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 "sysemu/char.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     CharDriverState *chr;
  24     VHostNetState *vhost_net;
  25     guint watch;
  26     uint64_t acked_features;
  27 } VhostUserState;
  28
  29 typedef struct VhostUserChardevProps {
  30     bool is_socket;
  31     bool is_unix;
  32 } VhostUserChardevProps;
  33
  34 VHostNetState *vhost_user_get_vhost_net(NetClientState *nc)
  35 {
  36     VhostUserState *s = DO_UPCAST(VhostUserState, nc, nc);
  37     assert(nc->info->type == NET_CLIENT_OPTIONS_KIND_VHOST_USER);
  38     return s->vhost_net;
  39 }
  40
  41 uint64_t vhost_user_get_acked_features(NetClientState *nc)
  42 {
  43     VhostUserState *s = DO_UPCAST(VhostUserState, nc, nc);
  44     assert(nc->info->type == NET_CLIENT_OPTIONS_KIND_VHOST_USER);
  45     return s->acked_features;
  46 }
  47
  48 static int vhost_user_running(VhostUserState *s)
  49 {
  50     return (s->vhost_net) ? 1 : 0;
  51 }
  52
  53 static void vhost_user_stop(int queues, NetClientState *ncs[])
  54 {
  55     VhostUserState *s;
  56     int i;
  57
  58     for (i = 0; i < queues; i++) {
  59         assert (ncs[i]->info->type == NET_CLIENT_OPTIONS_KIND_VHOST_USER);
  60
  61         s = DO_UPCAST(VhostUserState, nc, ncs[i]);
  62         if (!vhost_user_running(s)) {
  63             continue;
  64         }
  65
  66         if (s->vhost_net) {
  67             /* save acked features */
  68             s->acked_features = vhost_net_get_acked_features(s->vhost_net);
  69             vhost_net_cleanup(s->vhost_net);
  70             s->vhost_net = NULL;
  71         }
  72     }
  73 }
  74
  75 static int vhost_user_start(int queues, NetClientState *ncs[])
  76 {
  77     VhostNetOptions options;
  78     VhostUserState *s;
  79     int max_queues;
  80     int i;
  81
  82     options.backend_type = VHOST_BACKEND_TYPE_USER;
  83
  84     for (i = 0; i < queues; i++) {
  85         assert (ncs[i]->info->type == NET_CLIENT_OPTIONS_KIND_VHOST_USER);
  86
  87         s = DO_UPCAST(VhostUserState, nc, ncs[i]);
  88         if (vhost_user_running(s)) {
  89             continue;
  90         }
  91
  92         options.net_backend = ncs[i];
  93         options.opaque      = s->chr;
  94         s->vhost_net = vhost_net_init(&options);
  95         if (!s->vhost_net) {
  96             error_report("failed to init vhost_net for queue %d", i);
  97             goto err;
  98         }
  99
 100         if (i == 0) {
 101             max_queues = vhost_net_get_max_queues(s->vhost_net);
 102             if (queues > max_queues) {
 103                 error_report("you are asking more queues than supported: %d",
 104                              max_queues);
 105                 goto err;
 106             }
 107         }
 108     }
 109
 110     return 0;
 111
 112 err:
 113     vhost_user_stop(i + 1, 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         s->vhost_net = NULL;
 153     }
 154     if (s->chr) {
 155         qemu_chr_add_handlers(s->chr, NULL, NULL, NULL, NULL);
 156         qemu_chr_fe_release(s->chr);
 157         s->chr = NULL;
 158     }
 159
 160     qemu_purge_queued_packets(nc);
 161 }
 162
 163 static bool vhost_user_has_vnet_hdr(NetClientState *nc)
 164 {
 165     assert(nc->info->type == NET_CLIENT_OPTIONS_KIND_VHOST_USER);
 166
 167     return true;
 168 }
 169
 170 static bool vhost_user_has_ufo(NetClientState *nc)
 171 {
 172     assert(nc->info->type == NET_CLIENT_OPTIONS_KIND_VHOST_USER);
 173
 174     return true;
 175 }
 176
 177 static NetClientInfo net_vhost_user_info = {
 178         .type = NET_CLIENT_OPTIONS_KIND_VHOST_USER,
 179         .size = sizeof(VhostUserState),
 180         .receive = vhost_user_receive,
 181         .cleanup = vhost_user_cleanup,
 182         .has_vnet_hdr = vhost_user_has_vnet_hdr,
 183         .has_ufo = vhost_user_has_ufo,
 184 };
 185
 186 static gboolean net_vhost_user_watch(GIOChannel *chan, GIOCondition cond,
 187                                            void *opaque)
 188 {
 189     VhostUserState *s = opaque;
 190     uint8_t buf[1];
 191
 192     /* We don't actually want to read anything, but CHR_EVENT_CLOSED will be
 193      * raised as a side-effect of the read.
 194      */
 195     qemu_chr_fe_read_all(s->chr, buf, sizeof(buf));
 196
 197     return FALSE;
 198 }
 199
 200 static void net_vhost_user_event(void *opaque, int event)
 201 {
 202     const char *name = opaque;
 203     NetClientState *ncs[MAX_QUEUE_NUM];
 204     VhostUserState *s;
 205     Error *err = NULL;
 206     int queues;
 207
 208     queues = qemu_find_net_clients_except(name, ncs,
 209                                           NET_CLIENT_OPTIONS_KIND_NIC,
 210                                           MAX_QUEUE_NUM);
 211     assert(queues < MAX_QUEUE_NUM);
 212
 213     s = DO_UPCAST(VhostUserState, nc, ncs[0]);
 214     trace_vhost_user_event(s->chr->label, event);
 215     switch (event) {
 216     case CHR_EVENT_OPENED:
 217         s->watch = qemu_chr_fe_add_watch(s->chr, G_IO_HUP,
 218                                          net_vhost_user_watch, s);
 219         if (vhost_user_start(queues, ncs) < 0) {
 220             qemu_chr_disconnect(s->chr);
 221             return;
 222         }
 223         qmp_set_link(name, true, &err);
 224         break;
 225     case CHR_EVENT_CLOSED:
 226         qmp_set_link(name, false, &err);
 227         vhost_user_stop(queues, ncs);
 228         g_source_remove(s->watch);
 229         s->watch = 0;
 230         break;
 231     }
 232
 233     if (err) {
 234         error_report_err(err);
 235     }
 236 }
 237
 238 static int net_vhost_user_init(NetClientState *peer, const char *device,
 239                                const char *name, CharDriverState *chr,
 240                                int queues)
 241 {
 242     NetClientState *nc;
 243     VhostUserState *s;
 244     int i;
 245
 246     assert(name);
 247     assert(queues > 0);
 248
 249     for (i = 0; i < queues; i++) {
 250         nc = qemu_new_net_client(&net_vhost_user_info, peer, device, name);
 251
 252         snprintf(nc->info_str, sizeof(nc->info_str), "vhost-user%d to %s",
 253                  i, chr->label);
 254
 255         nc->queue_index = i;
 256
 257         s = DO_UPCAST(VhostUserState, nc, nc);
 258         s->chr = chr;
 259     }
 260
 261     qemu_chr_add_handlers(chr, NULL, NULL, net_vhost_user_event, nc[0].name);
 262
 263     return 0;
 264 }
 265
 266 static int net_vhost_chardev_opts(void *opaque,
 267                                   const char *name, const char *value,
 268                                   Error **errp)
 269 {
 270     VhostUserChardevProps *props = opaque;
 271
 272     if (strcmp(name, "backend") == 0 && strcmp(value, "socket") == 0) {
 273         props->is_socket = true;
 274     } else if (strcmp(name, "path") == 0) {
 275         props->is_unix = true;
 276     } else if (strcmp(name, "server") == 0) {
 277     } else {
 278         error_setg(errp,
 279                    "vhost-user does not support a chardev with option %s=%s",
 280                    name, value);
 281         return -1;
 282     }
 283     return 0;
 284 }
 285
 286 static CharDriverState *net_vhost_parse_chardev(
 287     const NetdevVhostUserOptions *opts, Error **errp)
 288 {
 289     CharDriverState *chr = qemu_chr_find(opts->chardev);
 290     VhostUserChardevProps props;
 291
 292     if (chr == NULL) {
 293         error_setg(errp, "chardev \"%s\" not found", opts->chardev);
 294         return NULL;
 295     }
 296
 297     /* inspect chardev opts */
 298     memset(&props, 0, sizeof(props));
 299     if (qemu_opt_foreach(chr->opts, net_vhost_chardev_opts, &props, errp)) {
 300         return NULL;
 301     }
 302
 303     if (!props.is_socket || !props.is_unix) {
 304         error_setg(errp, "chardev \"%s\" is not a unix socket",
 305                    opts->chardev);
 306         return NULL;
 307     }
 308
 309     qemu_chr_fe_claim_no_fail(chr);
 310
 311     return chr;
 312 }
 313
 314 static int net_vhost_check_net(void *opaque, QemuOpts *opts, Error **errp)
 315 {
 316     const char *name = opaque;
 317     const char *driver, *netdev;
 318     const char virtio_name[] = "virtio-net-";
 319
 320     driver = qemu_opt_get(opts, "driver");
 321     netdev = qemu_opt_get(opts, "netdev");
 322
 323     if (!driver || !netdev) {
 324         return 0;
 325     }
 326
 327     if (strcmp(netdev, name) == 0 &&
 328         strncmp(driver, virtio_name, strlen(virtio_name)) != 0) {
 329         error_setg(errp, "vhost-user requires frontend driver virtio-net-*");
 330         return -1;
 331     }
 332
 333     return 0;
 334 }
 335
 336 int net_init_vhost_user(const NetClientOptions *opts, const char *name,
 337                         NetClientState *peer, Error **errp)
 338 {
 339     int queues;
 340     const NetdevVhostUserOptions *vhost_user_opts;
 341     CharDriverState *chr;
 342
 343     assert(opts->type == NET_CLIENT_OPTIONS_KIND_VHOST_USER);
 344     vhost_user_opts = opts->u.vhost_user.data;
 345
 346     chr = net_vhost_parse_chardev(vhost_user_opts, errp);
 347     if (!chr) {
 348         return -1;
 349     }
 350
 351     /* verify net frontend */
 352     if (qemu_opts_foreach(qemu_find_opts("device"), net_vhost_check_net,
 353                           (char *)name, errp)) {
 354         return -1;
 355     }
 356
 357     queues = vhost_user_opts->has_queues ? vhost_user_opts->queues : 1;
 358     if (queues < 1 || queues > MAX_QUEUE_NUM) {
 359         error_setg(errp,
 360                    "vhost-user number of queues must be in range [1, %d]",
 361                    MAX_QUEUE_NUM);
 362         return -1;
 363     }
 364
 365     return net_vhost_user_init(peer, "vhost_user", name, chr, queues);
 366 }