net/vhost-user.c

   1 /*
   2  * vhost-user.c
   3  *
   4  * Copyright (c) 2013 Virtual Open Systems Sarl.
   5  *
   6  * This work is licensed under the terms of the GNU GPL, version 2 or later.
   7  * See the COPYING file in the top-level directory.
   8  *
   9  */
  10
  11 #include "qemu/osdep.h"
  12 #include "clients.h"
  13 #include "net/vhost_net.h"
  14 #include "net/vhost-user.h"
  15 #include "hw/virtio/vhost-user.h"
  16 #include "chardev/char-fe.h"
  17 #include "qapi/error.h"
  18 #include "qapi/qapi-commands-net.h"
  19 #include "qemu/config-file.h"
  20 #include "qemu/error-report.h"
  21 #include "qemu/option.h"
  22 #include "trace.h"
  23
  24 typedef struct NetVhostUserState {
  25     NetClientState nc;
  26     CharBackend chr; /* only queue index 0 */
  27     VhostUserState *vhost_user;
  28     VHostNetState *vhost_net;
  29     guint watch;
  30     uint64_t acked_features;
  31     bool started;
  32 } NetVhostUserState;
  33
  34 VHostNetState *vhost_user_get_vhost_net(NetClientState *nc)
  35 {
  36     NetVhostUserState *s = DO_UPCAST(NetVhostUserState, nc, nc);
  37     assert(nc->info->type == NET_CLIENT_DRIVER_VHOST_USER);
  38     return s->vhost_net;
  39 }
  40
  41 uint64_t vhost_user_get_acked_features(NetClientState *nc)
  42 {
  43     NetVhostUserState *s = DO_UPCAST(NetVhostUserState, nc, nc);
  44     assert(nc->info->type == NET_CLIENT_DRIVER_VHOST_USER);
  45     return s->acked_features;
  46 }
  47
  48 static void vhost_user_stop(int queues, NetClientState *ncs[])
  49 {
  50     NetVhostUserState *s;
  51     int i;
  52
  53     for (i = 0; i < queues; i++) {
  54         assert(ncs[i]->info->type == NET_CLIENT_DRIVER_VHOST_USER);
  55
  56         s = DO_UPCAST(NetVhostUserState, nc, ncs[i]);
  57
  58         if (s->vhost_net) {
  59             /* save acked features */
  60             uint64_t features = vhost_net_get_acked_features(s->vhost_net);
  61             if (features) {
  62                 s->acked_features = features;
  63             }
  64             vhost_net_cleanup(s->vhost_net);
  65         }
  66     }
  67 }
  68
  69 static int vhost_user_start(int queues, NetClientState *ncs[],
  70                             VhostUserState *be)
  71 {
  72     VhostNetOptions options;
  73     struct vhost_net *net = NULL;
  74     NetVhostUserState *s;
  75     int max_queues;
  76     int i;
  77
  78     options.backend_type = VHOST_BACKEND_TYPE_USER;
  79
  80     for (i = 0; i < queues; i++) {
  81         assert(ncs[i]->info->type == NET_CLIENT_DRIVER_VHOST_USER);
  82
  83         s = DO_UPCAST(NetVhostUserState, nc, ncs[i]);
  84
  85         options.net_backend = ncs[i];
  86         options.opaque      = be;
  87         options.busyloop_timeout = 0;
  88         net = vhost_net_init(&options);
  89         if (!net) {
  90             error_report("failed to init vhost_net for queue %d", i);
  91             goto err;
  92         }
  93
  94         if (i == 0) {
  95             max_queues = vhost_net_get_max_queues(net);
  96             if (queues > max_queues) {
  97                 error_report("you are asking more queues than supported: %d",
  98                              max_queues);
  99                 goto err;
 100             }
 101         }
 102
 103         if (s->vhost_net) {
 104             vhost_net_cleanup(s->vhost_net);
 105             g_free(s->vhost_net);
 106         }
 107         s->vhost_net = net;
 108     }
 109
 110     return 0;
 111
 112 err:
 113     if (net) {
 114         vhost_net_cleanup(net);
 115         g_free(net);
 116     }
 117     vhost_user_stop(i, ncs);
 118     return -1;
 119 }
 120
 121 static ssize_t vhost_user_receive(NetClientState *nc, const uint8_t *buf,
 122                                   size_t size)
 123 {
 124     /* In case of RARP (message size is 60) notify backup to send a fake RARP.
 125        This fake RARP will be sent by backend only for guest
 126        without GUEST_ANNOUNCE capability.
 127      */
 128     if (size == 60) {
 129         NetVhostUserState *s = DO_UPCAST(NetVhostUserState, nc, nc);
 130         int r;
 131         static int display_rarp_failure = 1;
 132         char mac_addr[6];
 133
 134         /* extract guest mac address from the RARP message */
 135         memcpy(mac_addr, &buf[6], 6);
 136
 137         r = vhost_net_notify_migration_done(s->vhost_net, mac_addr);
 138
 139         if ((r != 0) && (display_rarp_failure)) {
 140             fprintf(stderr,
 141                     "Vhost user backend fails to broadcast fake RARP\n");
 142             fflush(stderr);
 143             display_rarp_failure = 0;
 144         }
 145     }
 146
 147     return size;
 148 }
 149
 150 static void net_vhost_user_cleanup(NetClientState *nc)
 151 {
 152     NetVhostUserState *s = DO_UPCAST(NetVhostUserState, nc, nc);
 153
 154     if (s->vhost_net) {
 155         vhost_net_cleanup(s->vhost_net);
 156         g_free(s->vhost_net);
 157         s->vhost_net = NULL;
 158     }
 159     if (nc->queue_index == 0) {
 160         if (s->watch) {
 161             g_source_remove(s->watch);
 162             s->watch = 0;
 163         }
 164         qemu_chr_fe_deinit(&s->chr, true);
 165         if (s->vhost_user) {
 166             vhost_user_cleanup(s->vhost_user);
 167             g_free(s->vhost_user);
 168             s->vhost_user = NULL;
 169         }
 170     }
 171
 172     qemu_purge_queued_packets(nc);
 173 }
 174
 175 static bool vhost_user_has_vnet_hdr(NetClientState *nc)
 176 {
 177     assert(nc->info->type == NET_CLIENT_DRIVER_VHOST_USER);
 178
 179     return true;
 180 }
 181
 182 static bool vhost_user_has_ufo(NetClientState *nc)
 183 {
 184     assert(nc->info->type == NET_CLIENT_DRIVER_VHOST_USER);
 185
 186     return true;
 187 }
 188
 189 static NetClientInfo net_vhost_user_info = {
 190         .type = NET_CLIENT_DRIVER_VHOST_USER,
 191         .size = sizeof(NetVhostUserState),
 192         .receive = vhost_user_receive,
 193         .cleanup = net_vhost_user_cleanup,
 194         .has_vnet_hdr = vhost_user_has_vnet_hdr,
 195         .has_ufo = vhost_user_has_ufo,
 196 };
 197
 198 static gboolean net_vhost_user_watch(GIOChannel *chan, GIOCondition cond,
 199                                            void *opaque)
 200 {
 201     NetVhostUserState *s = opaque;
 202
 203     qemu_chr_fe_disconnect(&s->chr);
 204
 205     return TRUE;
 206 }
 207
 208 static void net_vhost_user_event(void *opaque, int event);
 209
 210 static void chr_closed_bh(void *opaque)
 211 {
 212     const char *name = opaque;
 213     NetClientState *ncs[MAX_QUEUE_NUM];
 214     NetVhostUserState *s;
 215     Error *err = NULL;
 216     int queues;
 217
 218     queues = qemu_find_net_clients_except(name, ncs,
 219                                           NET_CLIENT_DRIVER_NIC,
 220                                           MAX_QUEUE_NUM);
 221     assert(queues < MAX_QUEUE_NUM);
 222
 223     s = DO_UPCAST(NetVhostUserState, nc, ncs[0]);
 224
 225     qmp_set_link(name, false, &err);
 226     vhost_user_stop(queues, ncs);
 227
 228     qemu_chr_fe_set_handlers(&s->chr, NULL, NULL, net_vhost_user_event,
 229                              NULL, opaque, NULL, true);
 230
 231     if (err) {
 232         error_report_err(err);
 233     }
 234 }
 235
 236 static void net_vhost_user_event(void *opaque, int event)
 237 {
 238     const char *name = opaque;
 239     NetClientState *ncs[MAX_QUEUE_NUM];
 240     NetVhostUserState *s;
 241     Chardev *chr;
 242     Error *err = NULL;
 243     int queues;
 244
 245     queues = qemu_find_net_clients_except(name, ncs,
 246                                           NET_CLIENT_DRIVER_NIC,
 247                                           MAX_QUEUE_NUM);
 248     assert(queues < MAX_QUEUE_NUM);
 249
 250     s = DO_UPCAST(NetVhostUserState, nc, ncs[0]);
 251     chr = qemu_chr_fe_get_driver(&s->chr);
 252     trace_vhost_user_event(chr->label, event);
 253     switch (event) {
 254     case CHR_EVENT_OPENED:
 255         if (vhost_user_start(queues, ncs, s->vhost_user) < 0) {
 256             qemu_chr_fe_disconnect(&s->chr);
 257             return;
 258         }
 259         s->watch = qemu_chr_fe_add_watch(&s->chr, G_IO_HUP,
 260                                          net_vhost_user_watch, s);
 261         qmp_set_link(name, true, &err);
 262         s->started = true;
 263         break;
 264     case CHR_EVENT_CLOSED:
 265         /* a close event may happen during a read/write, but vhost
 266          * code assumes the vhost_dev remains setup, so delay the
 267          * stop & clear to idle.
 268          * FIXME: better handle failure in vhost code, remove bh
 269          */
 270         if (s->watch) {
 271             AioContext *ctx = qemu_get_current_aio_context();
 272
 273             g_source_remove(s->watch);
 274             s->watch = 0;
 275             qemu_chr_fe_set_handlers(&s->chr, NULL, NULL, NULL, NULL,
 276                                      NULL, NULL, false);
 277
 278             aio_bh_schedule_oneshot(ctx, chr_closed_bh, opaque);
 279         }
 280         break;
 281     }
 282
 283     if (err) {
 284         error_report_err(err);
 285     }
 286 }
 287
 288 static int net_vhost_user_init(NetClientState *peer, const char *device,
 289                                const char *name, Chardev *chr,
 290                                int queues)
 291 {
 292     Error *err = NULL;
 293     NetClientState *nc, *nc0 = NULL;
 294     VhostUserState *user = NULL;
 295     NetVhostUserState *s = NULL;
 296     int i;
 297
 298     assert(name);
 299     assert(queues > 0);
 300
 301     user = vhost_user_init();
 302     if (!user) {
 303         error_report("failed to init vhost_user");
 304         goto err;
 305     }
 306
 307     for (i = 0; i < queues; i++) {
 308         nc = qemu_new_net_client(&net_vhost_user_info, peer, device, name);
 309         snprintf(nc->info_str, sizeof(nc->info_str), "vhost-user%d to %s",
 310                  i, chr->label);
 311         nc->queue_index = i;
 312         if (!nc0) {
 313             nc0 = nc;
 314             s = DO_UPCAST(NetVhostUserState, nc, nc);
 315             if (!qemu_chr_fe_init(&s->chr, chr, &err)) {
 316                 error_report_err(err);
 317                 goto err;
 318             }
 319             user->chr = &s->chr;
 320         }
 321         s = DO_UPCAST(NetVhostUserState, nc, nc);
 322         s->vhost_user = user;
 323     }
 324
 325     s = DO_UPCAST(NetVhostUserState, nc, nc0);
 326     do {
 327         if (qemu_chr_fe_wait_connected(&s->chr, &err) < 0) {
 328             error_report_err(err);
 329             goto err;
 330         }
 331         qemu_chr_fe_set_handlers(&s->chr, NULL, NULL,
 332                                  net_vhost_user_event, NULL, nc0->name, NULL,
 333                                  true);
 334     } while (!s->started);
 335
 336     assert(s->vhost_net);
 337
 338     return 0;
 339
 340 err:
 341     if (user) {
 342         vhost_user_cleanup(user);
 343         g_free(user);
 344         if (s) {
 345             s->vhost_user = NULL;
 346         }
 347     }
 348     if (nc0) {
 349         qemu_del_net_client(nc0);
 350     }
 351
 352     return -1;
 353 }
 354
 355 static Chardev *net_vhost_claim_chardev(
 356     const NetdevVhostUserOptions *opts, Error **errp)
 357 {
 358     Chardev *chr = qemu_chr_find(opts->chardev);
 359
 360     if (chr == NULL) {
 361         error_setg(errp, "chardev \"%s\" not found", opts->chardev);
 362         return NULL;
 363     }
 364
 365     if (!qemu_chr_has_feature(chr, QEMU_CHAR_FEATURE_RECONNECTABLE)) {
 366         error_setg(errp, "chardev \"%s\" is not reconnectable",
 367                    opts->chardev);
 368         return NULL;
 369     }
 370     if (!qemu_chr_has_feature(chr, QEMU_CHAR_FEATURE_FD_PASS)) {
 371         error_setg(errp, "chardev \"%s\" does not support FD passing",
 372                    opts->chardev);
 373         return NULL;
 374     }
 375
 376     return chr;
 377 }
 378
 379 static int net_vhost_check_net(void *opaque, QemuOpts *opts, Error **errp)
 380 {
 381     const char *name = opaque;
 382     const char *driver, *netdev;
 383
 384     driver = qemu_opt_get(opts, "driver");
 385     netdev = qemu_opt_get(opts, "netdev");
 386
 387     if (!driver || !netdev) {
 388         return 0;
 389     }
 390
 391     if (strcmp(netdev, name) == 0 &&
 392         !g_str_has_prefix(driver, "virtio-net-")) {
 393         error_setg(errp, "vhost-user requires frontend driver virtio-net-*");
 394         return -1;
 395     }
 396
 397     return 0;
 398 }
 399
 400 int net_init_vhost_user(const Netdev *netdev, const char *name,
 401                         NetClientState *peer, Error **errp)
 402 {
 403     int queues;
 404     const NetdevVhostUserOptions *vhost_user_opts;
 405     Chardev *chr;
 406
 407     assert(netdev->type == NET_CLIENT_DRIVER_VHOST_USER);
 408     vhost_user_opts = &netdev->u.vhost_user;
 409
 410     chr = net_vhost_claim_chardev(vhost_user_opts, errp);
 411     if (!chr) {
 412         return -1;
 413     }
 414
 415     /* verify net frontend */
 416     if (qemu_opts_foreach(qemu_find_opts("device"), net_vhost_check_net,
 417                           (char *)name, errp)) {
 418         return -1;
 419     }
 420
 421     queues = vhost_user_opts->has_queues ? vhost_user_opts->queues : 1;
 422     if (queues < 1 || queues > MAX_QUEUE_NUM) {
 423         error_setg(errp,
 424                    "vhost-user number of queues must be in range [1, %d]",
 425                    MAX_QUEUE_NUM);
 426         return -1;
 427     }
 428
 429     return net_vhost_user_init(peer, "vhost_user", name, chr, queues);
 430 }