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