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         options.busyloop_timeout = 0;
  95         s->vhost_net = vhost_net_init(&options);
  96         if (!s->vhost_net) {
  97             error_report("failed to init vhost_net for queue %d", i);
  98             goto err;
  99         }
 100
 101         if (i == 0) {
 102             max_queues = vhost_net_get_max_queues(s->vhost_net);
 103             if (queues > max_queues) {
 104                 error_report("you are asking more queues than supported: %d",
 105                              max_queues);
 106                 goto err;
 107             }
 108         }
 109     }
 110
 111     return 0;
 112
 113 err:
 114     vhost_user_stop(i + 1, ncs);
 115     return -1;
 116 }
 117
 118 static ssize_t vhost_user_receive(NetClientState *nc, const uint8_t *buf,
 119                                   size_t size)
 120 {
 121     /* In case of RARP (message size is 60) notify backup to send a fake RARP.
 122        This fake RARP will be sent by backend only for guest
 123        without GUEST_ANNOUNCE capability.
 124      */
 125     if (size == 60) {
 126         VhostUserState *s = DO_UPCAST(VhostUserState, nc, nc);
 127         int r;
 128         static int display_rarp_failure = 1;
 129         char mac_addr[6];
 130
 131         /* extract guest mac address from the RARP message */
 132         memcpy(mac_addr, &buf[6], 6);
 133
 134         r = vhost_net_notify_migration_done(s->vhost_net, mac_addr);
 135
 136         if ((r != 0) && (display_rarp_failure)) {
 137             fprintf(stderr,
 138                     "Vhost user backend fails to broadcast fake RARP\n");
 139             fflush(stderr);
 140             display_rarp_failure = 0;
 141         }
 142     }
 143
 144     return size;
 145 }
 146
 147 static void vhost_user_cleanup(NetClientState *nc)
 148 {
 149     VhostUserState *s = DO_UPCAST(VhostUserState, nc, nc);
 150
 151     if (s->vhost_net) {
 152         vhost_net_cleanup(s->vhost_net);
 153         s->vhost_net = NULL;
 154     }
 155     if (s->chr) {
 156         qemu_chr_add_handlers(s->chr, NULL, NULL, NULL, NULL);
 157         qemu_chr_fe_release(s->chr);
 158         s->chr = NULL;
 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_OPTIONS_KIND_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_OPTIONS_KIND_VHOST_USER);
 174
 175     return true;
 176 }
 177
 178 static NetClientInfo net_vhost_user_info = {
 179         .type = NET_CLIENT_OPTIONS_KIND_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     uint8_t buf[1];
 192
 193     /* We don't actually want to read anything, but CHR_EVENT_CLOSED will be
 194      * raised as a side-effect of the read.
 195      */
 196     qemu_chr_fe_read_all(s->chr, buf, sizeof(buf));
 197
 198     return FALSE;
 199 }
 200
 201 static void net_vhost_user_event(void *opaque, int event)
 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_OPTIONS_KIND_NIC,
 211                                           MAX_QUEUE_NUM);
 212     assert(queues < MAX_QUEUE_NUM);
 213
 214     s = DO_UPCAST(VhostUserState, nc, ncs[0]);
 215     trace_vhost_user_event(s->chr->label, event);
 216     switch (event) {
 217     case CHR_EVENT_OPENED:
 218         s->watch = qemu_chr_fe_add_watch(s->chr, G_IO_HUP,
 219                                          net_vhost_user_watch, s);
 220         if (vhost_user_start(queues, ncs) < 0) {
 221             qemu_chr_disconnect(s->chr);
 222             return;
 223         }
 224         qmp_set_link(name, true, &err);
 225         break;
 226     case CHR_EVENT_CLOSED:
 227         qmp_set_link(name, false, &err);
 228         vhost_user_stop(queues, ncs);
 229         g_source_remove(s->watch);
 230         s->watch = 0;
 231         break;
 232     }
 233
 234     if (err) {
 235         error_report_err(err);
 236     }
 237 }
 238
 239 static int net_vhost_user_init(NetClientState *peer, const char *device,
 240                                const char *name, CharDriverState *chr,
 241                                int queues)
 242 {
 243     NetClientState *nc;
 244     VhostUserState *s;
 245     int i;
 246
 247     assert(name);
 248     assert(queues > 0);
 249
 250     for (i = 0; i < queues; i++) {
 251         nc = qemu_new_net_client(&net_vhost_user_info, peer, device, name);
 252
 253         snprintf(nc->info_str, sizeof(nc->info_str), "vhost-user%d to %s",
 254                  i, chr->label);
 255
 256         nc->queue_index = i;
 257
 258         s = DO_UPCAST(VhostUserState, nc, nc);
 259         s->chr = chr;
 260     }
 261
 262     qemu_chr_add_handlers(chr, NULL, NULL, net_vhost_user_event, nc[0].name);
 263
 264     return 0;
 265 }
 266
 267 static int net_vhost_chardev_opts(void *opaque,
 268                                   const char *name, const char *value,
 269                                   Error **errp)
 270 {
 271     VhostUserChardevProps *props = opaque;
 272
 273     if (strcmp(name, "backend") == 0 && strcmp(value, "socket") == 0) {
 274         props->is_socket = true;
 275     } else if (strcmp(name, "path") == 0) {
 276         props->is_unix = true;
 277     } else if (strcmp(name, "server") == 0) {
 278     } else {
 279         error_setg(errp,
 280                    "vhost-user does not support a chardev with option %s=%s",
 281                    name, value);
 282         return -1;
 283     }
 284     return 0;
 285 }
 286
 287 static CharDriverState *net_vhost_parse_chardev(
 288     const NetdevVhostUserOptions *opts, Error **errp)
 289 {
 290     CharDriverState *chr = qemu_chr_find(opts->chardev);
 291     VhostUserChardevProps props;
 292
 293     if (chr == NULL) {
 294         error_setg(errp, "chardev \"%s\" not found", opts->chardev);
 295         return NULL;
 296     }
 297
 298     /* inspect chardev opts */
 299     memset(&props, 0, sizeof(props));
 300     if (qemu_opt_foreach(chr->opts, net_vhost_chardev_opts, &props, errp)) {
 301         return NULL;
 302     }
 303
 304     if (!props.is_socket || !props.is_unix) {
 305         error_setg(errp, "chardev \"%s\" is not a unix socket",
 306                    opts->chardev);
 307         return NULL;
 308     }
 309
 310     qemu_chr_fe_claim_no_fail(chr);
 311
 312     return chr;
 313 }
 314
 315 static int net_vhost_check_net(void *opaque, QemuOpts *opts, Error **errp)
 316 {
 317     const char *name = opaque;
 318     const char *driver, *netdev;
 319     const char virtio_name[] = "virtio-net-";
 320
 321     driver = qemu_opt_get(opts, "driver");
 322     netdev = qemu_opt_get(opts, "netdev");
 323
 324     if (!driver || !netdev) {
 325         return 0;
 326     }
 327
 328     if (strcmp(netdev, name) == 0 &&
 329         strncmp(driver, virtio_name, strlen(virtio_name)) != 0) {
 330         error_setg(errp, "vhost-user requires frontend driver virtio-net-*");
 331         return -1;
 332     }
 333
 334     return 0;
 335 }
 336
 337 int net_init_vhost_user(const NetClientOptions *opts, const char *name,
 338                         NetClientState *peer, Error **errp)
 339 {
 340     int queues;
 341     const NetdevVhostUserOptions *vhost_user_opts;
 342     CharDriverState *chr;
 343
 344     assert(opts->type == NET_CLIENT_OPTIONS_KIND_VHOST_USER);
 345     vhost_user_opts = opts->u.vhost_user.data;
 346
 347     chr = net_vhost_parse_chardev(vhost_user_opts, errp);
 348     if (!chr) {
 349         return -1;
 350     }
 351
 352     /* verify net frontend */
 353     if (qemu_opts_foreach(qemu_find_opts("device"), net_vhost_check_net,
 354                           (char *)name, errp)) {
 355         return -1;
 356     }
 357
 358     queues = vhost_user_opts->has_queues ? vhost_user_opts->queues : 1;
 359     if (queues < 1 || queues > MAX_QUEUE_NUM) {
 360         error_setg(errp,
 361                    "vhost-user number of queues must be in range [1, %d]",
 362                    MAX_QUEUE_NUM);
 363         return -1;
 364     }
 365
 366     return net_vhost_user_init(peer, "vhost_user", name, chr, queues);
 367 }