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
vhost_user_get_vhost_net(NetClientState * nc)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
vhost_user_get_acked_features(NetClientState * nc)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
vhost_user_stop(int queues,NetClientState * ncs[])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
vhost_user_start(int queues,NetClientState * ncs[],VhostUserState * be)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
vhost_user_receive(NetClientState * nc,const uint8_t * buf,size_t size)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
net_vhost_user_cleanup(NetClientState * nc)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
vhost_user_set_vnet_endianness(NetClientState * nc,bool enable)175 static int vhost_user_set_vnet_endianness(NetClientState *nc,
176 bool enable)
177 {
178 /* Nothing to do. If the server supports
179 * VHOST_USER_PROTOCOL_F_CROSS_ENDIAN, it will get the
180 * vnet header endianness from there. If it doesn't, negotiation
181 * fails.
182 */
183 return 0;
184 }
185
vhost_user_has_vnet_hdr(NetClientState * nc)186 static bool vhost_user_has_vnet_hdr(NetClientState *nc)
187 {
188 assert(nc->info->type == NET_CLIENT_DRIVER_VHOST_USER);
189
190 return true;
191 }
192
vhost_user_has_ufo(NetClientState * nc)193 static bool vhost_user_has_ufo(NetClientState *nc)
194 {
195 assert(nc->info->type == NET_CLIENT_DRIVER_VHOST_USER);
196
197 return true;
198 }
199
200 static NetClientInfo net_vhost_user_info = {
201 .type = NET_CLIENT_DRIVER_VHOST_USER,
202 .size = sizeof(NetVhostUserState),
203 .receive = vhost_user_receive,
204 .cleanup = net_vhost_user_cleanup,
205 .has_vnet_hdr = vhost_user_has_vnet_hdr,
206 .has_ufo = vhost_user_has_ufo,
207 .set_vnet_be = vhost_user_set_vnet_endianness,
208 .set_vnet_le = vhost_user_set_vnet_endianness,
209 };
210
net_vhost_user_watch(GIOChannel * chan,GIOCondition cond,void * opaque)211 static gboolean net_vhost_user_watch(GIOChannel *chan, GIOCondition cond,
212 void *opaque)
213 {
214 NetVhostUserState *s = opaque;
215
216 qemu_chr_fe_disconnect(&s->chr);
217
218 return TRUE;
219 }
220
221 static void net_vhost_user_event(void *opaque, int event);
222
chr_closed_bh(void * opaque)223 static void chr_closed_bh(void *opaque)
224 {
225 const char *name = opaque;
226 NetClientState *ncs[MAX_QUEUE_NUM];
227 NetVhostUserState *s;
228 Error *err = NULL;
229 int queues;
230
231 queues = qemu_find_net_clients_except(name, ncs,
232 NET_CLIENT_DRIVER_NIC,
233 MAX_QUEUE_NUM);
234 assert(queues < MAX_QUEUE_NUM);
235
236 s = DO_UPCAST(NetVhostUserState, nc, ncs[0]);
237
238 if (s->vhost_net) {
239 s->acked_features = vhost_net_get_acked_features(s->vhost_net);
240 }
241
242 qmp_set_link(name, false, &err);
243
244 qemu_chr_fe_set_handlers(&s->chr, NULL, NULL, net_vhost_user_event,
245 NULL, opaque, NULL, true);
246
247 if (err) {
248 error_report_err(err);
249 }
250 }
251
net_vhost_user_event(void * opaque,int event)252 static void net_vhost_user_event(void *opaque, int event)
253 {
254 const char *name = opaque;
255 NetClientState *ncs[MAX_QUEUE_NUM];
256 NetVhostUserState *s;
257 Chardev *chr;
258 Error *err = NULL;
259 int queues;
260
261 queues = qemu_find_net_clients_except(name, ncs,
262 NET_CLIENT_DRIVER_NIC,
263 MAX_QUEUE_NUM);
264 assert(queues < MAX_QUEUE_NUM);
265
266 s = DO_UPCAST(NetVhostUserState, nc, ncs[0]);
267 chr = qemu_chr_fe_get_driver(&s->chr);
268 trace_vhost_user_event(chr->label, event);
269 switch (event) {
270 case CHR_EVENT_OPENED:
271 if (vhost_user_start(queues, ncs, s->vhost_user) < 0) {
272 qemu_chr_fe_disconnect(&s->chr);
273 return;
274 }
275 s->watch = qemu_chr_fe_add_watch(&s->chr, G_IO_HUP,
276 net_vhost_user_watch, s);
277 qmp_set_link(name, true, &err);
278 s->started = true;
279 break;
280 case CHR_EVENT_CLOSED:
281 /* a close event may happen during a read/write, but vhost
282 * code assumes the vhost_dev remains setup, so delay the
283 * stop & clear to idle.
284 * FIXME: better handle failure in vhost code, remove bh
285 */
286 if (s->watch) {
287 AioContext *ctx = qemu_get_current_aio_context();
288
289 g_source_remove(s->watch);
290 s->watch = 0;
291 qemu_chr_fe_set_handlers(&s->chr, NULL, NULL, NULL, NULL,
292 NULL, NULL, false);
293
294 aio_bh_schedule_oneshot(ctx, chr_closed_bh, opaque);
295 }
296 break;
297 }
298
299 if (err) {
300 error_report_err(err);
301 }
302 }
303
net_vhost_user_init(NetClientState * peer,const char * device,const char * name,Chardev * chr,int queues)304 static int net_vhost_user_init(NetClientState *peer, const char *device,
305 const char *name, Chardev *chr,
306 int queues)
307 {
308 Error *err = NULL;
309 NetClientState *nc, *nc0 = NULL;
310 NetVhostUserState *s = NULL;
311 VhostUserState *user;
312 int i;
313
314 assert(name);
315 assert(queues > 0);
316
317 user = g_new0(struct VhostUserState, 1);
318 for (i = 0; i < queues; i++) {
319 nc = qemu_new_net_client(&net_vhost_user_info, peer, device, name);
320 snprintf(nc->info_str, sizeof(nc->info_str), "vhost-user%d to %s",
321 i, chr->label);
322 nc->queue_index = i;
323 if (!nc0) {
324 nc0 = nc;
325 s = DO_UPCAST(NetVhostUserState, nc, nc);
326 if (!qemu_chr_fe_init(&s->chr, chr, &err) ||
327 !vhost_user_init(user, &s->chr, &err)) {
328 error_report_err(err);
329 goto err;
330 }
331 }
332 s = DO_UPCAST(NetVhostUserState, nc, nc);
333 s->vhost_user = user;
334 }
335
336 s = DO_UPCAST(NetVhostUserState, nc, nc0);
337 do {
338 if (qemu_chr_fe_wait_connected(&s->chr, &err) < 0) {
339 error_report_err(err);
340 goto err;
341 }
342 qemu_chr_fe_set_handlers(&s->chr, NULL, NULL,
343 net_vhost_user_event, NULL, nc0->name, NULL,
344 true);
345 } while (!s->started);
346
347 assert(s->vhost_net);
348
349 return 0;
350
351 err:
352 if (user) {
353 vhost_user_cleanup(user);
354 g_free(user);
355 if (s) {
356 s->vhost_user = NULL;
357 }
358 }
359 if (nc0) {
360 qemu_del_net_client(nc0);
361 }
362
363 return -1;
364 }
365
net_vhost_claim_chardev(const NetdevVhostUserOptions * opts,Error ** errp)366 static Chardev *net_vhost_claim_chardev(
367 const NetdevVhostUserOptions *opts, Error **errp)
368 {
369 Chardev *chr = qemu_chr_find(opts->chardev);
370
371 if (chr == NULL) {
372 error_setg(errp, "chardev \"%s\" not found", opts->chardev);
373 return NULL;
374 }
375
376 if (!qemu_chr_has_feature(chr, QEMU_CHAR_FEATURE_RECONNECTABLE)) {
377 error_setg(errp, "chardev \"%s\" is not reconnectable",
378 opts->chardev);
379 return NULL;
380 }
381 if (!qemu_chr_has_feature(chr, QEMU_CHAR_FEATURE_FD_PASS)) {
382 error_setg(errp, "chardev \"%s\" does not support FD passing",
383 opts->chardev);
384 return NULL;
385 }
386
387 return chr;
388 }
389
net_vhost_check_net(void * opaque,QemuOpts * opts,Error ** errp)390 static int net_vhost_check_net(void *opaque, QemuOpts *opts, Error **errp)
391 {
392 const char *name = opaque;
393 const char *driver, *netdev;
394
395 driver = qemu_opt_get(opts, "driver");
396 netdev = qemu_opt_get(opts, "netdev");
397
398 if (!driver || !netdev) {
399 return 0;
400 }
401
402 if (strcmp(netdev, name) == 0 &&
403 !g_str_has_prefix(driver, "virtio-net-")) {
404 error_setg(errp, "vhost-user requires frontend driver virtio-net-*");
405 return -1;
406 }
407
408 return 0;
409 }
410
net_init_vhost_user(const Netdev * netdev,const char * name,NetClientState * peer,Error ** errp)411 int net_init_vhost_user(const Netdev *netdev, const char *name,
412 NetClientState *peer, Error **errp)
413 {
414 int queues;
415 const NetdevVhostUserOptions *vhost_user_opts;
416 Chardev *chr;
417
418 assert(netdev->type == NET_CLIENT_DRIVER_VHOST_USER);
419 vhost_user_opts = &netdev->u.vhost_user;
420
421 chr = net_vhost_claim_chardev(vhost_user_opts, errp);
422 if (!chr) {
423 return -1;
424 }
425
426 /* verify net frontend */
427 if (qemu_opts_foreach(qemu_find_opts("device"), net_vhost_check_net,
428 (char *)name, errp)) {
429 return -1;
430 }
431
432 queues = vhost_user_opts->has_queues ? vhost_user_opts->queues : 1;
433 if (queues < 1 || queues > MAX_QUEUE_NUM) {
434 error_setg(errp,
435 "vhost-user number of queues must be in range [1, %d]",
436 MAX_QUEUE_NUM);
437 return -1;
438 }
439
440 return net_vhost_user_init(peer, "vhost_user", name, chr, queues);
441 }
442