup:
818
    VIR_FREE(base_dir_prefix);
819
    VIR_FREE(sock_dir_prefix);
820
    return ret;
D
Daniel P. Berrange 已提交
821 822
}

823 824 825 826 827 828
static void virshErrorHandler(void *opaque ATTRIBUTE_UNUSED, virErrorPtr err ATTRIBUTE_UNUSED)
{
    /* Don't do anything, since logging infrastructure already
     * took care of reporting the error */
}

829
static struct qemud_server *qemudInitialize(void) {
D
Daniel P. Berrange 已提交
830 831
    struct qemud_server *server;

832
    if (VIR_ALLOC(server) < 0) {
833
        VIR_ERROR0(_("Failed to allocate struct qemud_server"));
D
Daniel P. Berrange 已提交
834
        return NULL;
835
    }
D
Daniel P. Berrange 已提交
836

837 838 839
    server->privileged = geteuid() == 0 ? 1 : 0;
    server->sigread = server->sigwrite = -1;

840
    if (virMutexInit(&server->lock) < 0) {
841
        VIR_ERROR0(_("cannot initialize mutex"));
842
        VIR_FREE(server);
843
        return NULL;
844 845
    }
    if (virCondInit(&server->job) < 0) {
846
        VIR_ERROR0(_("cannot initialize condition variable"));
847
        virMutexDestroy(&server->lock);
848
        VIR_FREE(server);
849
        return NULL;
850 851
    }

852
    if (virEventInit() < 0) {
853
        VIR_ERROR0(_("Failed to initialize event system"));
854 855 856
        virMutexDestroy(&server->lock);
        if (virCondDestroy(&server->job) < 0)
        {}
857 858 859 860
        VIR_FREE(server);
        return NULL;
    }

861 862 863 864 865 866 867
    /*
     * Note that the order is important: the first ones have a higher
     * priority when calling virStateInitialize. We must register
     * the network, storage and nodedev drivers before any domain
     * drivers, since their resources must be auto-started before
     * any domains can be auto-started.
     */
868 869 870
#ifdef WITH_DRIVER_MODULES
    /* We don't care if any of these fail, because the whole point
     * is to allow users to only install modules they want to use.
D
Dan Kenigsberg 已提交
871
     * If they try to open a connection for a module that
872 873 874 875
     * is not loaded they'll get a suitable error at that point
     */
    virDriverLoadModule("network");
    virDriverLoadModule("storage");
876
    virDriverLoadModule("nodedev");
877
    virDriverLoadModule("secret");
878 879 880
    virDriverLoadModule("qemu");
    virDriverLoadModule("lxc");
    virDriverLoadModule("uml");
D
Daniel Veillard 已提交
881
    virDriverLoadModule("one");
882
    virDriverLoadModule("nwfilter");
883
#else
884
# ifdef WITH_NETWORK
885
    networkRegister();
886 887
# endif
# ifdef WITH_NETCF
888
    interfaceRegister();
889 890
# endif
# ifdef WITH_STORAGE_DIR
891
    storageRegister();
892 893
# endif
# if defined(WITH_NODE_DEVICES)
894
    nodedevRegister();
895 896
# endif
# ifdef WITH_SECRETS
897
    secretRegister();
898
# endif
899 900 901
# ifdef WITH_NWFILTER
    nwfilterRegister();
# endif
902
# ifdef WITH_QEMU
903
    qemuRegister();
904 905
# endif
# ifdef WITH_LXC
906
    lxcRegister();
907 908
# endif
# ifdef WITH_UML
909
    umlRegister();
910 911
# endif
# ifdef WITH_ONE
D
Daniel Veillard 已提交
912
    oneRegister();
913
# endif
914 915
#endif

916 917 918 919 920 921
    virEventRegisterImpl(virEventAddHandleImpl,
                         virEventUpdateHandleImpl,
                         virEventRemoveHandleImpl,
                         virEventAddTimeoutImpl,
                         virEventUpdateTimeoutImpl,
                         virEventRemoveTimeoutImpl);
922

923 924 925
    return server;
}

926
static int qemudNetworkInit(struct qemud_server *server) {
927 928
    char *sockname = NULL;
    char *roSockname = NULL;
929
#if HAVE_SASL
930 931 932
    int err;
#endif /* HAVE_SASL */

933
    if (qemudInitPaths(server, &sockname, &roSockname) < 0)
934 935 936
        goto cleanup;

    if (qemudListenUnix(server, sockname, 0, auth_unix_rw) < 0)
937
        goto cleanup;
938
    sockname = NULL;
939

940
    if (roSockname != NULL && qemudListenUnix(server, roSockname, 1, auth_unix_ro) < 0)
941
        goto cleanup;
942
    roSockname = NULL;
943 944 945 946 947 948 949

#if HAVE_SASL
    if (auth_unix_rw == REMOTE_AUTH_SASL ||
        auth_unix_ro == REMOTE_AUTH_SASL ||
        auth_tcp == REMOTE_AUTH_SASL ||
        auth_tls == REMOTE_AUTH_SASL) {
        if ((err = sasl_server_init(NULL, "libvirt")) != SASL_OK) {
950 951
            VIR_ERROR(_("Failed to initialize SASL authentication %s"),
                      sasl_errstring(err, NULL, NULL));
952 953
            goto cleanup;
        }
954 955 956
    }
#endif

957
#if HAVE_POLKIT0
958 959 960
    if (auth_unix_rw == REMOTE_AUTH_POLKIT ||
        auth_unix_ro == REMOTE_AUTH_POLKIT) {
        DBusError derr;
961 962 963 964

        dbus_connection_set_change_sigpipe(FALSE);
        dbus_threads_init_default();

965 966 967
        dbus_error_init(&derr);
        server->sysbus = dbus_bus_get(DBUS_BUS_SYSTEM, &derr);
        if (!(server->sysbus)) {
968 969
            VIR_ERROR(_("Failed to connect to system bus for PolicyKit auth: %s"),
                      derr.message);
970 971 972
            dbus_error_free(&derr);
            goto cleanup;
        }
973
        dbus_connection_set_exit_on_disconnect(server->sysbus, FALSE);
974 975 976
    }
#endif

977
    if (ipsock) {
978
        if (listen_tcp && remoteListenTCP (server, listen_addr, tcp_port, QEMUD_SOCK_TYPE_TCP, auth_tcp) < 0)
979 980 981 982 983 984
            goto cleanup;

        if (listen_tls) {
            if (remoteInitializeGnuTLS () < 0)
                goto cleanup;

985
            if (remoteListenTCP (server, listen_addr, tls_port, QEMUD_SOCK_TYPE_TLS, auth_tls) < 0)
986 987
                goto cleanup;
        }
D
Daniel P. Berrange 已提交
988 989
    }

990
#ifdef HAVE_AVAHI
991
    if (server->privileged && mdns_adv) {
992
        struct libvirtd_mdns_group *group;
993
        struct qemud_socket *sock;
994
        int port = 0;
995
        int ret;
996 997 998 999

        server->mdns = libvirtd_mdns_new();

        if (!mdns_name) {
1000 1001
            char *groupname, *localhost, *tmp;

1002 1003
            localhost = virGetHostname(NULL);
            if (localhost == NULL)
1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018
                /* we couldn't resolve the hostname; assume that we are
                 * running in disconnected operation, and report a less
                 * useful Avahi string
                 */
                ret = virAsprintf(&groupname, "Virtualization Host");
            else {
                /* Extract the host part of the potentially FQDN */
                if ((tmp = strchr(localhost, '.')))
                    *tmp = '\0';
                ret = virAsprintf(&groupname, "Virtualization Host %s",
                                  localhost);
            }
            VIR_FREE(localhost);
            if (ret < 0) {
                virReportOOMError();
1019
                goto cleanup;
1020
            }
1021
            group = libvirtd_mdns_add_group(server->mdns, groupname);
1022
            VIR_FREE(groupname);
1023 1024 1025 1026
        } else {
            group = libvirtd_mdns_add_group(server->mdns, mdns_name);
        }

1027
        /*
1028 1029 1030 1031 1032 1033
         * See if there's a TLS enabled port we can advertise. Cowardly
         * don't bother to advertise TCP since we don't want people using
         * them for real world apps
         */
        sock = server->sockets;
        while (sock) {
1034 1035 1036
            if (virSocketGetPort(&sock->addr) != -1 &&
                sock->type == QEMUD_SOCK_TYPE_TLS) {
                port = virSocketGetPort(&sock->addr);
1037 1038 1039 1040
                break;
            }
            sock = sock->next;
        }
1041

1042 1043 1044 1045 1046 1047 1048 1049 1050
        /*
         * Add the primary entry - we choose SSH because its most likely to always
         * be available
         */
        libvirtd_mdns_add_entry(group, "_libvirt._tcp", port);
        libvirtd_mdns_start(server->mdns);
    }
#endif

1051
    return 0;
D
Daniel P. Berrange 已提交
1052 1053

 cleanup:
1054 1055
    VIR_FREE(sockname);
    VIR_FREE(roSockname);
1056
    return -1;
D
Daniel P. Berrange 已提交
1057 1058
}

1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077
static int qemudNetworkEnable(struct qemud_server *server) {
    struct qemud_socket *sock;

    sock = server->sockets;
    while (sock) {
        if ((sock->watch = virEventAddHandleImpl(sock->fd,
                                                 VIR_EVENT_HANDLE_READABLE |
                                                 VIR_EVENT_HANDLE_ERROR |
                                                 VIR_EVENT_HANDLE_HANGUP,
                                                 qemudDispatchServerEvent,
                                                 server, NULL)) < 0) {
            VIR_ERROR0(_("Failed to add server event callback"));
            return -1;
        }

        sock = sock->next;
    }
    return 0;
}
1078

1079

1080 1081 1082
static gnutls_session_t
remoteInitializeTLSSession (void)
{
1083 1084
    gnutls_session_t session;
    int err;
1085

1086 1087
    err = gnutls_init (&session, GNUTLS_SERVER);
    if (err != 0) goto failed;
1088

1089 1090 1091 1092 1093
    /* avoid calling all the priority functions, since the defaults
     * are adequate.
     */
    err = gnutls_set_default_priority (session);
    if (err != 0) goto failed;
1094

1095 1096
    err = gnutls_credentials_set (session, GNUTLS_CRD_CERTIFICATE, x509_cred);
    if (err != 0) goto failed;
1097

1098 1099 1100
    /* request client certificate if any.
     */
    gnutls_certificate_server_set_request (session, GNUTLS_CERT_REQUEST);
1101

1102
    gnutls_dh_set_prime_bits (session, DH_BITS);
1103

1104
    return session;
1105

1106 1107 1108 1109
failed:
    VIR_ERROR(_("remoteInitializeTLSSession: %s"),
              gnutls_strerror (err));
    return NULL;
1110 1111 1112 1113
}

/* Check DN is on tls_allowed_dn_list. */
static int
1114
remoteCheckDN (const char *dname)
1115
{
1116
    char **wildcards;
1117 1118 1119 1120 1121 1122 1123

    /* If the list is not set, allow any DN. */
    wildcards = tls_allowed_dn_list;
    if (!wildcards)
        return 1;

    while (*wildcards) {
1124
        if (fnmatch (*wildcards, dname, 0) == 0)
1125 1126 1127 1128 1129
            return 1;
        wildcards++;
    }

    /* Print the client's DN. */
1130
    DEBUG(_("remoteCheckDN: failed: client DN is %s"), dname);
1131 1132 1133 1134 1135

    return 0; // Not found.
}

static int
1136
remoteCheckCertificate(struct qemud_client *client)
1137 1138 1139 1140 1141 1142
{
    int ret;
    unsigned int status;
    const gnutls_datum_t *certs;
    unsigned int nCerts, i;
    time_t now;
1143 1144 1145 1146
    char name[256];
    size_t namesize = sizeof name;

    memset(name, 0, namesize);
1147

1148 1149
    if ((ret = gnutls_certificate_verify_peers2 (client->tlssession, &status)) < 0){
        VIR_ERROR(_("Failed to verify certificate peers: %s"),
1150
                  gnutls_strerror (ret));
1151
        goto authdeny;
1152 1153 1154 1155
    }

    if (status != 0) {
        if (status & GNUTLS_CERT_INVALID)
1156
            VIR_ERROR0(_("The client certificate is not trusted."));
1157 1158

        if (status & GNUTLS_CERT_SIGNER_NOT_FOUND)
1159
            VIR_ERROR0(_("The client certificate has unknown issuer."));
1160 1161

        if (status & GNUTLS_CERT_REVOKED)
1162
            VIR_ERROR0(_("The client certificate has been revoked."));
1163

1164
#ifndef GNUTLS_1_0_COMPAT
1165
        if (status & GNUTLS_CERT_INSECURE_ALGORITHM)
1166
            VIR_ERROR0(_("The client certificate uses an insecure algorithm."));
1167
#endif
1168

1169
        goto authdeny;
1170 1171
    }

1172 1173 1174
    if (gnutls_certificate_type_get(client->tlssession) != GNUTLS_CRT_X509) {
        VIR_ERROR0(_("Only x509 certificates are supported"));
        goto authdeny;
1175 1176
    }

1177 1178 1179
    if (!(certs = gnutls_certificate_get_peers(client->tlssession, &nCerts))) {
        VIR_ERROR0(_("The certificate has no peers"));
        goto authdeny;
1180 1181 1182 1183 1184 1185 1186 1187
    }

    now = time (NULL);

    for (i = 0; i < nCerts; i++) {
        gnutls_x509_crt_t cert;

        if (gnutls_x509_crt_init (&cert) < 0) {
1188 1189
            VIR_ERROR0(_("Unable to initialize certificate"));
            goto authfail;
1190 1191 1192
        }

        if (gnutls_x509_crt_import(cert, &certs[i], GNUTLS_X509_FMT_DER) < 0) {
1193
            VIR_ERROR0(_("Unable to load certificate"));
1194
            gnutls_x509_crt_deinit (cert);
1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216
            goto authfail;
        }

        if (i == 0) {
            ret = gnutls_x509_crt_get_dn (cert, name, &namesize);
            if (ret != 0) {
                VIR_ERROR(_("Failed to get certificate distinguished name: %s"),
                          gnutls_strerror(ret));
                gnutls_x509_crt_deinit (cert);
                goto authfail;
            }

            if (!remoteCheckDN (name)) {
                /* This is the most common error: make it informative. */
                VIR_ERROR0(_("Client's Distinguished Name is not on the list "
                             "of allowed clients (tls_allowed_dn_list).  Use "
                             "'certtool -i --infile clientcert.pem' to view the"
                             "Distinguished Name field in the client certificate,"
                             "or run this daemon with --verbose option."));
                gnutls_x509_crt_deinit (cert);
                goto authdeny;
            }
1217
        }
1218

1219
        if (gnutls_x509_crt_get_expiration_time (cert) < now) {
1220
            VIR_ERROR0(_("The client certificate has expired"));
1221
            gnutls_x509_crt_deinit (cert);
1222
            goto authdeny;
1223
        }
1224

1225
        if (gnutls_x509_crt_get_activation_time (cert) > now) {
1226
            VIR_ERROR0(_("The client certificate is not yet active"));
1227
            gnutls_x509_crt_deinit (cert);
1228
            goto authdeny;
1229 1230 1231
        }
    }

M
Matthias Bolte 已提交
1232
    PROBE(CLIENT_TLS_ALLOW, "fd=%d, name=%s", client->fd, (char *)name);
1233
    return 0;
1234 1235

authdeny:
M
Matthias Bolte 已提交
1236
    PROBE(CLIENT_TLS_DENY, "fd=%d, name=%s", client->fd, (char *)name);
1237 1238 1239
    return -1;

authfail:
1240
    PROBE(CLIENT_TLS_FAIL, "fd=%d", client->fd);
1241
    return -1;
1242 1243 1244 1245 1246 1247
}

/* Check the client's access. */
static int
remoteCheckAccess (struct qemud_client *client)
{
1248 1249
    struct qemud_client_message *confirm;

1250
    /* Verify client certificate. */
1251
    if (remoteCheckCertificate (client) == -1) {
1252 1253
        VIR_ERROR0(_("remoteCheckCertificate: "
                     "failed to verify client's certificate"));
1254
        if (!tls_no_verify_certificate) return -1;
1255 1256
        else VIR_INFO0(_("remoteCheckCertificate: tls_no_verify_certificate "
                          "is set so the bad certificate is ignored"));
1257 1258
    }

1259 1260 1261 1262 1263 1264 1265 1266 1267
    if (client->tx) {
        VIR_INFO("%s",
                 _("client had unexpected data pending tx after access check"));
        return -1;
    }

    if (VIR_ALLOC(confirm) < 0)
        return -1;

1268 1269 1270 1271
    /* Checks have succeeded.  Write a '\1' byte back to the client to
     * indicate this (otherwise the socket is abruptly closed).
     * (NB. The '\1' byte is sent in an encrypted record).
     */
1272 1273 1274 1275 1276 1277
    confirm->async = 1;
    confirm->bufferLength = 1;
    confirm->bufferOffset = 0;
    confirm->buffer[0] = '\1';

    client->tx = confirm;
1278 1279
    return 0;
}
D
Daniel P. Berrange 已提交
1280

1281 1282
#if HAVE_POLKIT
int qemudGetSocketIdentity(int fd, uid_t *uid, pid_t *pid) {
1283
# ifdef SO_PEERCRED
1284 1285 1286 1287
    struct ucred cr;
    unsigned int cr_len = sizeof (cr);

    if (getsockopt (fd, SOL_SOCKET, SO_PEERCRED, &cr, &cr_len) < 0) {
1288
        char ebuf[1024];
1289
        VIR_ERROR(_("Failed to verify client credentials: %s"),
1290
                  virStrerror(errno, ebuf, sizeof ebuf));
1291 1292 1293 1294 1295
        return -1;
    }

    *pid = cr.pid;
    *uid = cr.uid;
1296
# else
1297
    /* XXX Many more OS support UNIX socket credentials we could port to. See dbus ....*/
1298 1299
#  error "UNIX socket credentials not supported/implemented on this platform yet..."
# endif
1300 1301 1302 1303
    return 0;
}
#endif

1304

D
Daniel P. Berrange 已提交
1305 1306
static int qemudDispatchServer(struct qemud_server *server, struct qemud_socket *sock) {
    int fd;
1307 1308
    virSocketAddr addr;
    char *addrstr = NULL;
1309
    struct qemud_client *client = NULL;
1310
    int no_slow_start = 1;
1311
    int i;
D
Daniel P. Berrange 已提交
1312

1313 1314
    addr.len = sizeof(addr.data.stor);
    if ((fd = accept(sock->fd, &addr.data.sa, &addr.len)) < 0) {
1315
        char ebuf[1024];
D
Daniel P. Berrange 已提交
1316 1317
        if (errno == EAGAIN)
            return 0;
1318 1319
        VIR_ERROR(_("Failed to accept connection: %s"),
                  virStrerror(errno, ebuf, sizeof ebuf));
D
Daniel P. Berrange 已提交
1320 1321
        return -1;
    }
1322 1323 1324 1325
    if (!(addrstr = virSocketFormatAddrFull(&addr, true, ";"))) {
        VIR_ERROR0(_("Failed to format addresss: out of memory"));
        goto error;
    }
D
Daniel P. Berrange 已提交
1326

1327 1328
    PROBE(CLIENT_CONNECT, "fd=%d, readonly=%d localAddr=%s remoteAddr=%s",
          fd, sock->readonly, sock->addrstr, addrstr);
1329

1330
    if (server->nclients >= max_clients) {
1331 1332
        VIR_ERROR(_("Too many active clients (%d), dropping connection from %s"),
                  max_clients, addrstr);
1333
        goto error;
1334 1335
    }

1336 1337
    if (VIR_RESIZE_N(server->clients, server->nclients_max,
                     server->nclients, 1) < 0) {
1338
        VIR_ERROR0(_("Out of memory allocating clients"));
1339
        goto error;
1340 1341
    }

J
John Levon 已提交
1342 1343 1344 1345 1346 1347 1348 1349 1350
#ifdef __sun
    {
        ucred_t *ucred = NULL;
        const priv_set_t *privs;

        if (getpeerucred (fd, &ucred) == -1 ||
            (privs = ucred_getprivset (ucred, PRIV_EFFECTIVE)) == NULL) {
            if (ucred != NULL)
                ucred_free (ucred);
1351
            goto error;
J
John Levon 已提交
1352 1353 1354 1355
        }

        if (!priv_ismember (privs, PRIV_VIRT_MANAGE)) {
            ucred_free (ucred);
1356
            goto error;
J
John Levon 已提交
1357 1358 1359 1360 1361 1362
        }

        ucred_free (ucred);
    }
#endif /* __sun */

1363 1364 1365 1366
    /* Disable Nagle.  Unix sockets will ignore this. */
    setsockopt (fd, IPPROTO_TCP, TCP_NODELAY, (void *)&no_slow_start,
                sizeof no_slow_start);

1367 1368
    if (virSetCloseExec(fd) < 0 ||
        virSetNonBlock(fd) < 0) {
1369
        goto error;
D
Daniel P. Berrange 已提交
1370 1371
    }

1372
    if (VIR_ALLOC(client) < 0)
1373
        goto error;
1374
    if (virMutexInit(&client->lock) < 0) {
1375
        VIR_ERROR0(_("cannot initialize mutex"));
1376
        goto error;
1377
    }
1378

1379
    client->magic = QEMUD_CLIENT_MAGIC;
D
Daniel P. Berrange 已提交
1380 1381
    client->fd = fd;
    client->readonly = sock->readonly;
1382
    client->type = sock->type;
1383
    client->auth = sock->auth;
1384 1385 1386
    client->addr = addr;
    client->addrstr = addrstr;
    addrstr = NULL;
1387

1388 1389 1390 1391
    for (i = 0 ; i < VIR_DOMAIN_EVENT_ID_LAST ; i++) {
        client->domainEventCallbackID[i] = -1;
    }

1392 1393
    /* Prepare one for packet receive */
    if (VIR_ALLOC(client->rx) < 0)
1394
        goto error;
1395 1396 1397
    client->rx->bufferLength = REMOTE_MESSAGE_HEADER_XDR_LEN;


1398 1399 1400 1401 1402 1403 1404 1405 1406 1407
#if HAVE_POLKIT
    /* Only do policy checks for non-root - allow root user
       through with no checks, as a fail-safe - root can easily
       change policykit policy anyway, so its pointless trying
       to restrict root */
    if (client->auth == REMOTE_AUTH_POLKIT) {
        uid_t uid;
        pid_t pid;

        if (qemudGetSocketIdentity(client->fd, &uid, &pid) < 0)
1408
            goto error;
1409

1410
        /* Client is running as root, so disable auth */
1411
        if (uid == 0) {
1412
            VIR_INFO(_("Turn off polkit auth for privileged client pid %d from %s"),
E
Eric Blake 已提交
1413
                     pid, client->addrstr);
1414 1415 1416 1417 1418
            client->auth = REMOTE_AUTH_NONE;
        }
    }
#endif

1419
    if (client->type != QEMUD_SOCK_TYPE_TLS) {
1420
        /* Plain socket, so prepare to read first message */
1421
        if (qemudRegisterClientEvent (server, client) < 0)
1422
            goto error;
1423 1424 1425
    } else {
        int ret;

1426 1427
        client->tlssession = remoteInitializeTLSSession ();
        if (client->tlssession == NULL)
1428
            goto error;
1429

1430
        gnutls_transport_set_ptr (client->tlssession,
1431 1432 1433
                                  (gnutls_transport_ptr_t) (long) fd);

        /* Begin the TLS handshake. */
1434
        ret = gnutls_handshake (client->tlssession);
1435
        if (ret == 0) {
1436 1437
            client->handshake = 0;

1438 1439
            /* Unlikely, but ...  Next step is to check the certificate. */
            if (remoteCheckAccess (client) == -1)
1440
                goto error;
D
Daniel P. Berrange 已提交
1441

1442
            /* Handshake & cert check OK,  so prepare to read first message */
1443
            if (qemudRegisterClientEvent(server, client) < 0)
1444
                goto error;
1445
        } else if (ret == GNUTLS_E_INTERRUPTED || ret == GNUTLS_E_AGAIN) {
1446 1447
            /* Most likely, need to do more handshake data */
            client->handshake = 1;
D
Daniel P. Berrange 已提交
1448

1449
            if (qemudRegisterClientEvent (server, client) < 0)
1450
                goto error;
1451
        } else {
1452
            PROBE(CLIENT_TLS_FAIL, "fd=%d", client->fd);
1453
            VIR_ERROR(_("TLS handshake failed for client %s: %s"),
E
Eric Blake 已提交
1454
                      client->addrstr, gnutls_strerror (ret));
1455
            goto error;
1456 1457
        }
    }
D
Daniel P. Berrange 已提交
1458

1459
    server->clients[server->nclients++] = client;
D
Daniel P. Berrange 已提交
1460

1461 1462 1463 1464 1465 1466 1467 1468 1469 1470 1471 1472 1473
    if (server->nclients > server->nactiveworkers &&
        server->nactiveworkers < server->nworkers) {
        for (i = 0 ; i < server->nworkers ; i++) {
            if (!server->workers[i].hasThread) {
                if (qemudStartWorker(server, &server->workers[i]) < 0)
                    return -1;
                server->nactiveworkers++;
                break;
            }
        }
    }


D
Daniel P. Berrange 已提交
1474
    return 0;
1475

1476 1477 1478
error:
    if (client) {
        if (client->tlssession) gnutls_deinit (client->tlssession);
1479 1480
        if (client) {
            VIR_FREE(client->addrstr);
1481
            VIR_FREE(client->rx);
1482
        }
1483 1484
        VIR_FREE(client);
    }
1485
    VIR_FREE(addrstr);
1486
    VIR_FORCE_CLOSE(fd);
1487
    PROBE(CLIENT_DISCONNECT, "fd=%d", fd);
1488
    return -1;
D
Daniel P. Berrange 已提交
1489 1490 1491
}


1492 1493 1494 1495 1496 1497 1498
/*
 * You must hold lock for at least the client
 * We don't free stuff here, merely disconnect the client's
 * network socket & resources.
 * We keep the libvirt connection open until any async
 * jobs have finished, then clean it up elsehwere
 */
1499
void qemudDispatchClientFailure(struct qemud_client *client) {
1500 1501 1502 1503
    if (client->watch != -1) {
        virEventRemoveHandleImpl(client->watch);
        client->watch = -1;
    }
D
Daniel P. Berrange 已提交
1504

1505
    /* Deregister event delivery callback */
1506 1507 1508 1509 1510 1511 1512 1513 1514 1515 1516
    if (client->conn) {
        int i;

        for (i = 0 ; i < VIR_DOMAIN_EVENT_ID_LAST ; i++) {
            if (client->domainEventCallbackID[i] != -1) {
                DEBUG("Deregistering to relay remote events %d", i);
                virConnectDomainEventDeregisterAny(client->conn,
                                                   client->domainEventCallbackID[i]);
            }
            client->domainEventCallbackID[i] = -1;
        }
1517 1518
    }

1519
#if HAVE_SASL
1520 1521 1522 1523
    if (client->saslconn) {
        sasl_dispose(&client->saslconn);
        client->saslconn = NULL;
    }
1524
    VIR_FREE(client->saslUsername);
1525
#endif
1526 1527 1528 1529 1530
    if (client->tlssession) {
        gnutls_deinit (client->tlssession);
        client->tlssession = NULL;
    }
    if (client->fd != -1) {
1531
        PROBE(CLIENT_DISCONNECT, "fd=%d", client->fd);
1532
        VIR_FORCE_CLOSE(client->fd);
1533
    }
1534
    VIR_FREE(client->addrstr);
1535 1536 1537 1538 1539 1540 1541 1542
}


/* Caller must hold server lock */
static struct qemud_client *qemudPendingJob(struct qemud_server *server)
{
    int i;
    for (i = 0 ; i < server->nclients ; i++) {
1543
        virMutexLock(&server->clients[i]->lock);
1544
        if (server->clients[i]->dx) {
1545 1546 1547
            /* Delibrately don't unlock client - caller wants the lock */
            return server->clients[i];
        }
1548
        virMutexUnlock(&server->clients[i]->lock);
1549 1550
    }
    return NULL;
D
Daniel P. Berrange 已提交
1551 1552
}

1553 1554
static void *qemudWorker(void *data)
{
1555 1556
    struct qemud_worker *worker = data;
    struct qemud_server *server = worker->server;
1557 1558

    while (1) {
1559
        struct qemud_client *client = NULL;
1560
        struct qemud_client_message *msg;
1561

1562
        virMutexLock(&server->lock);
1563 1564 1565
        while ((client = qemudPendingJob(server)) == NULL) {
            if (worker->quitRequest ||
                virCondWait(&server->job, &server->lock) < 0) {
1566 1567 1568 1569
                virMutexUnlock(&server->lock);
                return NULL;
            }
        }
1570
        if (worker->quitRequest) {
1571
            virMutexUnlock(&client->lock);
1572 1573 1574 1575
            virMutexUnlock(&server->lock);
            return NULL;
        }
        worker->processingCall = 1;
1576
        virMutexUnlock(&server->lock);
1577 1578 1579 1580

        /* We own a locked client now... */
        client->refs++;

1581
        /* Remove our message from dispatch queue while we use it */
1582
        msg = qemudClientMessageQueueServe(&client->dx);
1583 1584 1585

        /* This function drops the lock during dispatch,
         * and re-acquires it before returning */
1586
        if (remoteDispatchClientRequest (server, client, msg) < 0) {
1587
            VIR_FREE(msg);
1588 1589 1590 1591 1592
            qemudDispatchClientFailure(client);
            client->refs--;
            virMutexUnlock(&client->lock);
            continue;
        }
1593 1594

        client->refs--;
1595
        virMutexUnlock(&client->lock);
1596 1597 1598 1599

        virMutexLock(&server->lock);
        worker->processingCall = 0;
        virMutexUnlock(&server->lock);
1600 1601
    }
}
D
Daniel P. Berrange 已提交
1602

E
Eric Blake 已提交
1603 1604 1605 1606
static int
qemudStartWorker(struct qemud_server *server,
                 struct qemud_worker *worker)
{
1607
    pthread_attr_t attr;
E
Eric Blake 已提交
1608 1609 1610 1611
    int ret = -1;

    if (pthread_attr_init(&attr) != 0)
        return -1;
1612 1613 1614 1615
    /* We want to join workers, so don't detach them */
    /*pthread_attr_setdetachstate(&attr, 1);*/

    if (worker->hasThread)
E
Eric Blake 已提交
1616
        goto cleanup;
1617 1618 1619 1620 1621 1622 1623 1624 1625 1626 1627 1628

    worker->server = server;
    worker->hasThread = 1;
    worker->quitRequest = 0;
    worker->processingCall = 0;

    if (pthread_create(&worker->thread,
                       &attr,
                       qemudWorker,
                       worker) != 0) {
        worker->hasThread = 0;
        worker->server = NULL;
E
Eric Blake 已提交
1629
        goto cleanup;
1630 1631
    }

E
Eric Blake 已提交
1632 1633 1634 1635
    ret = 0;
cleanup:
    pthread_attr_destroy(&attr);
    return ret;
1636 1637
}

D
Daniel P. Berrange 已提交
1638

1639 1640 1641 1642 1643 1644 1645 1646 1647 1648 1649 1650 1651
/*
 * Read data into buffer using wire decoding (plain or TLS)
 *
 * Returns:
 *   -1 on error or EOF
 *    0 on EAGAIN
 *    n number of bytes
 */
static ssize_t qemudClientReadBuf(struct qemud_client *client,
                                  char *data, ssize_t len) {
    ssize_t ret;

    if (len < 0) {
1652 1653
        VIR_ERROR(_("unexpected negative length request %lld"),
                  (long long int) len);
1654 1655 1656
        qemudDispatchClientFailure(client);
        return -1;
    }
1657 1658 1659

    /*qemudDebug ("qemudClientRead: len = %d", len);*/

1660
    if (!client->tlssession) {
1661
        char ebuf[1024];
1662 1663 1664 1665 1666 1667
        ret = read (client->fd, data, len);
        if (ret == -1 && (errno == EAGAIN ||
                          errno == EINTR))
            return 0;
        if (ret <= 0) {
            if (ret != 0)
1668 1669
                VIR_ERROR(_("read: %s"),
                          virStrerror (errno, ebuf, sizeof ebuf));
1670
            qemudDispatchClientFailure(client);
1671 1672 1673
            return -1;
        }
    } else {
1674
        ret = gnutls_record_recv (client->tlssession, data, len);
1675 1676 1677 1678 1679 1680 1681 1682 1683

        if (ret < 0 && (ret == GNUTLS_E_AGAIN ||
                        ret == GNUTLS_E_INTERRUPTED))
            return 0;
        if (ret <= 0) {
            if (ret != 0)
                VIR_ERROR(_("gnutls_record_recv: %s"),
                          gnutls_strerror (ret));
            qemudDispatchClientFailure(client);
1684 1685
            return -1;
        }
D
Daniel P. Berrange 已提交
1686
    }
1687

1688 1689 1690
    return ret;
}

1691 1692 1693 1694 1695 1696 1697 1698 1699 1700 1701 1702 1703 1704 1705 1706 1707 1708
/*
 * Read data into buffer without decoding
 *
 * Returns:
 *   -1 on error or EOF
 *    0 on EAGAIN
 *    n number of bytes
 */
static ssize_t qemudClientReadPlain(struct qemud_client *client) {
    ssize_t ret;
    ret = qemudClientReadBuf(client,
                             client->rx->buffer + client->rx->bufferOffset,
                             client->rx->bufferLength - client->rx->bufferOffset);
    if (ret <= 0)
        return ret; /* -1 error, 0 eagain */

    client->rx->bufferOffset += ret;
    return ret;
D
Daniel P. Berrange 已提交
1709 1710
}

1711
#if HAVE_SASL
1712 1713 1714 1715 1716 1717 1718 1719 1720 1721
/*
 * Read data into buffer decoding with SASL
 *
 * Returns:
 *   -1 on error or EOF
 *    0 on EAGAIN
 *    n number of bytes
 */
static ssize_t qemudClientReadSASL(struct qemud_client *client) {
    ssize_t got, want;
1722 1723 1724 1725 1726 1727 1728 1729 1730 1731

    /* We're doing a SSF data read, so now its times to ensure
     * future writes are under SSF too.
     *
     * cf remoteSASLCheckSSF in remote.c
     */
    client->saslSSF |= QEMUD_SASL_SSF_WRITE;

    /* Need to read some more data off the wire */
    if (client->saslDecoded == NULL) {
1732
        int ret;
1733
        char encoded[8192];
1734 1735 1736 1737 1738 1739 1740 1741 1742 1743 1744 1745
        ssize_t encodedLen = sizeof(encoded);
        encodedLen = qemudClientReadBuf(client, encoded, encodedLen);

        if (encodedLen <= 0)
            return encodedLen;

        ret = sasl_decode(client->saslconn, encoded, encodedLen,
                          &client->saslDecoded, &client->saslDecodedLength);
        if (ret != SASL_OK) {
            VIR_ERROR(_("failed to decode SASL data %s"),
                      sasl_errstring(ret, NULL, NULL));
            qemudDispatchClientFailure(client);
1746
            return -1;
1747
        }
1748 1749 1750 1751 1752 1753

        client->saslDecodedOffset = 0;
    }

    /* Some buffered decoded data to return now */
    got = client->saslDecodedLength - client->saslDecodedOffset;
1754
    want = client->rx->bufferLength - client->rx->bufferOffset;
1755 1756 1757 1758

    if (want > got)
        want = got;

1759
    memcpy(client->rx->buffer + client->rx->bufferOffset,
1760 1761
           client->saslDecoded + client->saslDecodedOffset, want);
    client->saslDecodedOffset += want;
1762
    client->rx->bufferOffset += want;
1763 1764 1765 1766 1767 1768

    if (client->saslDecodedOffset == client->saslDecodedLength) {
        client->saslDecoded = NULL;
        client->saslDecodedOffset = client->saslDecodedLength = 0;
    }

1769
    return want;
1770 1771 1772
}
#endif

1773 1774 1775 1776 1777
/*
 * Read as much data off wire as possible till we fill our
 * buffer, or would block on I/O
 */
static ssize_t qemudClientRead(struct qemud_client *client) {
1778 1779
#if HAVE_SASL
    if (client->saslSSF & QEMUD_SASL_SSF_READ)
1780
        return qemudClientReadSASL(client);
1781 1782
    else
#endif
1783
        return qemudClientReadPlain(client);
1784 1785 1786
}


1787 1788 1789 1790 1791
/*
 * Read data until we get a complete message to process
 */
static void qemudDispatchClientRead(struct qemud_server *server,
                                    struct qemud_client *client) {
1792
    /*qemudDebug ("qemudDispatchClientRead: mode = %d", client->mode);*/
D
Daniel P. Berrange 已提交
1793

1794 1795 1796
readmore:
    if (qemudClientRead(client) < 0)
        return; /* Error */
1797

1798 1799
    if (client->rx->bufferOffset < client->rx->bufferLength)
        return; /* Still not read enough */
1800

1801 1802 1803 1804
    /* Either done with length word header */
    if (client->rx->bufferLength == REMOTE_MESSAGE_HEADER_XDR_LEN) {
        unsigned int len;
        XDR x;
1805

1806
        xdrmem_create(&x, client->rx->buffer, client->rx->bufferLength, XDR_DECODE);
1807

1808 1809
        if (!xdr_u_int(&x, &len)) {
            xdr_destroy (&x);
1810
            DEBUG0("Failed to decode packet length");
1811
            qemudDispatchClientFailure(client);
D
Daniel P. Berrange 已提交
1812 1813
            return;
        }
1814
        xdr_destroy (&x);
1815

1816 1817 1818
        if (len < REMOTE_MESSAGE_HEADER_XDR_LEN) {
            DEBUG("Packet length %u too small", len);
            qemudDispatchClientFailure(client);
1819
            return;
D
Daniel P. Berrange 已提交
1820
        }
1821

1822 1823 1824 1825 1826 1827
        /* Length includes the size of the length word itself */
        len -= REMOTE_MESSAGE_HEADER_XDR_LEN;

        if (len > REMOTE_MESSAGE_MAX) {
            DEBUG("Packet length %u too large", len);
            qemudDispatchClientFailure(client);
1828 1829 1830
            return;
        }

1831 1832
        /* Prepare to read rest of message */
        client->rx->bufferLength += len;
1833

1834
        qemudUpdateClientEvent(client);
D
Daniel P. Berrange 已提交
1835

1836 1837 1838 1839 1840
        /* Try and read payload immediately instead of going back
           into poll() because chances are the data is already
           waiting for us */
        goto readmore;
    } else {
1841 1842 1843 1844 1845 1846 1847 1848 1849 1850 1851 1852 1853
        /* Grab the completed message */
        struct qemud_client_message *msg = qemudClientMessageQueueServe(&client->rx);
        struct qemud_client_filter *filter;

        /* Decode the header so we can use it for routing decisions */
        if (remoteDecodeClientMessageHeader(msg) < 0) {
            VIR_FREE(msg);
            qemudDispatchClientFailure(client);
        }

        /* Check if any filters match this message */
        filter = client->filters;
        while (filter) {
1854 1855 1856
            int ret;
            ret = (filter->query)(client, msg, filter->opaque);
            if (ret == 1) {
1857 1858
                msg = NULL;
                break;
1859 1860 1861 1862
            } else if (ret == -1) {
                VIR_FREE(msg);
                qemudDispatchClientFailure(client);
                return;
1863 1864 1865 1866
            }
            filter = filter->next;
        }

1867
        /* Move completed message to the end of the dispatch queue */
1868 1869
        if (msg)
            qemudClientMessageQueuePush(&client->dx, msg);
1870 1871 1872 1873 1874 1875
        client->nrequests++;

        /* Possibly need to create another receive buffer */
        if ((client->nrequests < max_client_requests &&
             VIR_ALLOC(client->rx) < 0)) {
            qemudDispatchClientFailure(client);
D
Daniel P. Berrange 已提交
1876
        } else {
1877 1878 1879
            if (client->rx)
                client->rx->bufferLength = REMOTE_MESSAGE_HEADER_XDR_LEN;

1880 1881 1882 1883
            qemudUpdateClientEvent(client);

            /* Tell one of the workers to get on with it... */
            virCondSignal(&server->job);
D
Daniel P. Berrange 已提交
1884
        }
D
Daniel P. Berrange 已提交
1885 1886 1887 1888
    }
}


1889 1890 1891 1892 1893 1894 1895 1896 1897 1898 1899 1900 1901
/*
 * Send a chunk of data using wire encoding (plain or TLS)
 *
 * Returns:
 *   -1 on error
 *    0 on EAGAIN
 *    n number of bytes
 */
static ssize_t qemudClientWriteBuf(struct qemud_client *client,
                                   const char *data, ssize_t len) {
    ssize_t ret;

    if (len < 0) {
1902 1903
        VIR_ERROR(_("unexpected negative length request %lld"),
                  (long long int) len);
1904 1905 1906 1907
        qemudDispatchClientFailure(client);
        return -1;
    }

1908
    if (!client->tlssession) {
1909
        char ebuf[1024];
1910 1911 1912
        if ((ret = write(client->fd, data, len)) == -1) {
            if (errno == EAGAIN || errno == EINTR)
                return 0;
1913
            VIR_ERROR(_("write: %s"), virStrerror (errno, ebuf, sizeof ebuf));
1914
            qemudDispatchClientFailure(client);
1915 1916 1917
            return -1;
        }
    } else {
1918
        ret = gnutls_record_send (client->tlssession, data, len);
1919 1920 1921 1922 1923 1924 1925
        if (ret < 0) {
            if (ret == GNUTLS_E_INTERRUPTED ||
                ret == GNUTLS_E_AGAIN)
                return 0;

            VIR_ERROR(_("gnutls_record_send: %s"), gnutls_strerror (ret));
            qemudDispatchClientFailure(client);
1926 1927
            return -1;
        }
D
Daniel P. Berrange 已提交
1928
    }
1929 1930
    return ret;
}
1931

1932

1933 1934 1935 1936 1937 1938 1939 1940 1941 1942 1943 1944 1945 1946 1947 1948
/*
 * Send client->tx using no encoding
 *
 * Returns:
 *   -1 on error or EOF
 *    0 on EAGAIN
 *    n number of bytes
 */
static int qemudClientWritePlain(struct qemud_client *client) {
    int ret = qemudClientWriteBuf(client,
                                  client->tx->buffer + client->tx->bufferOffset,
                                  client->tx->bufferLength - client->tx->bufferOffset);
    if (ret <= 0)
        return ret; /* -1 error, 0 = egain */
    client->tx->bufferOffset += ret;
    return ret;
D
Daniel P. Berrange 已提交
1949 1950 1951
}


1952
#if HAVE_SASL
1953 1954 1955 1956 1957 1958 1959 1960 1961
/*
 * Send client->tx using SASL encoding
 *
 * Returns:
 *   -1 on error
 *    0 on EAGAIN
 *    n number of bytes
 */
static int qemudClientWriteSASL(struct qemud_client *client) {
1962 1963 1964 1965
    int ret;

    /* Not got any pending encoded data, so we need to encode raw stuff */
    if (client->saslEncoded == NULL) {
1966 1967 1968
        ret = sasl_encode(client->saslconn,
                          client->tx->buffer + client->tx->bufferOffset,
                          client->tx->bufferLength - client->tx->bufferOffset,
1969 1970 1971
                          &client->saslEncoded,
                          &client->saslEncodedLength);

1972 1973 1974 1975 1976 1977 1978
        if (ret != SASL_OK) {
            VIR_ERROR(_("failed to encode SASL data %s"),
                      sasl_errstring(ret, NULL, NULL));
            qemudDispatchClientFailure(client);
            return -1;
        }

1979 1980 1981 1982
        client->saslEncodedOffset = 0;
    }

    /* Send some of the encoded stuff out on the wire */
1983
    ret = qemudClientWriteBuf(client,
1984 1985 1986
                              client->saslEncoded + client->saslEncodedOffset,
                              client->saslEncodedLength - client->saslEncodedOffset);

1987 1988
    if (ret <= 0)
        return ret; /* -1 error, 0 == egain */
1989 1990 1991 1992 1993 1994 1995 1996

    /* Note how much we sent */
    client->saslEncodedOffset += ret;

    /* Sent all encoded, so update raw buffer to indicate completion */
    if (client->saslEncodedOffset == client->saslEncodedLength) {
        client->saslEncoded = NULL;
        client->saslEncodedOffset = client->saslEncodedLength = 0;
1997 1998 1999

        /* Mark as complete, so caller detects completion */
        client->tx->bufferOffset = client->tx->bufferLength;
2000 2001
    }

2002
    return ret;
2003 2004 2005
}
#endif

2006 2007 2008 2009 2010 2011 2012 2013 2014
/*
 * Send as much data in the client->tx as possible
 *
 * Returns:
 *   -1 on error or EOF
 *    0 on EAGAIN
 *    n number of bytes
 */
static ssize_t qemudClientWrite(struct qemud_client *client) {
2015 2016
#if HAVE_SASL
    if (client->saslSSF & QEMUD_SASL_SSF_WRITE)
2017
        return qemudClientWriteSASL(client);
2018 2019
    else
#endif
2020
        return qemudClientWritePlain(client);
2021 2022 2023
}


2024 2025 2026 2027
void
qemudClientMessageRelease(struct qemud_client *client,
                          struct qemud_client_message *msg)
{
2028 2029 2030
    if (msg->streamTX) {
        remoteStreamMessageFinished(client, msg);
    } else if (!msg->async)
2031 2032 2033 2034 2035 2036 2037 2038 2039 2040 2041 2042 2043 2044 2045 2046 2047 2048
        client->nrequests--;

    /* See if the recv queue is currently throttled */
    if (!client->rx &&
        client->nrequests < max_client_requests) {
        /* Reset message record for next RX attempt */
        memset(msg, 0, sizeof(*msg));
        client->rx = msg;
        /* Get ready to receive next message */
        client->rx->bufferLength = REMOTE_MESSAGE_HEADER_XDR_LEN;
    } else {
        VIR_FREE(msg);
    }

    qemudUpdateClientEvent(client);
}


2049 2050 2051 2052 2053
/*
 * Process all queued client->tx messages until
 * we would block on I/O
 */
static void
2054
qemudDispatchClientWrite(struct qemud_client *client) {
2055 2056
    while (client->tx) {
        ssize_t ret;
2057

2058 2059 2060 2061 2062 2063 2064 2065 2066 2067 2068 2069 2070 2071
        ret = qemudClientWrite(client);
        if (ret < 0) {
            qemudDispatchClientFailure(client);
            return;
        }
        if (ret == 0)
            return; /* Would block on write EAGAIN */

        if (client->tx->bufferOffset == client->tx->bufferLength) {
            struct qemud_client_message *reply;

            /* Get finished reply from head of tx queue */
            reply = qemudClientMessageQueueServe(&client->tx);

2072
            qemudClientMessageRelease(client, reply);
2073

2074 2075
            if (client->closing)
                qemudDispatchClientFailure(client);
2076
         }
2077
    }
2078
}
2079

2080
static void
2081
qemudDispatchClientHandshake(struct qemud_client *client) {
2082 2083 2084 2085
    int ret;
    /* Continue the handshake. */
    ret = gnutls_handshake (client->tlssession);
    if (ret == 0) {
2086 2087
        client->handshake = 0;

2088 2089 2090
        /* Finished.  Next step is to check the certificate. */
        if (remoteCheckAccess (client) == -1)
            qemudDispatchClientFailure(client);
2091 2092
        else
            qemudUpdateClientEvent(client);
2093 2094 2095 2096 2097
    } else if (ret == GNUTLS_E_AGAIN ||
               ret == GNUTLS_E_INTERRUPTED) {
        /* Carry on waiting for more handshake. Update
           the events just in case handshake data flow
           direction has changed */
2098
        qemudUpdateClientEvent (client);
2099
    } else {
2100
        PROBE(CLIENT_TLS_FAIL, "fd=%d", client->fd);
2101 2102 2103 2104
        /* Fatal error in handshake */
        VIR_ERROR(_("TLS handshake failed: %s"),
                  gnutls_strerror (ret));
        qemudDispatchClientFailure(client);
D
Daniel P. Berrange 已提交
2105 2106 2107
    }
}

2108
static void
2109
qemudDispatchClientEvent(int watch, int fd, int events, void *opaque) {
D
Daniel P. Berrange 已提交
2110
    struct qemud_server *server = (struct qemud_server *)opaque;
2111 2112
    struct qemud_client *client = NULL;
    int i;
2113

2114
    virMutexLock(&server->lock);
2115

2116
    for (i = 0 ; i < server->nclients ; i++) {
2117
        virMutexLock(&server->clients[i]->lock);
2118 2119
        if (server->clients[i]->watch == watch) {
            client = server->clients[i];
D
Daniel P. Berrange 已提交
2120
            break;
2121
        }
2122
        virMutexUnlock(&server->clients[i]->lock);
D
Daniel P. Berrange 已提交
2123
    }
2124

2125 2126
    virMutexUnlock(&server->lock);

2127
    if (!client) {
D
Daniel P. Berrange 已提交
2128
        return;
2129 2130
    }

2131 2132
    if (client->fd != fd) {
        virMutexUnlock(&client->lock);
2133
        return;
2134 2135 2136 2137 2138
    }

    if (events & (VIR_EVENT_HANDLE_WRITABLE |
                  VIR_EVENT_HANDLE_READABLE)) {
        if (client->handshake) {
2139
            qemudDispatchClientHandshake(client);
2140 2141
        } else {
            if (events & VIR_EVENT_HANDLE_WRITABLE)
2142
                qemudDispatchClientWrite(client);
2143 2144 2145 2146 2147 2148 2149 2150 2151 2152
            if (events & VIR_EVENT_HANDLE_READABLE)
                qemudDispatchClientRead(server, client);
        }
    }

    /* NB, will get HANGUP + READABLE at same time upon
     * disconnect */
    if (events & (VIR_EVENT_HANDLE_ERROR |
                  VIR_EVENT_HANDLE_HANGUP))
        qemudDispatchClientFailure(client);
2153

2154
    virMutexUnlock(&client->lock);
D
Daniel P. Berrange 已提交
2155 2156
}

2157 2158 2159 2160 2161 2162

/*
 * @client: a locked client object
 */
static int
qemudCalculateHandleMode(struct qemud_client *client) {
2163 2164 2165
    int mode = 0;

    if (client->handshake) {
2166
        if (gnutls_record_get_direction (client->tlssession) == 0)
2167
            mode |= VIR_EVENT_HANDLE_READABLE;
2168
        else
2169 2170 2171 2172 2173 2174
            mode |= VIR_EVENT_HANDLE_WRITABLE;
    } else {
        /* If there is a message on the rx queue then
         * we're wanting more input */
        if (client->rx)
            mode |= VIR_EVENT_HANDLE_READABLE;
2175

2176 2177 2178 2179
        /* If there are one or more messages to send back to client,
           then monitor for writability on socket */
        if (client->tx)
            mode |= VIR_EVENT_HANDLE_WRITABLE;
2180 2181
    }

2182 2183 2184 2185 2186 2187 2188 2189 2190 2191 2192 2193 2194 2195 2196 2197 2198 2199
    return mode;
}

/*
 * @server: a locked or unlocked server object
 * @client: a locked client object
 */
int qemudRegisterClientEvent(struct qemud_server *server,
                             struct qemud_client *client) {
    int mode;

    mode = qemudCalculateHandleMode(client);

    if ((client->watch = virEventAddHandleImpl(client->fd,
                                               mode,
                                               qemudDispatchClientEvent,
                                               server, NULL)) < 0)
        return -1;
D
Daniel P. Berrange 已提交
2200 2201 2202 2203

    return 0;
}

2204 2205 2206 2207 2208 2209 2210 2211 2212 2213 2214 2215
/*
 * @client: a locked client object
 */
void qemudUpdateClientEvent(struct qemud_client *client) {
    int mode;

    mode = qemudCalculateHandleMode(client);

    virEventUpdateHandleImpl(client->watch, mode);
}


2216
static void
2217
qemudDispatchServerEvent(int watch, int fd, int events, void *opaque) {
D
Daniel P. Berrange 已提交
2218
    struct qemud_server *server = (struct qemud_server *)opaque;
2219 2220
    struct qemud_socket *sock;

2221
    virMutexLock(&server->lock);
2222 2223

    sock = server->sockets;
D
Daniel P. Berrange 已提交
2224

D
Daniel P. Berrange 已提交
2225
    while (sock) {
2226
        if (sock->watch == watch)
D
Daniel P. Berrange 已提交
2227 2228 2229
            break;

        sock = sock->next;
D
Daniel P. Berrange 已提交
2230
    }
D
Daniel P. Berrange 已提交
2231

2232
    if (sock && sock->fd == fd && events)
D
Daniel P. Berrange 已提交
2233
        qemudDispatchServer(server, sock);
2234

2235
    virMutexUnlock(&server->lock);
D
Daniel P. Berrange 已提交
2236 2237 2238
}


2239
static int qemudOneLoop(void) {
2240
    sig_atomic_t errors;
D
Daniel P. Berrange 已提交
2241

D
Daniel P. Berrange 已提交
2242
    if (virEventRunOnce() < 0)
D
Daniel P. Berrange 已提交
2243 2244
        return -1;

2245 2246 2247
    /* Check for any signal handling errors and log them. */
    errors = sig_errors;
    if (errors) {
2248
        char ebuf[1024];
2249
        sig_errors -= errors;
2250
        VIR_ERROR(_("Signal handler reported %d errors: last error: %s"),
2251
                  errors, virStrerror (sig_lasterrno, ebuf, sizeof ebuf));
2252 2253 2254
        return -1;
    }

D
Daniel P. Berrange 已提交
2255 2256 2257
    return 0;
}

2258
static void qemudInactiveTimer(int timerid, void *data) {
2259
    struct qemud_server *server = (struct qemud_server *)data;
2260 2261 2262 2263 2264 2265 2266

    if (virStateActive() ||
        server->clients) {
        DEBUG0("Timer expired but still active, not shutting down");
        virEventUpdateTimeoutImpl(timerid, -1);
    } else {
        DEBUG0("Timer expired and inactive, shutting down");
2267
        server->quitEventThread = 1;
2268 2269 2270
    }
}

2271 2272 2273 2274 2275 2276 2277 2278 2279 2280 2281 2282 2283 2284 2285 2286 2287
static void qemudFreeClient(struct qemud_client *client) {
    while (client->rx) {
        struct qemud_client_message *msg
            = qemudClientMessageQueueServe(&client->rx);
        VIR_FREE(msg);
    }
    while (client->dx) {
        struct qemud_client_message *msg
            = qemudClientMessageQueueServe(&client->dx);
        VIR_FREE(msg);
    }
    while (client->tx) {
        struct qemud_client_message *msg
            = qemudClientMessageQueueServe(&client->tx);
        VIR_FREE(msg);
    }

2288 2289 2290
    while (client->streams)
        remoteRemoveClientStream(client, client->streams);

2291 2292 2293
    if (client->conn)
        virConnectClose(client->conn);
    virMutexDestroy(&client->lock);
E
Eric Blake 已提交
2294
    VIR_FREE(client->addrstr);
2295 2296 2297
    VIR_FREE(client);
}

2298 2299
static void *qemudRunLoop(void *opaque) {
    struct qemud_server *server = opaque;
2300
    int timerid = -1;
2301
    int i;
2302
    int timerActive = 0;
2303

2304
    virMutexLock(&server->lock);
2305

2306 2307 2308 2309 2310
    if (timeout > 0 &&
        (timerid = virEventAddTimeoutImpl(-1,
                                          qemudInactiveTimer,
                                          server, NULL)) < 0) {
        VIR_ERROR0(_("Failed to register shutdown timeout"));
2311
        return NULL;
2312 2313
    }

2314 2315 2316 2317
    if (min_workers > max_workers)
        max_workers = min_workers;

    server->nworkers = max_workers;
2318
    if (VIR_ALLOC_N(server->workers, server->nworkers) < 0) {
2319
        VIR_ERROR0(_("Failed to allocate workers"));
2320
        return NULL;
2321 2322
    }

2323 2324 2325 2326
    for (i = 0 ; i < min_workers ; i++) {
        if (qemudStartWorker(server, &server->workers[i]) < 0)
            goto cleanup;
        server->nactiveworkers++;
2327
    }
2328

2329
    for (;!server->quitEventThread;) {
2330 2331 2332 2333
        /* A shutdown timeout is specified, so check
         * if any drivers have active state, if not
         * shutdown after timeout seconds
         */
2334 2335 2336 2337 2338 2339 2340 2341 2342 2343 2344 2345 2346 2347 2348
        if (timeout > 0) {
            if (timerActive) {
                if (server->clients) {
                    DEBUG("Deactivating shutdown timer %d", timerid);
                    virEventUpdateTimeoutImpl(timerid, -1);
                    timerActive = 0;
                }
            } else {
                if (!virStateActive() &&
                    !server->clients) {
                    DEBUG("Activating shutdown timer %d", timerid);
                    virEventUpdateTimeoutImpl(timerid, timeout * 1000);
                    timerActive = 1;
                }
            }
2349 2350
        }

2351
        virMutexUnlock(&server->lock);
2352 2353
        if (qemudOneLoop() < 0) {
            virMutexLock(&server->lock);
2354
            DEBUG0("Loop iteration error, exiting");
2355
            break;
2356
        }
2357
        virMutexLock(&server->lock);
2358 2359 2360 2361

    reprocess:
        for (i = 0 ; i < server->nclients ; i++) {
            int inactive;
2362
            virMutexLock(&server->clients[i]->lock);
2363 2364
            inactive = server->clients[i]->fd == -1
                && server->clients[i]->refs == 0;
2365
            virMutexUnlock(&server->clients[i]->lock);
2366
            if (inactive) {
2367
                qemudFreeClient(server->clients[i]);
2368
                server->nclients--;
2369
                if (i < server->nclients)
2370 2371
                    memmove(server->clients + i,
                            server->clients + i + 1,
2372 2373
                            sizeof (*server->clients) * (server->nclients - i));

2374 2375
                VIR_SHRINK_N(server->clients, server->nclients_max,
                             server->nclients_max - server->nclients);
2376
                goto reprocess;
2377 2378
            }
        }
D
Daniel P. Berrange 已提交
2379

2380 2381 2382 2383 2384 2385 2386 2387 2388 2389 2390 2391 2392 2393 2394 2395 2396 2397 2398
        /* If number of active workers exceeds both the min_workers
         * threshold and the number of clients, then kill some
         * off */
        for (i = 0 ; (i < server->nworkers &&
                      server->nactiveworkers > server->nclients &&
                      server->nactiveworkers > min_workers) ; i++) {

            if (server->workers[i].hasThread &&
                !server->workers[i].processingCall) {
                server->workers[i].quitRequest = 1;

                virCondBroadcast(&server->job);
                virMutexUnlock(&server->lock);
                pthread_join(server->workers[i].thread, NULL);
                virMutexLock(&server->lock);
                server->workers[i].hasThread = 0;
                server->nactiveworkers--;
            }
        }
2399 2400
    }

2401
cleanup:
2402
    for (i = 0 ; i < server->nworkers ; i++) {
2403 2404 2405 2406 2407 2408
        if (!server->workers[i].hasThread)
            continue;

        server->workers[i].quitRequest = 1;
        virCondBroadcast(&server->job);

2409
        virMutexUnlock(&server->lock);
2410
        pthread_join(server->workers[i].thread, NULL);
2411
        virMutexLock(&server->lock);
2412
        server->workers[i].hasThread = 0;
2413
    }
2414
    VIR_FREE(server->workers);
2415 2416 2417 2418
    for (i = 0; i < server->nclients; i++)
        qemudFreeClient(server->clients[i]);
    server->nclients = 0;
    VIR_SHRINK_N(server->clients, server->nclients_max, server->nclients_max);
2419

2420
    virMutexUnlock(&server->lock);
2421
    return NULL;
D
Daniel P. Berrange 已提交
2422 2423
}

2424

E
Eric Blake 已提交
2425 2426 2427
static int
qemudStartEventLoop(struct qemud_server *server)
{
2428
    pthread_attr_t attr;
E
Eric Blake 已提交
2429 2430 2431 2432
    int ret = -1;

    if (pthread_attr_init(&attr) != 0)
        return -1;
2433 2434 2435 2436 2437 2438 2439
    /* We want to join the eventloop, so don't detach it */
    /*pthread_attr_setdetachstate(&attr, 1);*/

    if (pthread_create(&server->eventThread,
                       &attr,
                       qemudRunLoop,
                       server) != 0)
E
Eric Blake 已提交
2440
        goto cleanup;
2441 2442 2443

    server->hasEventThread = 1;

E
Eric Blake 已提交
2444 2445 2446 2447
    ret = 0;
cleanup:
    pthread_attr_destroy(&attr);
    return ret;
2448 2449 2450
}


D
Daniel P. Berrange 已提交
2451
static void qemudCleanup(struct qemud_server *server) {
2452 2453
    struct qemud_socket *sock;

2454 2455
    VIR_FORCE_CLOSE(server->sigread);
    VIR_FORCE_CLOSE(server->sigwrite);
2456 2457

    sock = server->sockets;
D
Daniel P. Berrange 已提交
2458
    while (sock) {
2459
        struct qemud_socket *next = sock->next;
2460 2461
        if (sock->watch)
            virEventRemoveHandleImpl(sock->watch);
2462
        VIR_FORCE_CLOSE(sock->fd);
2463 2464 2465 2466 2467 2468 2469

        /* Unlink unix domain sockets which are not in
         * the abstract namespace */
        if (sock->path &&
            sock->path[0] != '@')
            unlink(sock->path);
        VIR_FREE(sock->path);
2470
        VIR_FREE(sock->addrstr);
2471

2472
        VIR_FREE(sock);
2473
        sock = next;
D
Daniel P. Berrange 已提交
2474
    }
2475
    VIR_FREE(server->logDir);
2476

2477
#ifdef HAVE_SASL
2478 2479 2480
    if (server->saslUsernameWhitelist) {
        char **list = server->saslUsernameWhitelist;
        while (*list) {
2481
            VIR_FREE(*list);
2482 2483
            list++;
        }
2484
        VIR_FREE(server->saslUsernameWhitelist);
2485
    }
2486
#endif
2487

2488 2489 2490 2491 2492
#if HAVE_POLKIT0
        if (server->sysbus)
            dbus_connection_unref(server->sysbus);
#endif

2493
    virStateCleanup();
2494

2495 2496 2497 2498 2499 2500
    if (virCondDestroy(&server->job) < 0) {
        ;
    }
    virMutexDestroy(&server->lock);

    VIR_FREE(server);
D
Daniel P. Berrange 已提交
2501 2502
}

2503 2504 2505 2506 2507 2508 2509 2510 2511 2512 2513 2514 2515 2516 2517 2518
/* Allocate an array of malloc'd strings from the config file, filename
 * (used only in diagnostics), using handle "conf".  Upon error, return -1
 * and free any allocated memory.  Otherwise, save the array in *list_arg
 * and return 0.
 */
static int
remoteConfigGetStringList(virConfPtr conf, const char *key, char ***list_arg,
                          const char *filename)
{
    char **list;
    virConfValuePtr p = virConfGetValue (conf, key);
    if (!p)
        return 0;

    switch (p->type) {
    case VIR_CONF_STRING:
2519
        if (VIR_ALLOC_N(list, 2) < 0) {
2520
            VIR_ERROR(_("failed to allocate memory for %s config list"), key);
2521 2522 2523 2524 2525
            return -1;
        }
        list[0] = strdup (p->str);
        list[1] = NULL;
        if (list[0] == NULL) {
2526
            VIR_ERROR(_("failed to allocate memory for %s config list value"),
2527
                      key);
2528
            VIR_FREE(list);
2529 2530 2531 2532 2533 2534 2535 2536 2537
            return -1;
        }
        break;

    case VIR_CONF_LIST: {
        int i, len = 0;
        virConfValuePtr pp;
        for (pp = p->list; pp; pp = pp->next)
            len++;
2538
        if (VIR_ALLOC_N(list, 1+len) < 0) {
2539
            VIR_ERROR(_("failed to allocate memory for %s config list"), key);
2540 2541 2542 2543
            return -1;
        }
        for (i = 0, pp = p->list; pp; ++i, pp = pp->next) {
            if (pp->type != VIR_CONF_STRING) {
2544
                VIR_ERROR(_("remoteReadConfigFile: %s: %s:"
2545
                            " must be a string or list of strings"),
2546
                          filename, key);
2547
                VIR_FREE(list);
2548 2549 2550 2551 2552 2553
                return -1;
            }
            list[i] = strdup (pp->str);
            if (list[i] == NULL) {
                int j;
                for (j = 0 ; j < i ; j++)
2554 2555
                    VIR_FREE(list[j]);
                VIR_FREE(list);
2556 2557
                VIR_ERROR(_("failed to allocate memory for %s config list value"),
                          key);
2558 2559 2560 2561 2562 2563 2564 2565 2566
                return -1;
            }

        }
        list[i] = NULL;
        break;
    }

    default:
2567
        VIR_ERROR(_("remoteReadConfigFile: %s: %s:"
2568
                    " must be a string or list of strings"),
2569 2570 2571 2572 2573 2574 2575 2576 2577 2578 2579 2580 2581 2582
                  filename, key);
        return -1;
    }

    *list_arg = list;
    return 0;
}

/* A helper function used by each of the following macros.  */
static int
checkType (virConfValuePtr p, const char *filename,
           const char *key, virConfType required_type)
{
    if (p->type != required_type) {
2583
        VIR_ERROR(_("remoteReadConfigFile: %s: %s: invalid type:"
2584
                    " got %s; expected %s"), filename, key,
2585 2586 2587 2588 2589 2590 2591 2592 2593 2594 2595 2596 2597 2598 2599 2600 2601 2602 2603
                  virConfTypeName (p->type),
                  virConfTypeName (required_type));
        return -1;
    }
    return 0;
}

/* If there is no config data for the key, #var_name, then do nothing.
   If there is valid data of type VIR_CONF_STRING, and strdup succeeds,
   store the result in var_name.  Otherwise, (i.e. invalid type, or strdup
   failure), give a diagnostic and "goto" the cleanup-and-fail label.  */
#define GET_CONF_STR(conf, filename, var_name)                          \
    do {                                                                \
        virConfValuePtr p = virConfGetValue (conf, #var_name);          \
        if (p) {                                                        \
            if (checkType (p, filename, #var_name, VIR_CONF_STRING) < 0) \
                goto free_and_fail;                                     \
            (var_name) = strdup (p->str);                               \
            if ((var_name) == NULL) {                                   \
2604
                char ebuf[1024];                                        \
2605
                VIR_ERROR(_("remoteReadConfigFile: %s"),		\
2606
                          virStrerror(errno, ebuf, sizeof ebuf));       \
2607 2608 2609 2610 2611 2612 2613 2614 2615 2616 2617 2618 2619 2620 2621 2622
                goto free_and_fail;                                     \
            }                                                           \
        }                                                               \
    } while (0)

/* Like GET_CONF_STR, but for integral values.  */
#define GET_CONF_INT(conf, filename, var_name)                          \
    do {                                                                \
        virConfValuePtr p = virConfGetValue (conf, #var_name);          \
        if (p) {                                                        \
            if (checkType (p, filename, #var_name, VIR_CONF_LONG) < 0)  \
                goto free_and_fail;                                     \
            (var_name) = p->l;                                          \
        }                                                               \
    } while (0)

2623 2624 2625 2626 2627 2628 2629 2630

static int remoteConfigGetAuth(virConfPtr conf, const char *key, int *auth, const char *filename) {
    virConfValuePtr p;

    p = virConfGetValue (conf, key);
    if (!p)
        return 0;

2631
    if (checkType (p, filename, key, VIR_CONF_STRING) < 0)
2632 2633 2634 2635 2636 2637 2638 2639 2640 2641
        return -1;

    if (!p->str)
        return 0;

    if (STREQ(p->str, "none")) {
        *auth = REMOTE_AUTH_NONE;
#if HAVE_SASL
    } else if (STREQ(p->str, "sasl")) {
        *auth = REMOTE_AUTH_SASL;
2642 2643 2644 2645
#endif
#if HAVE_POLKIT
    } else if (STREQ(p->str, "polkit")) {
        *auth = REMOTE_AUTH_POLKIT;
2646 2647
#endif
    } else {
2648
        VIR_ERROR(_("remoteReadConfigFile: %s: %s: unsupported auth %s"),
2649
                  filename, key, p->str);
2650 2651 2652 2653 2654 2655
        return -1;
    }

    return 0;
}

2656 2657 2658 2659 2660 2661 2662 2663 2664 2665 2666 2667 2668 2669 2670 2671 2672 2673 2674 2675
#ifdef HAVE_SASL
static inline int
remoteReadSaslAllowedUsernameList (virConfPtr conf,
                                   struct qemud_server *server,
                                   const char *filename)
{
    return
        remoteConfigGetStringList (conf, "sasl_allowed_username_list",
                                   &server->saslUsernameWhitelist, filename);
}
#else
static inline int
remoteReadSaslAllowedUsernameList (virConfPtr conf ATTRIBUTE_UNUSED,
                                   struct qemud_server *server ATTRIBUTE_UNUSED,
                                   const char *filename ATTRIBUTE_UNUSED)
{
    return 0;
}
#endif

2676 2677 2678 2679 2680 2681
/*
 * Set up the logging environment
 * By default if daemonized all errors go to syslog and the logging
 * is also saved onto the logfile libvird.log, but if verbose or error
 * debugging is asked for then output informations or debug.
 */
2682
static int
2683 2684 2685
qemudSetLogging(virConfPtr conf, const char *filename)
{
    int log_level = 0;
2686 2687
    char *log_filters = NULL;
    char *log_outputs = NULL;
2688 2689
    int ret = -1;

2690 2691
    virLogReset();

2692
    /*
2693 2694 2695 2696 2697 2698 2699 2700 2701 2702 2703 2704
     * Libvirtd's order of precedence is:
     * cmdline > environment > config
     *
     * In order to achieve this, we must process configuration in
     * different order for the log level versus the filters and
     * outputs. Because filters and outputs append, we have to look at
     * the environment first and then only check the config file if
     * there was no result from the environment. The default output is
     * then applied only if there was no setting from either of the
     * first two. Because we don't have a way to determine if the log
     * level has been set, we must process variables in the opposite
     * order, each one overriding the previous.
2705
     */
2706
    GET_CONF_INT (conf, filename, log_level);
2707 2708 2709
    if (log_level != 0)
        virLogSetDefaultPriority(log_level);

2710
    virLogSetFromEnv();
2711 2712 2713 2714

    if (virLogGetNbFilters() == 0) {
        GET_CONF_STR (conf, filename, log_filters);
        virLogParseFilters(log_filters);
2715
    }
2716

2717 2718 2719
    if (virLogGetNbOutputs() == 0) {
        GET_CONF_STR (conf, filename, log_outputs);
        virLogParseOutputs(log_outputs);
2720
    }
2721 2722

    /*
2723 2724
     * If no defined outputs, then direct to syslog when running
     * as daemon. Otherwise the default output is stderr.
2725
     */
2726
    if (virLogGetNbOutputs() == 0) {
2727
        char *tmp = NULL;
2728
        if (godaemon) {
2729 2730
            if (virAsprintf (&tmp, "%d:syslog:libvirtd",
                             virLogGetDefaultPriority()) < 0)
2731 2732
                goto free_and_fail;
        } else {
2733 2734
            if (virAsprintf (&tmp, "%d:stderr",
                             virLogGetDefaultPriority()) < 0)
2735
                goto free_and_fail;
2736
        }
2737 2738
        virLogParseOutputs(tmp);
        VIR_FREE(tmp);
2739
    }
2740 2741 2742 2743 2744 2745 2746

    /*
     * Command line override for --verbose
     */
    if ((verbose) && (virLogGetDefaultPriority() > VIR_LOG_INFO))
        virLogSetDefaultPriority(VIR_LOG_INFO);

2747 2748
    ret = 0;

2749 2750 2751
free_and_fail:
    VIR_FREE(log_filters);
    VIR_FREE(log_outputs);
2752
    return(ret);
2753
}
2754

2755 2756 2757 2758
/* Read the config file if it exists.
 * Only used in the remote case, hence the name.
 */
static int
2759
remoteReadConfigFile (struct qemud_server *server, const char *filename)
2760 2761 2762
{
    virConfPtr conf;

2763 2764 2765 2766 2767
    /* The following variable names must match the corresponding
       configuration strings.  */
    char *unix_sock_ro_perms = NULL;
    char *unix_sock_rw_perms = NULL;
    char *unix_sock_group = NULL;
2768
    char *buf = NULL;
2769
    char *host_uuid = NULL;
2770

2771 2772
#if HAVE_POLKIT
    /* Change the default back to no auth for non-root */
2773
    if (!server->privileged && auth_unix_rw == REMOTE_AUTH_POLKIT)
2774
        auth_unix_rw = REMOTE_AUTH_NONE;
2775
    if (!server->privileged && auth_unix_ro == REMOTE_AUTH_POLKIT)
2776 2777 2778
        auth_unix_ro = REMOTE_AUTH_NONE;
#endif

2779
    conf = virConfReadFile (filename, 0);
2780
    if (!conf) return -1;
2781

2782 2783 2784
    /*
     * First get all the logging settings and activate them
     */
2785 2786
    if (qemudSetLogging(conf, filename) < 0)
        goto free_and_fail;
2787

2788 2789
    GET_CONF_INT (conf, filename, listen_tcp);
    GET_CONF_INT (conf, filename, listen_tls);
2790 2791
    GET_CONF_STR (conf, filename, tls_port);
    GET_CONF_STR (conf, filename, tcp_port);
2792
    GET_CONF_STR (conf, filename, listen_addr);
J
Jim Meyering 已提交
2793

2794
    if (remoteConfigGetAuth(conf, "auth_unix_rw", &auth_unix_rw, filename) < 0)
J
Jim Meyering 已提交
2795
        goto free_and_fail;
2796 2797 2798 2799 2800 2801 2802
#if HAVE_POLKIT
    /* Change default perms to be wide-open if PolicyKit is enabled.
     * Admin can always override in config file
     */
    if (auth_unix_rw == REMOTE_AUTH_POLKIT)
        unix_sock_rw_mask = 0777;
#endif
2803
    if (remoteConfigGetAuth(conf, "auth_unix_ro", &auth_unix_ro, filename) < 0)
J
Jim Meyering 已提交
2804
        goto free_and_fail;
2805
    if (remoteConfigGetAuth(conf, "auth_tcp", &auth_tcp, filename) < 0)
J
Jim Meyering 已提交
2806
        goto free_and_fail;
2807
    if (remoteConfigGetAuth(conf, "auth_tls", &auth_tls, filename) < 0)
J
Jim Meyering 已提交
2808
        goto free_and_fail;
2809

2810 2811
    GET_CONF_STR (conf, filename, unix_sock_group);
    if (unix_sock_group) {
2812
        if (!server->privileged) {
2813
            VIR_WARN0("Cannot set group when not running as root");
2814
        } else {
2815
            int ret;
2816
            struct group grpdata, *grp;
2817 2818 2819 2820 2821 2822
            size_t maxbuf = sysconf(_SC_GETGR_R_SIZE_MAX);

            if (maxbuf == -1)
                maxbuf = 1024;

            if (VIR_ALLOC_N(buf, maxbuf) < 0) {
2823
                VIR_ERROR0(_("Failed to allocate memory for buffer"));
2824 2825 2826 2827 2828 2829 2830 2831
                goto free_and_fail;
            }

            while ((ret = getgrnam_r(unix_sock_group, &grpdata,
                                     buf, maxbuf,
                                     &grp)) == ERANGE) {
                    maxbuf *= 2;
                    if (maxbuf > 65536 || VIR_REALLOC_N(buf, maxbuf) < 0) {
2832
                        VIR_ERROR0(_("Failed to reallocate enough memory for buffer"));
2833 2834 2835 2836 2837
                        goto free_and_fail;
                    }
            }

            if (ret != 0 || !grp) {
2838
                VIR_ERROR(_("Failed to lookup group '%s'"), unix_sock_group);
2839
                goto free_and_fail;
2840 2841
            }
            unix_sock_gid = grp->gr_gid;
2842
            VIR_FREE(buf);
2843
        }
2844
        VIR_FREE(unix_sock_group);
2845 2846
    }

2847 2848
    GET_CONF_STR (conf, filename, unix_sock_ro_perms);
    if (unix_sock_ro_perms) {
2849
        if (virStrToLong_i (unix_sock_ro_perms, NULL, 8, &unix_sock_ro_mask) != 0) {
2850
            VIR_ERROR(_("Failed to parse mode '%s'"), unix_sock_ro_perms);
2851
            goto free_and_fail;
2852
        }
2853
        VIR_FREE(unix_sock_ro_perms);
2854 2855
    }

2856 2857
    GET_CONF_STR (conf, filename, unix_sock_rw_perms);
    if (unix_sock_rw_perms) {
2858
        if (virStrToLong_i (unix_sock_rw_perms, NULL, 8, &unix_sock_rw_mask) != 0) {
2859
            VIR_ERROR(_("Failed to parse mode '%s'"), unix_sock_rw_perms);
2860
            goto free_and_fail;
2861
        }
2862
        VIR_FREE(unix_sock_rw_perms);
2863 2864
    }

2865 2866
    GET_CONF_STR (conf, filename, unix_sock_dir);

2867 2868
    GET_CONF_INT (conf, filename, mdns_adv);
    GET_CONF_STR (conf, filename, mdns_name);
2869

2870
    GET_CONF_INT (conf, filename, tls_no_verify_certificate);
2871

2872 2873 2874 2875
    GET_CONF_STR (conf, filename, key_file);
    GET_CONF_STR (conf, filename, cert_file);
    GET_CONF_STR (conf, filename, ca_file);
    GET_CONF_STR (conf, filename, crl_file);
2876

2877 2878 2879
    if (remoteConfigGetStringList (conf, "tls_allowed_dn_list",
                                   &tls_allowed_dn_list, filename) < 0)
        goto free_and_fail;
2880

2881
    if (remoteReadSaslAllowedUsernameList (conf, server, filename) < 0)
2882
        goto free_and_fail;
2883

2884 2885 2886 2887 2888

    GET_CONF_INT (conf, filename, min_workers);
    GET_CONF_INT (conf, filename, max_workers);
    GET_CONF_INT (conf, filename, max_clients);

2889 2890 2891
    GET_CONF_INT (conf, filename, max_requests);
    GET_CONF_INT (conf, filename, max_client_requests);

2892 2893 2894
    GET_CONF_INT (conf, filename, audit_level);
    GET_CONF_INT (conf, filename, audit_logging);

2895
    GET_CONF_STR (conf, filename, host_uuid);
2896 2897
    if (virSetHostUUIDStr(host_uuid)) {
        VIR_ERROR(_("invalid host UUID: %s"), host_uuid);
2898
        goto free_and_fail;
2899
    }
2900 2901 2902

    VIR_FREE(host_uuid);

2903 2904
    virConfFree (conf);
    return 0;
2905

2906 2907
 free_and_fail:
    virConfFree (conf);
2908
    VIR_FREE(host_uuid);
2909 2910 2911 2912 2913
    VIR_FREE(mdns_name);
    VIR_FREE(unix_sock_ro_perms);
    VIR_FREE(unix_sock_rw_perms);
    VIR_FREE(unix_sock_group);
    VIR_FREE(buf);
2914

2915 2916 2917 2918
    /* Don't bother trying to free listen_addr, tcp_port, tls_port, key_file,
       cert_file, ca_file, or crl_file, since they are initialized to
       non-malloc'd strings.  Besides, these are static variables, and callers
       are unlikely to call this function more than once, so there wouldn't
2919 2920 2921 2922 2923
       even be a real leak.  */

    if (tls_allowed_dn_list) {
        int i;
        for (i = 0; tls_allowed_dn_list[i]; i++)
2924 2925
            VIR_FREE(tls_allowed_dn_list[i]);
        VIR_FREE(tls_allowed_dn_list);
2926 2927 2928
    }

    return -1;
2929 2930
}

2931 2932
/* Display version information. */
static void
2933
version (void)
2934 2935 2936 2937
{
    printf ("%s (%s) %s\n", argv0, PACKAGE_NAME, PACKAGE_VERSION);
}

J
John Levon 已提交
2938 2939 2940 2941 2942 2943 2944 2945
#ifdef __sun
static int
qemudSetupPrivs (void)
{
    chown ("/var/run/libvirt", SYSTEM_UID, SYSTEM_UID);

    if (__init_daemon_priv (PU_RESETGROUPS | PU_CLEARLIMITSET,
        SYSTEM_UID, SYSTEM_UID, PRIV_XVM_CONTROL, NULL)) {
2946
        VIR_ERROR0(_("additional privileges are required"));
J
John Levon 已提交
2947 2948 2949 2950 2951
        return -1;
    }

    if (priv_set (PRIV_OFF, PRIV_ALLSETS, PRIV_FILE_LINK_ANY, PRIV_PROC_INFO,
        PRIV_PROC_SESSION, PRIV_PROC_EXEC, PRIV_PROC_FORK, NULL)) {
2952
        VIR_ERROR0(_("failed to set reduced privileges"));
J
John Levon 已提交
2953 2954 2955 2956 2957 2958
        return -1;
    }

    return 0;
}
#else
2959
# define qemudSetupPrivs() 0
J
John Levon 已提交
2960 2961
#endif

2962 2963 2964 2965 2966 2967 2968 2969 2970 2971 2972 2973 2974 2975 2976 2977 2978 2979 2980 2981 2982 2983 2984 2985 2986 2987 2988 2989 2990 2991 2992 2993 2994 2995 2996 2997 2998 2999 3000 3001 3002 3003 3004 3005 3006 3007 3008 3009 3010 3011 3012 3013 3014 3015 3016

/*
 * Doing anything non-trivial in signal handlers is pretty dangerous,
 * since there are very few async-signal safe POSIX funtions. To
 * deal with this we setup a very simple signal handler. It simply
 * writes the signal number to a pipe. The main event loop then sees
 * the signal on the pipe and can safely do the processing from
 * event loop context
 */
static int
daemonSetupSignals(struct qemud_server *server)
{
    struct sigaction sig_action;
    int sigpipe[2];

    if (pipe(sigpipe) < 0)
        return -1;

    if (virSetNonBlock(sigpipe[0]) < 0 ||
        virSetNonBlock(sigpipe[1]) < 0 ||
        virSetCloseExec(sigpipe[0]) < 0 ||
        virSetCloseExec(sigpipe[1]) < 0) {
        char ebuf[1024];
        VIR_ERROR(_("Failed to create pipe: %s"),
                  virStrerror(errno, ebuf, sizeof ebuf));
        goto error;
    }

    sig_action.sa_sigaction = sig_handler;
    sig_action.sa_flags = SA_SIGINFO;
    sigemptyset(&sig_action.sa_mask);

    sigaction(SIGHUP, &sig_action, NULL);
    sigaction(SIGINT, &sig_action, NULL);
    sigaction(SIGQUIT, &sig_action, NULL);
    sigaction(SIGTERM, &sig_action, NULL);

    sig_action.sa_handler = SIG_IGN;
    sigaction(SIGPIPE, &sig_action, NULL);

    if (virEventAddHandleImpl(sigpipe[0],
                              VIR_EVENT_HANDLE_READABLE,
                              qemudDispatchSignalEvent,
                              server, NULL) < 0) {
        VIR_ERROR0(_("Failed to register callback for signal pipe"));
        goto error;
    }

    server->sigread = sigpipe[0];
    server->sigwrite = sigpipe[1];
    sigwrite = sigpipe[1];

    return 0;

error:
3017 3018
    VIR_FORCE_CLOSE(sigpipe[0]);
    VIR_FORCE_CLOSE(sigpipe[1]);
3019 3020 3021
    return -1;
}

3022 3023
/* Print command-line usage. */
static void
3024
usage (void)
3025 3026
{
    fprintf (stderr,
3027
             _("\n\
3028 3029 3030 3031 3032 3033
Usage:\n\
  %s [options]\n\
\n\
Options:\n\
  -v | --verbose         Verbose messages.\n\
  -d | --daemon          Run as a daemon & write PID file.\n\
3034 3035
  -l | --listen          Listen for TCP/IP connections.\n\
  -t | --timeout <secs>  Exit after timeout period.\n\
3036
  -f | --config <file>   Configuration file.\n\
3037
     | --version         Display version information.\n\
3038 3039
  -p | --pid-file <file> Change name of PID file.\n\
\n\
3040
libvirt management daemon:\n\
3041 3042 3043
\n\
  Default paths:\n\
\n\
W
Wolfgang Mauerer 已提交
3044
    Configuration file (unless overridden by -f):\n\
3045
      %s/libvirt/libvirtd.conf\n\
3046
\n\
3047
    Sockets (as root):\n\
3048 3049
      %s/run/libvirt/libvirt-sock\n\
      %s/run/libvirt/libvirt-sock-ro\n\
3050 3051 3052
\n\
    Sockets (as non-root):\n\
      $HOME/.libvirt/libvirt-sock (in UNIX abstract namespace)\n\
3053 3054
\n\
    TLS:\n\
3055 3056 3057
      CA certificate:     %s\n\
      Server certificate: %s\n\
      Server private key: %s\n\
3058 3059 3060
\n\
    PID file (unless overridden by --pid-file):\n\
      %s\n\
3061 3062
\n"),
               argv0,
3063 3064 3065
               SYSCONFDIR,
               LOCALSTATEDIR,
               LOCALSTATEDIR,
3066 3067 3068 3069 3070 3071
               LIBVIRT_CACERT,
               LIBVIRT_SERVERCERT,
               LIBVIRT_SERVERKEY,
               (REMOTE_PID_FILE[0] != '\0'
                ? REMOTE_PID_FILE
                : _("(disabled in ./configure)")));
3072 3073
}

3074 3075 3076 3077
enum {
    OPT_VERSION = 129
};

D
Daniel P. Berrange 已提交
3078 3079
#define MAX_LISTEN 5
int main(int argc, char **argv) {
3080
    struct qemud_server *server = NULL;
3081
    const char *pid_file = NULL;
3082
    const char *remote_config_file = NULL;
3083
    int statuswrite = -1;
3084
    int ret = 1;
3085
    argv0 = argv[0];
D
Daniel P. Berrange 已提交
3086 3087 3088 3089

    struct option opts[] = {
        { "verbose", no_argument, &verbose, 1},
        { "daemon", no_argument, &godaemon, 1},
3090
        { "listen", no_argument, &ipsock, 1},
3091
        { "config", required_argument, NULL, 'f'},
3092 3093
        { "timeout", required_argument, NULL, 't'},
        { "pid-file", required_argument, NULL, 'p'},
3094
        { "version", no_argument, NULL, OPT_VERSION },
3095
        { "help", no_argument, NULL, '?' },
D
Daniel P. Berrange 已提交
3096 3097 3098
        {0, 0, 0, 0}
    };

E
Eric Blake 已提交
3099 3100 3101 3102 3103 3104
    if (setlocale (LC_ALL, "") == NULL ||
        bindtextdomain (PACKAGE, LOCALEDIR) == NULL ||
        textdomain(PACKAGE) == NULL ||
        virInitialize() < 0) {
        fprintf(stderr, _("%s: initialization failed\n"), argv0);
        exit(EXIT_FAILURE);
3105
    }
D
Daniel P. Berrange 已提交
3106

3107 3108 3109 3110
    /* Set error logging priority to debug, so client errors don't
     * show up as errors in the daemon log */
    virErrorSetLogPriority(VIR_LOG_DEBUG);

D
Daniel P. Berrange 已提交
3111 3112 3113 3114 3115
    while (1) {
        int optidx = 0;
        int c;
        char *tmp;

3116
        c = getopt_long(argc, argv, "ldf:p:t:v", opts, &optidx);
D
Daniel P. Berrange 已提交
3117 3118 3119 3120 3121 3122 3123 3124 3125 3126 3127 3128 3129 3130 3131

        if (c == -1) {
            break;
        }

        switch (c) {
        case 0:
            /* Got one of the flags */
            break;
        case 'v':
            verbose = 1;
            break;
        case 'd':
            godaemon = 1;
            break;
3132 3133
        case 'l':
            ipsock = 1;
D
Daniel P. Berrange 已提交
3134 3135 3136
            break;

        case 't':
3137
            if (virStrToLong_i(optarg, &tmp, 10, &timeout) != 0
3138 3139 3140
                || timeout <= 0
                /* Ensure that we can multiply by 1000 without overflowing.  */
                || timeout > INT_MAX / 1000)
D
Daniel P. Berrange 已提交
3141 3142
                timeout = -1;
            break;
3143 3144

        case 'p':
3145 3146 3147 3148 3149
            pid_file = optarg;
            break;

        case 'f':
            remote_config_file = optarg;
3150 3151
            break;

3152
        case OPT_VERSION:
3153
            version ();
3154 3155
            return 0;

D
Daniel P. Berrange 已提交
3156
        case '?':
3157
            usage ();
D
Daniel P. Berrange 已提交
3158 3159 3160
            return 2;

        default:
3161 3162
            fprintf (stderr, _("%s: internal error: unknown flag: %c\n"),
                     argv0, c);
3163
            exit (EXIT_FAILURE);
D
Daniel P. Berrange 已提交
3164 3165 3166
        }
    }

3167 3168
    if (remote_config_file == NULL) {
        static const char *default_config_file
3169
            = SYSCONFDIR "/libvirt/libvirtd.conf";
3170
        remote_config_file =
3171
            (access(default_config_file, R_OK) == 0
3172 3173 3174 3175
             ? default_config_file
             : "/dev/null");
    }

3176
    if (godaemon) {
3177
        char ebuf[1024];
3178 3179 3180 3181 3182 3183 3184

        if (chdir("/") < 0) {
            VIR_ERROR(_("cannot change to root directory: %s"),
                      virStrerror(errno, ebuf, sizeof(ebuf)));
            goto error;
        }

3185
        if ((statuswrite = daemonForkIntoBackground()) < 0) {
3186 3187
            VIR_ERROR(_("Failed to fork as daemon: %s"),
                      virStrerror(errno, ebuf, sizeof ebuf));
3188
            goto error;
3189 3190 3191 3192 3193
        }
    }

    /* If running as root and no PID file is set, use the default */
    if (pid_file == NULL &&
3194
        geteuid() == 0 &&
3195 3196 3197 3198 3199
        REMOTE_PID_FILE[0] != '\0')
        pid_file = REMOTE_PID_FILE;

    /* If we have a pidfile set, claim it now, exiting if already taken */
    if (pid_file != NULL &&
3200 3201 3202
        qemudWritePidFile (pid_file) < 0) {
        pid_file = NULL; /* Prevent unlinking of someone else's pid ! */
        ret = VIR_DAEMON_ERR_PIDFILE;
3203
        goto error;
3204
    }
3205

J
John Levon 已提交
3206
    /* Ensure the rundir exists (on tmpfs on some systems) */
3207
    if (geteuid() == 0) {
3208
        const char *rundir = LOCALSTATEDIR "/run/libvirt";
J
John Levon 已提交
3209 3210 3211

        if (mkdir (rundir, 0755)) {
            if (errno != EEXIST) {
3212 3213 3214
                char ebuf[1024];
                VIR_ERROR(_("unable to create rundir %s: %s"), rundir,
                          virStrerror(errno, ebuf, sizeof(ebuf)));
3215 3216
                ret = VIR_DAEMON_ERR_RUNDIR;
                goto error;
J
John Levon 已提交
3217 3218 3219 3220
            }
        }
    }

3221 3222 3223 3224 3225 3226
    /* Beyond this point, nothing should rely on using
     * getuid/geteuid() == 0, for privilege level checks.
     * It must all use the flag 'server->privileged'
     * which is also passed into all libvirt stateful
     * drivers
     */
3227 3228
    if (qemudSetupPrivs() < 0) {
        ret = VIR_DAEMON_ERR_PRIVS;
3229
        goto error;
3230
    }
J
John Levon 已提交
3231

3232
    if (!(server = qemudInitialize())) {
3233
        ret = VIR_DAEMON_ERR_INIT;
3234
        goto error;
3235
    }
3236

3237 3238
    if ((daemonSetupSignals(server)) < 0) {
        ret = VIR_DAEMON_ERR_SIGNAL;
3239
        goto error;
3240
    }
3241

3242
    /* Read the config file (if it exists). */
3243 3244
    if (remoteReadConfigFile (server, remote_config_file) < 0) {
        ret = VIR_DAEMON_ERR_CONFIG;
3245
        goto error;
3246
    }
D
Daniel P. Berrange 已提交
3247

3248 3249 3250 3251 3252 3253 3254 3255 3256 3257
    if (audit_level) {
        if (virAuditOpen() < 0) {
            if (audit_level > 1) {
                ret = VIR_DAEMON_ERR_AUDIT;
                goto error;
            }
        }
    }
    virAuditLog(audit_logging);

3258
    /* setup the hooks if any */
3259
    if (virHookInitialize() < 0) {
3260 3261 3262 3263
        ret = VIR_DAEMON_ERR_HOOKS;
        goto error;
    }

3264 3265 3266
    /* Disable error func, now logging is setup */
    virSetErrorFunc(NULL, virshErrorHandler);

3267 3268 3269 3270 3271 3272 3273 3274
    /*
     * Call the daemon startup hook
     * TODO: should we abort the daemon startup if the script returned
     *       an error ?
     */
    virHookCall(VIR_HOOK_DRIVER_DAEMON, "-", VIR_HOOK_DAEMON_OP_START,
                0, "start", NULL);

3275
    if (qemudNetworkInit(server) < 0) {
3276
        ret = VIR_DAEMON_ERR_NETWORK;
3277
        goto error;
3278 3279
    }

3280 3281 3282 3283 3284 3285 3286 3287 3288
    /* Tell parent of daemon that basic initialization is complete
     * In particular we're ready to accept net connections & have
     * written the pidfile
     */
    if (statuswrite != -1) {
        char status = 0;
        while (write(statuswrite, &status, 1) == -1 &&
               errno == EINTR)
            ;
3289
        VIR_FORCE_CLOSE(statuswrite);
3290 3291
    }

3292 3293 3294
    /* Start the event loop in a background thread, since
     * state initialization needs events to be being processed */
    if (qemudStartEventLoop(server) < 0) {
3295
        VIR_ERROR0(_("Event thread startup failed"));
3296 3297 3298
        goto error;
    }

3299 3300 3301 3302 3303
    /* Start the stateful HV drivers
     * This is delibrately done after telling the parent process
     * we're ready, since it can take a long time and this will
     * seriously delay OS bootup process */
    if (virStateInitialize(server->privileged) < 0) {
3304
        VIR_ERROR0(_("Driver state initialization failed"));
3305
        goto shutdown;
3306
    }
D
Daniel P. Berrange 已提交
3307

3308 3309 3310
    /* Start accepting new clients from network */
    virMutexLock(&server->lock);
    if (qemudNetworkEnable(server) < 0) {
3311
        VIR_ERROR0(_("Network event loop enablement failed"));
3312 3313 3314 3315
        goto shutdown;
    }
    virMutexUnlock(&server->lock);

3316 3317
    ret = 0;

3318 3319 3320 3321 3322 3323 3324 3325 3326 3327 3328 3329 3330 3331 3332 3333
shutdown:
    /* In a non-0 shutdown scenario we need to tell event loop
     * to quit immediately. Otherwise in normal case we just
     * sit in the thread join forever. Sure this means the
     * main thread doesn't do anything useful ever, but that's
     * not too much of drain on resources
     */
    if (ret != 0) {
        virMutexLock(&server->lock);
        if (server->hasEventThread)
            /* This SIGQUIT triggers the shutdown process */
            kill(getpid(), SIGQUIT);
        virMutexUnlock(&server->lock);
    }
    pthread_join(server->eventThread, NULL);

3334 3335 3336
    virHookCall(VIR_HOOK_DRIVER_DAEMON, "-", VIR_HOOK_DAEMON_OP_SHUTDOWN,
                0, "shutdown", NULL);

3337
error:
3338 3339 3340 3341 3342 3343 3344 3345
    if (statuswrite != -1) {
        if (ret != 0) {
            /* Tell parent of daemon what failed */
            char status = ret;
            while (write(statuswrite, &status, 1) == -1 &&
                   errno == EINTR)
                ;
        }
3346
        VIR_FORCE_CLOSE(statuswrite);
3347
    }
3348 3349
    if (server)
        qemudCleanup(server);
3350 3351
    if (pid_file)
        unlink (pid_file);
3352
    virLogShutdown();
3353
    return ret;
D
Daniel P. Berrange 已提交
3354
}