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