Merge tag 'kraxel-20220704-pull-request' of https://gitlab.com/kraxel/qemu into staging
[qemu.git] / 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 "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
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
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
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
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 options.nvqs = 2;
89 net = vhost_net_init(&options);
90 if (!net) {
91 error_report("failed to init vhost_net for queue %d", i);
92 goto err;
93 }
94
95 if (i == 0) {
96 max_queues = vhost_net_get_max_queues(net);
97 if (queues > max_queues) {
98 error_report("you are asking more queues than supported: %d",
99 max_queues);
100 goto err;
101 }
102 }
103
104 if (s->vhost_net) {
105 vhost_net_cleanup(s->vhost_net);
106 g_free(s->vhost_net);
107 }
108 s->vhost_net = net;
109 }
110
111 return 0;
112
113 err:
114 if (net) {
115 vhost_net_cleanup(net);
116 g_free(net);
117 }
118 vhost_user_stop(i, ncs);
119 return -1;
120 }
121
122 static ssize_t vhost_user_receive(NetClientState *nc, const uint8_t *buf,
123 size_t size)
124 {
125 /* In case of RARP (message size is 60) notify backup to send a fake RARP.
126 This fake RARP will be sent by backend only for guest
127 without GUEST_ANNOUNCE capability.
128 */
129 if (size == 60) {
130 NetVhostUserState *s = DO_UPCAST(NetVhostUserState, nc, nc);
131 int r;
132 static int display_rarp_failure = 1;
133 char mac_addr[6];
134
135 /* extract guest mac address from the RARP message */
136 memcpy(mac_addr, &buf[6], 6);
137
138 r = vhost_net_notify_migration_done(s->vhost_net, mac_addr);
139
140 if ((r != 0) && (display_rarp_failure)) {
141 fprintf(stderr,
142 "Vhost user backend fails to broadcast fake RARP\n");
143 fflush(stderr);
144 display_rarp_failure = 0;
145 }
146 }
147
148 return size;
149 }
150
151 static void net_vhost_user_cleanup(NetClientState *nc)
152 {
153 NetVhostUserState *s = DO_UPCAST(NetVhostUserState, nc, nc);
154
155 if (s->vhost_net) {
156 vhost_net_cleanup(s->vhost_net);
157 g_free(s->vhost_net);
158 s->vhost_net = NULL;
159 }
160 if (nc->queue_index == 0) {
161 if (s->watch) {
162 g_source_remove(s->watch);
163 s->watch = 0;
164 }
165 qemu_chr_fe_deinit(&s->chr, true);
166 if (s->vhost_user) {
167 vhost_user_cleanup(s->vhost_user);
168 g_free(s->vhost_user);
169 s->vhost_user = NULL;
170 }
171 }
172
173 qemu_purge_queued_packets(nc);
174 }
175
176 static int vhost_user_set_vnet_endianness(NetClientState *nc,
177 bool enable)
178 {
179 /* Nothing to do. If the server supports
180 * VHOST_USER_PROTOCOL_F_CROSS_ENDIAN, it will get the
181 * vnet header endianness from there. If it doesn't, negotiation
182 * fails.
183 */
184 return 0;
185 }
186
187 static bool vhost_user_has_vnet_hdr(NetClientState *nc)
188 {
189 assert(nc->info->type == NET_CLIENT_DRIVER_VHOST_USER);
190
191 return true;
192 }
193
194 static bool vhost_user_has_ufo(NetClientState *nc)
195 {
196 assert(nc->info->type == NET_CLIENT_DRIVER_VHOST_USER);
197
198 return true;
199 }
200
201 static bool vhost_user_check_peer_type(NetClientState *nc, ObjectClass *oc,
202 Error **errp)
203 {
204 const char *driver = object_class_get_name(oc);
205
206 if (!g_str_has_prefix(driver, "virtio-net-")) {
207 error_setg(errp, "vhost-user requires frontend driver virtio-net-*");
208 return false;
209 }
210
211 return true;
212 }
213
214 static NetClientInfo net_vhost_user_info = {
215 .type = NET_CLIENT_DRIVER_VHOST_USER,
216 .size = sizeof(NetVhostUserState),
217 .receive = vhost_user_receive,
218 .cleanup = net_vhost_user_cleanup,
219 .has_vnet_hdr = vhost_user_has_vnet_hdr,
220 .has_ufo = vhost_user_has_ufo,
221 .set_vnet_be = vhost_user_set_vnet_endianness,
222 .set_vnet_le = vhost_user_set_vnet_endianness,
223 .check_peer_type = vhost_user_check_peer_type,
224 };
225
226 static gboolean net_vhost_user_watch(void *do_not_use, GIOCondition cond,
227 void *opaque)
228 {
229 NetVhostUserState *s = opaque;
230
231 qemu_chr_fe_disconnect(&s->chr);
232
233 return TRUE;
234 }
235
236 static void net_vhost_user_event(void *opaque, QEMUChrEvent event);
237
238 static void chr_closed_bh(void *opaque)
239 {
240 const char *name = opaque;
241 NetClientState *ncs[MAX_QUEUE_NUM];
242 NetVhostUserState *s;
243 Error *err = NULL;
244 int queues, i;
245
246 queues = qemu_find_net_clients_except(name, ncs,
247 NET_CLIENT_DRIVER_NIC,
248 MAX_QUEUE_NUM);
249 assert(queues < MAX_QUEUE_NUM);
250
251 s = DO_UPCAST(NetVhostUserState, nc, ncs[0]);
252
253 for (i = queues -1; i >= 0; i--) {
254 s = DO_UPCAST(NetVhostUserState, nc, ncs[i]);
255
256 if (s->vhost_net) {
257 s->acked_features = vhost_net_get_acked_features(s->vhost_net);
258 }
259 }
260
261 qmp_set_link(name, false, &err);
262
263 qemu_chr_fe_set_handlers(&s->chr, NULL, NULL, net_vhost_user_event,
264 NULL, opaque, NULL, true);
265
266 if (err) {
267 error_report_err(err);
268 }
269 }
270
271 static void net_vhost_user_event(void *opaque, QEMUChrEvent event)
272 {
273 const char *name = opaque;
274 NetClientState *ncs[MAX_QUEUE_NUM];
275 NetVhostUserState *s;
276 Chardev *chr;
277 Error *err = NULL;
278 int queues;
279
280 queues = qemu_find_net_clients_except(name, ncs,
281 NET_CLIENT_DRIVER_NIC,
282 MAX_QUEUE_NUM);
283 assert(queues < MAX_QUEUE_NUM);
284
285 s = DO_UPCAST(NetVhostUserState, nc, ncs[0]);
286 chr = qemu_chr_fe_get_driver(&s->chr);
287 trace_vhost_user_event(chr->label, event);
288 switch (event) {
289 case CHR_EVENT_OPENED:
290 if (vhost_user_start(queues, ncs, s->vhost_user) < 0) {
291 qemu_chr_fe_disconnect(&s->chr);
292 return;
293 }
294 s->watch = qemu_chr_fe_add_watch(&s->chr, G_IO_HUP,
295 net_vhost_user_watch, s);
296 qmp_set_link(name, true, &err);
297 s->started = true;
298 break;
299 case CHR_EVENT_CLOSED:
300 /* a close event may happen during a read/write, but vhost
301 * code assumes the vhost_dev remains setup, so delay the
302 * stop & clear to idle.
303 * FIXME: better handle failure in vhost code, remove bh
304 */
305 if (s->watch) {
306 AioContext *ctx = qemu_get_current_aio_context();
307
308 g_source_remove(s->watch);
309 s->watch = 0;
310 qemu_chr_fe_set_handlers(&s->chr, NULL, NULL, NULL, NULL,
311 NULL, NULL, false);
312
313 aio_bh_schedule_oneshot(ctx, chr_closed_bh, opaque);
314 }
315 break;
316 case CHR_EVENT_BREAK:
317 case CHR_EVENT_MUX_IN:
318 case CHR_EVENT_MUX_OUT:
319 /* Ignore */
320 break;
321 }
322
323 if (err) {
324 error_report_err(err);
325 }
326 }
327
328 static int net_vhost_user_init(NetClientState *peer, const char *device,
329 const char *name, Chardev *chr,
330 int queues)
331 {
332 Error *err = NULL;
333 NetClientState *nc, *nc0 = NULL;
334 NetVhostUserState *s = NULL;
335 VhostUserState *user;
336 int i;
337
338 assert(name);
339 assert(queues > 0);
340
341 user = g_new0(struct VhostUserState, 1);
342 for (i = 0; i < queues; i++) {
343 nc = qemu_new_net_client(&net_vhost_user_info, peer, device, name);
344 snprintf(nc->info_str, sizeof(nc->info_str), "vhost-user%d to %s",
345 i, chr->label);
346 nc->queue_index = i;
347 if (!nc0) {
348 nc0 = nc;
349 s = DO_UPCAST(NetVhostUserState, nc, nc);
350 if (!qemu_chr_fe_init(&s->chr, chr, &err) ||
351 !vhost_user_init(user, &s->chr, &err)) {
352 error_report_err(err);
353 goto err;
354 }
355 }
356 s = DO_UPCAST(NetVhostUserState, nc, nc);
357 s->vhost_user = user;
358 }
359
360 s = DO_UPCAST(NetVhostUserState, nc, nc0);
361 do {
362 if (qemu_chr_fe_wait_connected(&s->chr, &err) < 0) {
363 error_report_err(err);
364 goto err;
365 }
366 qemu_chr_fe_set_handlers(&s->chr, NULL, NULL,
367 net_vhost_user_event, NULL, nc0->name, NULL,
368 true);
369 } while (!s->started);
370
371 assert(s->vhost_net);
372
373 return 0;
374
375 err:
376 if (user) {
377 vhost_user_cleanup(user);
378 g_free(user);
379 if (s) {
380 s->vhost_user = NULL;
381 }
382 }
383 if (nc0) {
384 qemu_del_net_client(nc0);
385 }
386
387 return -1;
388 }
389
390 static Chardev *net_vhost_claim_chardev(
391 const NetdevVhostUserOptions *opts, Error **errp)
392 {
393 Chardev *chr = qemu_chr_find(opts->chardev);
394
395 if (chr == NULL) {
396 error_setg(errp, "chardev \"%s\" not found", opts->chardev);
397 return NULL;
398 }
399
400 if (!qemu_chr_has_feature(chr, QEMU_CHAR_FEATURE_RECONNECTABLE)) {
401 error_setg(errp, "chardev \"%s\" is not reconnectable",
402 opts->chardev);
403 return NULL;
404 }
405 if (!qemu_chr_has_feature(chr, QEMU_CHAR_FEATURE_FD_PASS)) {
406 error_setg(errp, "chardev \"%s\" does not support FD passing",
407 opts->chardev);
408 return NULL;
409 }
410
411 return chr;
412 }
413
414 int net_init_vhost_user(const Netdev *netdev, const char *name,
415 NetClientState *peer, Error **errp)
416 {
417 int queues;
418 const NetdevVhostUserOptions *vhost_user_opts;
419 Chardev *chr;
420
421 assert(netdev->type == NET_CLIENT_DRIVER_VHOST_USER);
422 vhost_user_opts = &netdev->u.vhost_user;
423
424 chr = net_vhost_claim_chardev(vhost_user_opts, errp);
425 if (!chr) {
426 return -1;
427 }
428
429 queues = vhost_user_opts->has_queues ? vhost_user_opts->queues : 1;
430 if (queues < 1 || queues > MAX_QUEUE_NUM) {
431 error_setg(errp,
432 "vhost-user number of queues must be in range [1, %d]",
433 MAX_QUEUE_NUM);
434 return -1;
435 }
436
437 return net_vhost_user_init(peer, "vhost_user", name, chr, queues);
438 }