]>
Commit | Line | Data |
---|---|---|
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 | 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 | ||
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 | ||
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 | ||
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 | ||
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 | ||
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 | ||
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, QEMUChrEvent event); | |
222 | ||
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 | ||
252 | static void net_vhost_user_event(void *opaque, QEMUChrEvent 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 | case CHR_EVENT_BREAK: | |
298 | case CHR_EVENT_MUX_IN: | |
299 | case CHR_EVENT_MUX_OUT: | |
300 | /* Ignore */ | |
301 | break; | |
302 | } | |
303 | ||
304 | if (err) { | |
305 | error_report_err(err); | |
306 | } | |
307 | } | |
308 | ||
309 | static int net_vhost_user_init(NetClientState *peer, const char *device, | |
310 | const char *name, Chardev *chr, | |
311 | int queues) | |
312 | { | |
313 | Error *err = NULL; | |
314 | NetClientState *nc, *nc0 = NULL; | |
315 | NetVhostUserState *s = NULL; | |
316 | VhostUserState *user; | |
317 | int i; | |
318 | ||
319 | assert(name); | |
320 | assert(queues > 0); | |
321 | ||
322 | user = g_new0(struct VhostUserState, 1); | |
323 | for (i = 0; i < queues; i++) { | |
324 | nc = qemu_new_net_client(&net_vhost_user_info, peer, device, name); | |
325 | snprintf(nc->info_str, sizeof(nc->info_str), "vhost-user%d to %s", | |
326 | i, chr->label); | |
327 | nc->queue_index = i; | |
328 | if (!nc0) { | |
329 | nc0 = nc; | |
330 | s = DO_UPCAST(NetVhostUserState, nc, nc); | |
331 | if (!qemu_chr_fe_init(&s->chr, chr, &err) || | |
332 | !vhost_user_init(user, &s->chr, &err)) { | |
333 | error_report_err(err); | |
334 | goto err; | |
335 | } | |
336 | } | |
337 | s = DO_UPCAST(NetVhostUserState, nc, nc); | |
338 | s->vhost_user = user; | |
339 | } | |
340 | ||
341 | s = DO_UPCAST(NetVhostUserState, nc, nc0); | |
342 | do { | |
343 | if (qemu_chr_fe_wait_connected(&s->chr, &err) < 0) { | |
344 | error_report_err(err); | |
345 | goto err; | |
346 | } | |
347 | qemu_chr_fe_set_handlers(&s->chr, NULL, NULL, | |
348 | net_vhost_user_event, NULL, nc0->name, NULL, | |
349 | true); | |
350 | } while (!s->started); | |
351 | ||
352 | assert(s->vhost_net); | |
353 | ||
354 | return 0; | |
355 | ||
356 | err: | |
357 | if (user) { | |
358 | vhost_user_cleanup(user); | |
359 | g_free(user); | |
360 | if (s) { | |
361 | s->vhost_user = NULL; | |
362 | } | |
363 | } | |
364 | if (nc0) { | |
365 | qemu_del_net_client(nc0); | |
366 | } | |
367 | ||
368 | return -1; | |
369 | } | |
370 | ||
371 | static Chardev *net_vhost_claim_chardev( | |
372 | const NetdevVhostUserOptions *opts, Error **errp) | |
373 | { | |
374 | Chardev *chr = qemu_chr_find(opts->chardev); | |
375 | ||
376 | if (chr == NULL) { | |
377 | error_setg(errp, "chardev \"%s\" not found", opts->chardev); | |
378 | return NULL; | |
379 | } | |
380 | ||
381 | if (!qemu_chr_has_feature(chr, QEMU_CHAR_FEATURE_RECONNECTABLE)) { | |
382 | error_setg(errp, "chardev \"%s\" is not reconnectable", | |
383 | opts->chardev); | |
384 | return NULL; | |
385 | } | |
386 | if (!qemu_chr_has_feature(chr, QEMU_CHAR_FEATURE_FD_PASS)) { | |
387 | error_setg(errp, "chardev \"%s\" does not support FD passing", | |
388 | opts->chardev); | |
389 | return NULL; | |
390 | } | |
391 | ||
392 | return chr; | |
393 | } | |
394 | ||
395 | static int net_vhost_check_net(void *opaque, QemuOpts *opts, Error **errp) | |
396 | { | |
397 | const char *name = opaque; | |
398 | const char *driver, *netdev; | |
399 | ||
400 | driver = qemu_opt_get(opts, "driver"); | |
401 | netdev = qemu_opt_get(opts, "netdev"); | |
402 | ||
403 | if (!driver || !netdev) { | |
404 | return 0; | |
405 | } | |
406 | ||
407 | if (strcmp(netdev, name) == 0 && | |
408 | !g_str_has_prefix(driver, "virtio-net-")) { | |
409 | error_setg(errp, "vhost-user requires frontend driver virtio-net-*"); | |
410 | return -1; | |
411 | } | |
412 | ||
413 | return 0; | |
414 | } | |
415 | ||
416 | int net_init_vhost_user(const Netdev *netdev, const char *name, | |
417 | NetClientState *peer, Error **errp) | |
418 | { | |
419 | int queues; | |
420 | const NetdevVhostUserOptions *vhost_user_opts; | |
421 | Chardev *chr; | |
422 | ||
423 | assert(netdev->type == NET_CLIENT_DRIVER_VHOST_USER); | |
424 | vhost_user_opts = &netdev->u.vhost_user; | |
425 | ||
426 | chr = net_vhost_claim_chardev(vhost_user_opts, errp); | |
427 | if (!chr) { | |
428 | return -1; | |
429 | } | |
430 | ||
431 | /* verify net frontend */ | |
432 | if (qemu_opts_foreach(qemu_find_opts("device"), net_vhost_check_net, | |
433 | (char *)name, errp)) { | |
434 | return -1; | |
435 | } | |
436 | ||
437 | queues = vhost_user_opts->has_queues ? vhost_user_opts->queues : 1; | |
438 | if (queues < 1 || queues > MAX_QUEUE_NUM) { | |
439 | error_setg(errp, | |
440 | "vhost-user number of queues must be in range [1, %d]", | |
441 | MAX_QUEUE_NUM); | |
442 | return -1; | |
443 | } | |
444 | ||
445 | return net_vhost_user_init(peer, "vhost_user", name, chr, queues); | |
446 | } |