xref: /openbmc/qemu/net/net.c (revision 9abe3bdc)
1 /*
2  * QEMU System Emulator
3  *
4  * Copyright (c) 2003-2008 Fabrice Bellard
5  *
6  * Permission is hereby granted, free of charge, to any person obtaining a copy
7  * of this software and associated documentation files (the "Software"), to deal
8  * in the Software without restriction, including without limitation the rights
9  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10  * copies of the Software, and to permit persons to whom the Software is
11  * furnished to do so, subject to the following conditions:
12  *
13  * The above copyright notice and this permission notice shall be included in
14  * all copies or substantial portions of the Software.
15  *
16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
22  * THE SOFTWARE.
23  */
24 #include "config-host.h"
25 
26 #include "net/net.h"
27 #include "clients.h"
28 #include "hub.h"
29 #include "net/slirp.h"
30 #include "net/eth.h"
31 #include "util.h"
32 
33 #include "monitor/monitor.h"
34 #include "qemu-common.h"
35 #include "qemu/sockets.h"
36 #include "qemu/config-file.h"
37 #include "qmp-commands.h"
38 #include "hw/qdev.h"
39 #include "qemu/iov.h"
40 #include "qemu/main-loop.h"
41 #include "qapi-visit.h"
42 #include "qapi/opts-visitor.h"
43 #include "qapi/dealloc-visitor.h"
44 #include "sysemu/sysemu.h"
45 
46 /* Net bridge is currently not supported for W32. */
47 #if !defined(_WIN32)
48 # define CONFIG_NET_BRIDGE
49 #endif
50 
51 static VMChangeStateEntry *net_change_state_entry;
52 static QTAILQ_HEAD(, NetClientState) net_clients;
53 
54 const char *host_net_devices[] = {
55     "tap",
56     "socket",
57     "dump",
58 #ifdef CONFIG_NET_BRIDGE
59     "bridge",
60 #endif
61 #ifdef CONFIG_SLIRP
62     "user",
63 #endif
64 #ifdef CONFIG_VDE
65     "vde",
66 #endif
67     "vhost-user",
68     NULL,
69 };
70 
71 int default_net = 1;
72 
73 /***********************************************************/
74 /* network device redirectors */
75 
76 #if defined(DEBUG_NET)
77 static void hex_dump(FILE *f, const uint8_t *buf, int size)
78 {
79     int len, i, j, c;
80 
81     for(i=0;i<size;i+=16) {
82         len = size - i;
83         if (len > 16)
84             len = 16;
85         fprintf(f, "%08x ", i);
86         for(j=0;j<16;j++) {
87             if (j < len)
88                 fprintf(f, " %02x", buf[i+j]);
89             else
90                 fprintf(f, "   ");
91         }
92         fprintf(f, " ");
93         for(j=0;j<len;j++) {
94             c = buf[i+j];
95             if (c < ' ' || c > '~')
96                 c = '.';
97             fprintf(f, "%c", c);
98         }
99         fprintf(f, "\n");
100     }
101 }
102 #endif
103 
104 static int get_str_sep(char *buf, int buf_size, const char **pp, int sep)
105 {
106     const char *p, *p1;
107     int len;
108     p = *pp;
109     p1 = strchr(p, sep);
110     if (!p1)
111         return -1;
112     len = p1 - p;
113     p1++;
114     if (buf_size > 0) {
115         if (len > buf_size - 1)
116             len = buf_size - 1;
117         memcpy(buf, p, len);
118         buf[len] = '\0';
119     }
120     *pp = p1;
121     return 0;
122 }
123 
124 int parse_host_port(struct sockaddr_in *saddr, const char *str)
125 {
126     char buf[512];
127     struct hostent *he;
128     const char *p, *r;
129     int port;
130 
131     p = str;
132     if (get_str_sep(buf, sizeof(buf), &p, ':') < 0)
133         return -1;
134     saddr->sin_family = AF_INET;
135     if (buf[0] == '\0') {
136         saddr->sin_addr.s_addr = 0;
137     } else {
138         if (qemu_isdigit(buf[0])) {
139             if (!inet_aton(buf, &saddr->sin_addr))
140                 return -1;
141         } else {
142             if ((he = gethostbyname(buf)) == NULL)
143                 return - 1;
144             saddr->sin_addr = *(struct in_addr *)he->h_addr;
145         }
146     }
147     port = strtol(p, (char **)&r, 0);
148     if (r == p)
149         return -1;
150     saddr->sin_port = htons(port);
151     return 0;
152 }
153 
154 char *qemu_mac_strdup_printf(const uint8_t *macaddr)
155 {
156     return g_strdup_printf("%.2x:%.2x:%.2x:%.2x:%.2x:%.2x",
157                            macaddr[0], macaddr[1], macaddr[2],
158                            macaddr[3], macaddr[4], macaddr[5]);
159 }
160 
161 void qemu_format_nic_info_str(NetClientState *nc, uint8_t macaddr[6])
162 {
163     snprintf(nc->info_str, sizeof(nc->info_str),
164              "model=%s,macaddr=%02x:%02x:%02x:%02x:%02x:%02x",
165              nc->model,
166              macaddr[0], macaddr[1], macaddr[2],
167              macaddr[3], macaddr[4], macaddr[5]);
168 }
169 
170 static int mac_table[256] = {0};
171 
172 static void qemu_macaddr_set_used(MACAddr *macaddr)
173 {
174     int index;
175 
176     for (index = 0x56; index < 0xFF; index++) {
177         if (macaddr->a[5] == index) {
178             mac_table[index]++;
179         }
180     }
181 }
182 
183 static void qemu_macaddr_set_free(MACAddr *macaddr)
184 {
185     int index;
186     static const MACAddr base = { .a = { 0x52, 0x54, 0x00, 0x12, 0x34, 0 } };
187 
188     if (memcmp(macaddr->a, &base.a, (sizeof(base.a) - 1)) != 0) {
189         return;
190     }
191     for (index = 0x56; index < 0xFF; index++) {
192         if (macaddr->a[5] == index) {
193             mac_table[index]--;
194         }
195     }
196 }
197 
198 static int qemu_macaddr_get_free(void)
199 {
200     int index;
201 
202     for (index = 0x56; index < 0xFF; index++) {
203         if (mac_table[index] == 0) {
204             return index;
205         }
206     }
207 
208     return -1;
209 }
210 
211 void qemu_macaddr_default_if_unset(MACAddr *macaddr)
212 {
213     static const MACAddr zero = { .a = { 0,0,0,0,0,0 } };
214     static const MACAddr base = { .a = { 0x52, 0x54, 0x00, 0x12, 0x34, 0 } };
215 
216     if (memcmp(macaddr, &zero, sizeof(zero)) != 0) {
217         if (memcmp(macaddr->a, &base.a, (sizeof(base.a) - 1)) != 0) {
218             return;
219         } else {
220             qemu_macaddr_set_used(macaddr);
221             return;
222         }
223     }
224 
225     macaddr->a[0] = 0x52;
226     macaddr->a[1] = 0x54;
227     macaddr->a[2] = 0x00;
228     macaddr->a[3] = 0x12;
229     macaddr->a[4] = 0x34;
230     macaddr->a[5] = qemu_macaddr_get_free();
231     qemu_macaddr_set_used(macaddr);
232 }
233 
234 /**
235  * Generate a name for net client
236  *
237  * Only net clients created with the legacy -net option and NICs need this.
238  */
239 static char *assign_name(NetClientState *nc1, const char *model)
240 {
241     NetClientState *nc;
242     int id = 0;
243 
244     QTAILQ_FOREACH(nc, &net_clients, next) {
245         if (nc == nc1) {
246             continue;
247         }
248         if (strcmp(nc->model, model) == 0) {
249             id++;
250         }
251     }
252 
253     return g_strdup_printf("%s.%d", model, id);
254 }
255 
256 static void qemu_net_client_destructor(NetClientState *nc)
257 {
258     g_free(nc);
259 }
260 
261 static void qemu_net_client_setup(NetClientState *nc,
262                                   NetClientInfo *info,
263                                   NetClientState *peer,
264                                   const char *model,
265                                   const char *name,
266                                   NetClientDestructor *destructor)
267 {
268     nc->info = info;
269     nc->model = g_strdup(model);
270     if (name) {
271         nc->name = g_strdup(name);
272     } else {
273         nc->name = assign_name(nc, model);
274     }
275 
276     if (peer) {
277         assert(!peer->peer);
278         nc->peer = peer;
279         peer->peer = nc;
280     }
281     QTAILQ_INSERT_TAIL(&net_clients, nc, next);
282 
283     nc->incoming_queue = qemu_new_net_queue(nc);
284     nc->destructor = destructor;
285 }
286 
287 NetClientState *qemu_new_net_client(NetClientInfo *info,
288                                     NetClientState *peer,
289                                     const char *model,
290                                     const char *name)
291 {
292     NetClientState *nc;
293 
294     assert(info->size >= sizeof(NetClientState));
295 
296     nc = g_malloc0(info->size);
297     qemu_net_client_setup(nc, info, peer, model, name,
298                           qemu_net_client_destructor);
299 
300     return nc;
301 }
302 
303 NICState *qemu_new_nic(NetClientInfo *info,
304                        NICConf *conf,
305                        const char *model,
306                        const char *name,
307                        void *opaque)
308 {
309     NetClientState **peers = conf->peers.ncs;
310     NICState *nic;
311     int i, queues = MAX(1, conf->peers.queues);
312 
313     assert(info->type == NET_CLIENT_OPTIONS_KIND_NIC);
314     assert(info->size >= sizeof(NICState));
315 
316     nic = g_malloc0(info->size + sizeof(NetClientState) * queues);
317     nic->ncs = (void *)nic + info->size;
318     nic->conf = conf;
319     nic->opaque = opaque;
320 
321     for (i = 0; i < queues; i++) {
322         qemu_net_client_setup(&nic->ncs[i], info, peers[i], model, name,
323                               NULL);
324         nic->ncs[i].queue_index = i;
325     }
326 
327     return nic;
328 }
329 
330 NetClientState *qemu_get_subqueue(NICState *nic, int queue_index)
331 {
332     return nic->ncs + queue_index;
333 }
334 
335 NetClientState *qemu_get_queue(NICState *nic)
336 {
337     return qemu_get_subqueue(nic, 0);
338 }
339 
340 NICState *qemu_get_nic(NetClientState *nc)
341 {
342     NetClientState *nc0 = nc - nc->queue_index;
343 
344     return (NICState *)((void *)nc0 - nc->info->size);
345 }
346 
347 void *qemu_get_nic_opaque(NetClientState *nc)
348 {
349     NICState *nic = qemu_get_nic(nc);
350 
351     return nic->opaque;
352 }
353 
354 static void qemu_cleanup_net_client(NetClientState *nc)
355 {
356     QTAILQ_REMOVE(&net_clients, nc, next);
357 
358     if (nc->info->cleanup) {
359         nc->info->cleanup(nc);
360     }
361 }
362 
363 static void qemu_free_net_client(NetClientState *nc)
364 {
365     if (nc->incoming_queue) {
366         qemu_del_net_queue(nc->incoming_queue);
367     }
368     if (nc->peer) {
369         nc->peer->peer = NULL;
370     }
371     g_free(nc->name);
372     g_free(nc->model);
373     if (nc->destructor) {
374         nc->destructor(nc);
375     }
376 }
377 
378 void qemu_del_net_client(NetClientState *nc)
379 {
380     NetClientState *ncs[MAX_QUEUE_NUM];
381     int queues, i;
382 
383     assert(nc->info->type != NET_CLIENT_OPTIONS_KIND_NIC);
384 
385     /* If the NetClientState belongs to a multiqueue backend, we will change all
386      * other NetClientStates also.
387      */
388     queues = qemu_find_net_clients_except(nc->name, ncs,
389                                           NET_CLIENT_OPTIONS_KIND_NIC,
390                                           MAX_QUEUE_NUM);
391     assert(queues != 0);
392 
393     /* If there is a peer NIC, delete and cleanup client, but do not free. */
394     if (nc->peer && nc->peer->info->type == NET_CLIENT_OPTIONS_KIND_NIC) {
395         NICState *nic = qemu_get_nic(nc->peer);
396         if (nic->peer_deleted) {
397             return;
398         }
399         nic->peer_deleted = true;
400 
401         for (i = 0; i < queues; i++) {
402             ncs[i]->peer->link_down = true;
403         }
404 
405         if (nc->peer->info->link_status_changed) {
406             nc->peer->info->link_status_changed(nc->peer);
407         }
408 
409         for (i = 0; i < queues; i++) {
410             qemu_cleanup_net_client(ncs[i]);
411         }
412 
413         return;
414     }
415 
416     for (i = 0; i < queues; i++) {
417         qemu_cleanup_net_client(ncs[i]);
418         qemu_free_net_client(ncs[i]);
419     }
420 }
421 
422 void qemu_del_nic(NICState *nic)
423 {
424     int i, queues = MAX(nic->conf->peers.queues, 1);
425 
426     qemu_macaddr_set_free(&nic->conf->macaddr);
427 
428     /* If this is a peer NIC and peer has already been deleted, free it now. */
429     if (nic->peer_deleted) {
430         for (i = 0; i < queues; i++) {
431             qemu_free_net_client(qemu_get_subqueue(nic, i)->peer);
432         }
433     }
434 
435     for (i = queues - 1; i >= 0; i--) {
436         NetClientState *nc = qemu_get_subqueue(nic, i);
437 
438         qemu_cleanup_net_client(nc);
439         qemu_free_net_client(nc);
440     }
441 
442     g_free(nic);
443 }
444 
445 void qemu_foreach_nic(qemu_nic_foreach func, void *opaque)
446 {
447     NetClientState *nc;
448 
449     QTAILQ_FOREACH(nc, &net_clients, next) {
450         if (nc->info->type == NET_CLIENT_OPTIONS_KIND_NIC) {
451             if (nc->queue_index == 0) {
452                 func(qemu_get_nic(nc), opaque);
453             }
454         }
455     }
456 }
457 
458 bool qemu_has_ufo(NetClientState *nc)
459 {
460     if (!nc || !nc->info->has_ufo) {
461         return false;
462     }
463 
464     return nc->info->has_ufo(nc);
465 }
466 
467 bool qemu_has_vnet_hdr(NetClientState *nc)
468 {
469     if (!nc || !nc->info->has_vnet_hdr) {
470         return false;
471     }
472 
473     return nc->info->has_vnet_hdr(nc);
474 }
475 
476 bool qemu_has_vnet_hdr_len(NetClientState *nc, int len)
477 {
478     if (!nc || !nc->info->has_vnet_hdr_len) {
479         return false;
480     }
481 
482     return nc->info->has_vnet_hdr_len(nc, len);
483 }
484 
485 void qemu_using_vnet_hdr(NetClientState *nc, bool enable)
486 {
487     if (!nc || !nc->info->using_vnet_hdr) {
488         return;
489     }
490 
491     nc->info->using_vnet_hdr(nc, enable);
492 }
493 
494 void qemu_set_offload(NetClientState *nc, int csum, int tso4, int tso6,
495                           int ecn, int ufo)
496 {
497     if (!nc || !nc->info->set_offload) {
498         return;
499     }
500 
501     nc->info->set_offload(nc, csum, tso4, tso6, ecn, ufo);
502 }
503 
504 void qemu_set_vnet_hdr_len(NetClientState *nc, int len)
505 {
506     if (!nc || !nc->info->set_vnet_hdr_len) {
507         return;
508     }
509 
510     nc->info->set_vnet_hdr_len(nc, len);
511 }
512 
513 int qemu_can_send_packet(NetClientState *sender)
514 {
515     int vm_running = runstate_is_running();
516 
517     if (!vm_running) {
518         return 0;
519     }
520 
521     if (!sender->peer) {
522         return 1;
523     }
524 
525     if (sender->peer->receive_disabled) {
526         return 0;
527     } else if (sender->peer->info->can_receive &&
528                !sender->peer->info->can_receive(sender->peer)) {
529         return 0;
530     }
531     return 1;
532 }
533 
534 ssize_t qemu_deliver_packet(NetClientState *sender,
535                             unsigned flags,
536                             const uint8_t *data,
537                             size_t size,
538                             void *opaque)
539 {
540     NetClientState *nc = opaque;
541     ssize_t ret;
542 
543     if (nc->link_down) {
544         return size;
545     }
546 
547     if (nc->receive_disabled) {
548         return 0;
549     }
550 
551     if (flags & QEMU_NET_PACKET_FLAG_RAW && nc->info->receive_raw) {
552         ret = nc->info->receive_raw(nc, data, size);
553     } else {
554         ret = nc->info->receive(nc, data, size);
555     }
556 
557     if (ret == 0) {
558         nc->receive_disabled = 1;
559     }
560 
561     return ret;
562 }
563 
564 void qemu_purge_queued_packets(NetClientState *nc)
565 {
566     if (!nc->peer) {
567         return;
568     }
569 
570     qemu_net_queue_purge(nc->peer->incoming_queue, nc);
571 }
572 
573 static
574 void qemu_flush_or_purge_queued_packets(NetClientState *nc, bool purge)
575 {
576     nc->receive_disabled = 0;
577 
578     if (nc->peer && nc->peer->info->type == NET_CLIENT_OPTIONS_KIND_HUBPORT) {
579         if (net_hub_flush(nc->peer)) {
580             qemu_notify_event();
581         }
582     }
583     if (qemu_net_queue_flush(nc->incoming_queue)) {
584         /* We emptied the queue successfully, signal to the IO thread to repoll
585          * the file descriptor (for tap, for example).
586          */
587         qemu_notify_event();
588     } else if (purge) {
589         /* Unable to empty the queue, purge remaining packets */
590         qemu_net_queue_purge(nc->incoming_queue, nc);
591     }
592 }
593 
594 void qemu_flush_queued_packets(NetClientState *nc)
595 {
596     qemu_flush_or_purge_queued_packets(nc, false);
597 }
598 
599 static ssize_t qemu_send_packet_async_with_flags(NetClientState *sender,
600                                                  unsigned flags,
601                                                  const uint8_t *buf, int size,
602                                                  NetPacketSent *sent_cb)
603 {
604     NetQueue *queue;
605 
606 #ifdef DEBUG_NET
607     printf("qemu_send_packet_async:\n");
608     hex_dump(stdout, buf, size);
609 #endif
610 
611     if (sender->link_down || !sender->peer) {
612         return size;
613     }
614 
615     queue = sender->peer->incoming_queue;
616 
617     return qemu_net_queue_send(queue, sender, flags, buf, size, sent_cb);
618 }
619 
620 ssize_t qemu_send_packet_async(NetClientState *sender,
621                                const uint8_t *buf, int size,
622                                NetPacketSent *sent_cb)
623 {
624     return qemu_send_packet_async_with_flags(sender, QEMU_NET_PACKET_FLAG_NONE,
625                                              buf, size, sent_cb);
626 }
627 
628 void qemu_send_packet(NetClientState *nc, const uint8_t *buf, int size)
629 {
630     qemu_send_packet_async(nc, buf, size, NULL);
631 }
632 
633 ssize_t qemu_send_packet_raw(NetClientState *nc, const uint8_t *buf, int size)
634 {
635     return qemu_send_packet_async_with_flags(nc, QEMU_NET_PACKET_FLAG_RAW,
636                                              buf, size, NULL);
637 }
638 
639 static ssize_t nc_sendv_compat(NetClientState *nc, const struct iovec *iov,
640                                int iovcnt)
641 {
642     uint8_t buffer[NET_BUFSIZE];
643     size_t offset;
644 
645     offset = iov_to_buf(iov, iovcnt, 0, buffer, sizeof(buffer));
646 
647     return nc->info->receive(nc, buffer, offset);
648 }
649 
650 ssize_t qemu_deliver_packet_iov(NetClientState *sender,
651                                 unsigned flags,
652                                 const struct iovec *iov,
653                                 int iovcnt,
654                                 void *opaque)
655 {
656     NetClientState *nc = opaque;
657     int ret;
658 
659     if (nc->link_down) {
660         return iov_size(iov, iovcnt);
661     }
662 
663     if (nc->receive_disabled) {
664         return 0;
665     }
666 
667     if (nc->info->receive_iov) {
668         ret = nc->info->receive_iov(nc, iov, iovcnt);
669     } else {
670         ret = nc_sendv_compat(nc, iov, iovcnt);
671     }
672 
673     if (ret == 0) {
674         nc->receive_disabled = 1;
675     }
676 
677     return ret;
678 }
679 
680 ssize_t qemu_sendv_packet_async(NetClientState *sender,
681                                 const struct iovec *iov, int iovcnt,
682                                 NetPacketSent *sent_cb)
683 {
684     NetQueue *queue;
685 
686     if (sender->link_down || !sender->peer) {
687         return iov_size(iov, iovcnt);
688     }
689 
690     queue = sender->peer->incoming_queue;
691 
692     return qemu_net_queue_send_iov(queue, sender,
693                                    QEMU_NET_PACKET_FLAG_NONE,
694                                    iov, iovcnt, sent_cb);
695 }
696 
697 ssize_t
698 qemu_sendv_packet(NetClientState *nc, const struct iovec *iov, int iovcnt)
699 {
700     return qemu_sendv_packet_async(nc, iov, iovcnt, NULL);
701 }
702 
703 NetClientState *qemu_find_netdev(const char *id)
704 {
705     NetClientState *nc;
706 
707     QTAILQ_FOREACH(nc, &net_clients, next) {
708         if (nc->info->type == NET_CLIENT_OPTIONS_KIND_NIC)
709             continue;
710         if (!strcmp(nc->name, id)) {
711             return nc;
712         }
713     }
714 
715     return NULL;
716 }
717 
718 int qemu_find_net_clients_except(const char *id, NetClientState **ncs,
719                                  NetClientOptionsKind type, int max)
720 {
721     NetClientState *nc;
722     int ret = 0;
723 
724     QTAILQ_FOREACH(nc, &net_clients, next) {
725         if (nc->info->type == type) {
726             continue;
727         }
728         if (!id || !strcmp(nc->name, id)) {
729             if (ret < max) {
730                 ncs[ret] = nc;
731             }
732             ret++;
733         }
734     }
735 
736     return ret;
737 }
738 
739 static int nic_get_free_idx(void)
740 {
741     int index;
742 
743     for (index = 0; index < MAX_NICS; index++)
744         if (!nd_table[index].used)
745             return index;
746     return -1;
747 }
748 
749 int qemu_show_nic_models(const char *arg, const char *const *models)
750 {
751     int i;
752 
753     if (!arg || !is_help_option(arg)) {
754         return 0;
755     }
756 
757     fprintf(stderr, "qemu: Supported NIC models: ");
758     for (i = 0 ; models[i]; i++)
759         fprintf(stderr, "%s%c", models[i], models[i+1] ? ',' : '\n');
760     return 1;
761 }
762 
763 void qemu_check_nic_model(NICInfo *nd, const char *model)
764 {
765     const char *models[2];
766 
767     models[0] = model;
768     models[1] = NULL;
769 
770     if (qemu_show_nic_models(nd->model, models))
771         exit(0);
772     if (qemu_find_nic_model(nd, models, model) < 0)
773         exit(1);
774 }
775 
776 int qemu_find_nic_model(NICInfo *nd, const char * const *models,
777                         const char *default_model)
778 {
779     int i;
780 
781     if (!nd->model)
782         nd->model = g_strdup(default_model);
783 
784     for (i = 0 ; models[i]; i++) {
785         if (strcmp(nd->model, models[i]) == 0)
786             return i;
787     }
788 
789     error_report("Unsupported NIC model: %s", nd->model);
790     return -1;
791 }
792 
793 static int net_init_nic(const NetClientOptions *opts, const char *name,
794                         NetClientState *peer, Error **errp)
795 {
796     int idx;
797     NICInfo *nd;
798     const NetLegacyNicOptions *nic;
799 
800     assert(opts->kind == NET_CLIENT_OPTIONS_KIND_NIC);
801     nic = opts->nic;
802 
803     idx = nic_get_free_idx();
804     if (idx == -1 || nb_nics >= MAX_NICS) {
805         error_setg(errp, "too many NICs");
806         return -1;
807     }
808 
809     nd = &nd_table[idx];
810 
811     memset(nd, 0, sizeof(*nd));
812 
813     if (nic->has_netdev) {
814         nd->netdev = qemu_find_netdev(nic->netdev);
815         if (!nd->netdev) {
816             error_setg(errp, "netdev '%s' not found", nic->netdev);
817             return -1;
818         }
819     } else {
820         assert(peer);
821         nd->netdev = peer;
822     }
823     nd->name = g_strdup(name);
824     if (nic->has_model) {
825         nd->model = g_strdup(nic->model);
826     }
827     if (nic->has_addr) {
828         nd->devaddr = g_strdup(nic->addr);
829     }
830 
831     if (nic->has_macaddr &&
832         net_parse_macaddr(nd->macaddr.a, nic->macaddr) < 0) {
833         error_setg(errp, "invalid syntax for ethernet address");
834         return -1;
835     }
836     if (nic->has_macaddr &&
837         is_multicast_ether_addr(nd->macaddr.a)) {
838         error_setg(errp,
839                    "NIC cannot have multicast MAC address (odd 1st byte)");
840         return -1;
841     }
842     qemu_macaddr_default_if_unset(&nd->macaddr);
843 
844     if (nic->has_vectors) {
845         if (nic->vectors > 0x7ffffff) {
846             error_setg(errp, "invalid # of vectors: %"PRIu32, nic->vectors);
847             return -1;
848         }
849         nd->nvectors = nic->vectors;
850     } else {
851         nd->nvectors = DEV_NVECTORS_UNSPECIFIED;
852     }
853 
854     nd->used = 1;
855     nb_nics++;
856 
857     return idx;
858 }
859 
860 
861 static int (* const net_client_init_fun[NET_CLIENT_OPTIONS_KIND_MAX])(
862     const NetClientOptions *opts,
863     const char *name,
864     NetClientState *peer, Error **errp) = {
865         [NET_CLIENT_OPTIONS_KIND_NIC]       = net_init_nic,
866 #ifdef CONFIG_SLIRP
867         [NET_CLIENT_OPTIONS_KIND_USER]      = net_init_slirp,
868 #endif
869         [NET_CLIENT_OPTIONS_KIND_TAP]       = net_init_tap,
870         [NET_CLIENT_OPTIONS_KIND_SOCKET]    = net_init_socket,
871 #ifdef CONFIG_VDE
872         [NET_CLIENT_OPTIONS_KIND_VDE]       = net_init_vde,
873 #endif
874 #ifdef CONFIG_NETMAP
875         [NET_CLIENT_OPTIONS_KIND_NETMAP]    = net_init_netmap,
876 #endif
877         [NET_CLIENT_OPTIONS_KIND_DUMP]      = net_init_dump,
878 #ifdef CONFIG_NET_BRIDGE
879         [NET_CLIENT_OPTIONS_KIND_BRIDGE]    = net_init_bridge,
880 #endif
881         [NET_CLIENT_OPTIONS_KIND_HUBPORT]   = net_init_hubport,
882 #ifdef CONFIG_VHOST_NET_USED
883         [NET_CLIENT_OPTIONS_KIND_VHOST_USER] = net_init_vhost_user,
884 #endif
885 #ifdef CONFIG_L2TPV3
886         [NET_CLIENT_OPTIONS_KIND_L2TPV3]    = net_init_l2tpv3,
887 #endif
888 };
889 
890 
891 static int net_client_init1(const void *object, int is_netdev, Error **errp)
892 {
893     union {
894         const Netdev    *netdev;
895         const NetLegacy *net;
896     } u;
897     const NetClientOptions *opts;
898     const char *name;
899 
900     if (is_netdev) {
901         u.netdev = object;
902         opts = u.netdev->opts;
903         name = u.netdev->id;
904 
905         switch (opts->kind) {
906 #ifdef CONFIG_SLIRP
907         case NET_CLIENT_OPTIONS_KIND_USER:
908 #endif
909         case NET_CLIENT_OPTIONS_KIND_TAP:
910         case NET_CLIENT_OPTIONS_KIND_SOCKET:
911 #ifdef CONFIG_VDE
912         case NET_CLIENT_OPTIONS_KIND_VDE:
913 #endif
914 #ifdef CONFIG_NETMAP
915         case NET_CLIENT_OPTIONS_KIND_NETMAP:
916 #endif
917 #ifdef CONFIG_NET_BRIDGE
918         case NET_CLIENT_OPTIONS_KIND_BRIDGE:
919 #endif
920         case NET_CLIENT_OPTIONS_KIND_HUBPORT:
921 #ifdef CONFIG_VHOST_NET_USED
922         case NET_CLIENT_OPTIONS_KIND_VHOST_USER:
923 #endif
924 #ifdef CONFIG_L2TPV3
925         case NET_CLIENT_OPTIONS_KIND_L2TPV3:
926 #endif
927             break;
928 
929         default:
930             error_set(errp, QERR_INVALID_PARAMETER_VALUE, "type",
931                       "a netdev backend type");
932             return -1;
933         }
934     } else {
935         u.net = object;
936         opts = u.net->opts;
937         if (opts->kind == NET_CLIENT_OPTIONS_KIND_HUBPORT) {
938             error_set(errp, QERR_INVALID_PARAMETER_VALUE, "type",
939                       "a net type");
940             return -1;
941         }
942         /* missing optional values have been initialized to "all bits zero" */
943         name = u.net->has_id ? u.net->id : u.net->name;
944     }
945 
946     if (net_client_init_fun[opts->kind]) {
947         NetClientState *peer = NULL;
948 
949         /* Do not add to a vlan if it's a -netdev or a nic with a netdev=
950          * parameter. */
951         if (!is_netdev &&
952             (opts->kind != NET_CLIENT_OPTIONS_KIND_NIC ||
953              !opts->nic->has_netdev)) {
954             peer = net_hub_add_port(u.net->has_vlan ? u.net->vlan : 0, NULL);
955         }
956 
957         if (net_client_init_fun[opts->kind](opts, name, peer, errp) < 0) {
958             /* FIXME drop when all init functions store an Error */
959             if (errp && !*errp) {
960                 error_set(errp, QERR_DEVICE_INIT_FAILED,
961                           NetClientOptionsKind_lookup[opts->kind]);
962             }
963             return -1;
964         }
965     }
966     return 0;
967 }
968 
969 
970 static void net_visit(Visitor *v, int is_netdev, void **object, Error **errp)
971 {
972     if (is_netdev) {
973         visit_type_Netdev(v, (Netdev **)object, NULL, errp);
974     } else {
975         visit_type_NetLegacy(v, (NetLegacy **)object, NULL, errp);
976     }
977 }
978 
979 
980 int net_client_init(QemuOpts *opts, int is_netdev, Error **errp)
981 {
982     void *object = NULL;
983     Error *err = NULL;
984     int ret = -1;
985 
986     {
987         OptsVisitor *ov = opts_visitor_new(opts);
988 
989         net_visit(opts_get_visitor(ov), is_netdev, &object, &err);
990         opts_visitor_cleanup(ov);
991     }
992 
993     if (!err) {
994         ret = net_client_init1(object, is_netdev, &err);
995     }
996 
997     if (object) {
998         QapiDeallocVisitor *dv = qapi_dealloc_visitor_new();
999 
1000         net_visit(qapi_dealloc_get_visitor(dv), is_netdev, &object, NULL);
1001         qapi_dealloc_visitor_cleanup(dv);
1002     }
1003 
1004     error_propagate(errp, err);
1005     return ret;
1006 }
1007 
1008 
1009 static int net_host_check_device(const char *device)
1010 {
1011     int i;
1012     for (i = 0; host_net_devices[i]; i++) {
1013         if (!strncmp(host_net_devices[i], device,
1014                      strlen(host_net_devices[i]))) {
1015             return 1;
1016         }
1017     }
1018 
1019     return 0;
1020 }
1021 
1022 void hmp_host_net_add(Monitor *mon, const QDict *qdict)
1023 {
1024     const char *device = qdict_get_str(qdict, "device");
1025     const char *opts_str = qdict_get_try_str(qdict, "opts");
1026     Error *local_err = NULL;
1027     QemuOpts *opts;
1028 
1029     if (!net_host_check_device(device)) {
1030         monitor_printf(mon, "invalid host network device %s\n", device);
1031         return;
1032     }
1033 
1034     opts = qemu_opts_parse(qemu_find_opts("net"), opts_str ? opts_str : "", 0);
1035     if (!opts) {
1036         return;
1037     }
1038 
1039     qemu_opt_set(opts, "type", device, &error_abort);
1040 
1041     net_client_init(opts, 0, &local_err);
1042     if (local_err) {
1043         error_report_err(local_err);
1044         monitor_printf(mon, "adding host network device %s failed\n", device);
1045     }
1046 }
1047 
1048 void hmp_host_net_remove(Monitor *mon, const QDict *qdict)
1049 {
1050     NetClientState *nc;
1051     int vlan_id = qdict_get_int(qdict, "vlan_id");
1052     const char *device = qdict_get_str(qdict, "device");
1053 
1054     nc = net_hub_find_client_by_name(vlan_id, device);
1055     if (!nc) {
1056         error_report("Host network device '%s' on hub '%d' not found",
1057                      device, vlan_id);
1058         return;
1059     }
1060     if (nc->info->type == NET_CLIENT_OPTIONS_KIND_NIC) {
1061         error_report("invalid host network device '%s'", device);
1062         return;
1063     }
1064 
1065     qemu_del_net_client(nc->peer);
1066     qemu_del_net_client(nc);
1067 }
1068 
1069 void netdev_add(QemuOpts *opts, Error **errp)
1070 {
1071     net_client_init(opts, 1, errp);
1072 }
1073 
1074 int qmp_netdev_add(Monitor *mon, const QDict *qdict, QObject **ret)
1075 {
1076     Error *local_err = NULL;
1077     QemuOptsList *opts_list;
1078     QemuOpts *opts;
1079 
1080     opts_list = qemu_find_opts_err("netdev", &local_err);
1081     if (local_err) {
1082         goto exit_err;
1083     }
1084 
1085     opts = qemu_opts_from_qdict(opts_list, qdict, &local_err);
1086     if (local_err) {
1087         goto exit_err;
1088     }
1089 
1090     netdev_add(opts, &local_err);
1091     if (local_err) {
1092         qemu_opts_del(opts);
1093         goto exit_err;
1094     }
1095 
1096     return 0;
1097 
1098 exit_err:
1099     qerror_report_err(local_err);
1100     error_free(local_err);
1101     return -1;
1102 }
1103 
1104 void qmp_netdev_del(const char *id, Error **errp)
1105 {
1106     NetClientState *nc;
1107     QemuOpts *opts;
1108 
1109     nc = qemu_find_netdev(id);
1110     if (!nc) {
1111         error_set(errp, QERR_DEVICE_NOT_FOUND, id);
1112         return;
1113     }
1114 
1115     opts = qemu_opts_find(qemu_find_opts_err("netdev", NULL), id);
1116     if (!opts) {
1117         error_setg(errp, "Device '%s' is not a netdev", id);
1118         return;
1119     }
1120 
1121     qemu_del_net_client(nc);
1122     qemu_opts_del(opts);
1123 }
1124 
1125 void print_net_client(Monitor *mon, NetClientState *nc)
1126 {
1127     monitor_printf(mon, "%s: index=%d,type=%s,%s\n", nc->name,
1128                    nc->queue_index,
1129                    NetClientOptionsKind_lookup[nc->info->type],
1130                    nc->info_str);
1131 }
1132 
1133 RxFilterInfoList *qmp_query_rx_filter(bool has_name, const char *name,
1134                                       Error **errp)
1135 {
1136     NetClientState *nc;
1137     RxFilterInfoList *filter_list = NULL, *last_entry = NULL;
1138 
1139     QTAILQ_FOREACH(nc, &net_clients, next) {
1140         RxFilterInfoList *entry;
1141         RxFilterInfo *info;
1142 
1143         if (has_name && strcmp(nc->name, name) != 0) {
1144             continue;
1145         }
1146 
1147         /* only query rx-filter information of NIC */
1148         if (nc->info->type != NET_CLIENT_OPTIONS_KIND_NIC) {
1149             if (has_name) {
1150                 error_setg(errp, "net client(%s) isn't a NIC", name);
1151                 return NULL;
1152             }
1153             continue;
1154         }
1155 
1156         if (nc->info->query_rx_filter) {
1157             info = nc->info->query_rx_filter(nc);
1158             entry = g_malloc0(sizeof(*entry));
1159             entry->value = info;
1160 
1161             if (!filter_list) {
1162                 filter_list = entry;
1163             } else {
1164                 last_entry->next = entry;
1165             }
1166             last_entry = entry;
1167         } else if (has_name) {
1168             error_setg(errp, "net client(%s) doesn't support"
1169                        " rx-filter querying", name);
1170             return NULL;
1171         }
1172 
1173         if (has_name) {
1174             break;
1175         }
1176     }
1177 
1178     if (filter_list == NULL && has_name) {
1179         error_setg(errp, "invalid net client name: %s", name);
1180     }
1181 
1182     return filter_list;
1183 }
1184 
1185 void hmp_info_network(Monitor *mon, const QDict *qdict)
1186 {
1187     NetClientState *nc, *peer;
1188     NetClientOptionsKind type;
1189 
1190     net_hub_info(mon);
1191 
1192     QTAILQ_FOREACH(nc, &net_clients, next) {
1193         peer = nc->peer;
1194         type = nc->info->type;
1195 
1196         /* Skip if already printed in hub info */
1197         if (net_hub_id_for_client(nc, NULL) == 0) {
1198             continue;
1199         }
1200 
1201         if (!peer || type == NET_CLIENT_OPTIONS_KIND_NIC) {
1202             print_net_client(mon, nc);
1203         } /* else it's a netdev connected to a NIC, printed with the NIC */
1204         if (peer && type == NET_CLIENT_OPTIONS_KIND_NIC) {
1205             monitor_printf(mon, " \\ ");
1206             print_net_client(mon, peer);
1207         }
1208     }
1209 }
1210 
1211 void qmp_set_link(const char *name, bool up, Error **errp)
1212 {
1213     NetClientState *ncs[MAX_QUEUE_NUM];
1214     NetClientState *nc;
1215     int queues, i;
1216 
1217     queues = qemu_find_net_clients_except(name, ncs,
1218                                           NET_CLIENT_OPTIONS_KIND_MAX,
1219                                           MAX_QUEUE_NUM);
1220 
1221     if (queues == 0) {
1222         error_set(errp, QERR_DEVICE_NOT_FOUND, name);
1223         return;
1224     }
1225     nc = ncs[0];
1226 
1227     for (i = 0; i < queues; i++) {
1228         ncs[i]->link_down = !up;
1229     }
1230 
1231     if (nc->info->link_status_changed) {
1232         nc->info->link_status_changed(nc);
1233     }
1234 
1235     if (nc->peer) {
1236         /* Change peer link only if the peer is NIC and then notify peer.
1237          * If the peer is a HUBPORT or a backend, we do not change the
1238          * link status.
1239          *
1240          * This behavior is compatible with qemu vlans where there could be
1241          * multiple clients that can still communicate with each other in
1242          * disconnected mode. For now maintain this compatibility.
1243          */
1244         if (nc->peer->info->type == NET_CLIENT_OPTIONS_KIND_NIC) {
1245             for (i = 0; i < queues; i++) {
1246                 ncs[i]->peer->link_down = !up;
1247             }
1248         }
1249         if (nc->peer->info->link_status_changed) {
1250             nc->peer->info->link_status_changed(nc->peer);
1251         }
1252     }
1253 }
1254 
1255 static void net_vm_change_state_handler(void *opaque, int running,
1256                                         RunState state)
1257 {
1258     /* Complete all queued packets, to guarantee we don't modify
1259      * state later when VM is not running.
1260      */
1261     if (!running) {
1262         NetClientState *nc;
1263         NetClientState *tmp;
1264 
1265         QTAILQ_FOREACH_SAFE(nc, &net_clients, next, tmp) {
1266             qemu_flush_or_purge_queued_packets(nc, true);
1267         }
1268     }
1269 }
1270 
1271 void net_cleanup(void)
1272 {
1273     NetClientState *nc;
1274 
1275     /* We may del multiple entries during qemu_del_net_client(),
1276      * so QTAILQ_FOREACH_SAFE() is also not safe here.
1277      */
1278     while (!QTAILQ_EMPTY(&net_clients)) {
1279         nc = QTAILQ_FIRST(&net_clients);
1280         if (nc->info->type == NET_CLIENT_OPTIONS_KIND_NIC) {
1281             qemu_del_nic(qemu_get_nic(nc));
1282         } else {
1283             qemu_del_net_client(nc);
1284         }
1285     }
1286 
1287     qemu_del_vm_change_state_handler(net_change_state_entry);
1288 }
1289 
1290 void net_check_clients(void)
1291 {
1292     NetClientState *nc;
1293     int i;
1294 
1295     /* Don't warn about the default network setup that you get if
1296      * no command line -net or -netdev options are specified. There
1297      * are two cases that we would otherwise complain about:
1298      * (1) board doesn't support a NIC but the implicit "-net nic"
1299      * requested one
1300      * (2) CONFIG_SLIRP not set, in which case the implicit "-net nic"
1301      * sets up a nic that isn't connected to anything.
1302      */
1303     if (default_net) {
1304         return;
1305     }
1306 
1307     net_hub_check_clients();
1308 
1309     QTAILQ_FOREACH(nc, &net_clients, next) {
1310         if (!nc->peer) {
1311             fprintf(stderr, "Warning: %s %s has no peer\n",
1312                     nc->info->type == NET_CLIENT_OPTIONS_KIND_NIC ?
1313                     "nic" : "netdev", nc->name);
1314         }
1315     }
1316 
1317     /* Check that all NICs requested via -net nic actually got created.
1318      * NICs created via -device don't need to be checked here because
1319      * they are always instantiated.
1320      */
1321     for (i = 0; i < MAX_NICS; i++) {
1322         NICInfo *nd = &nd_table[i];
1323         if (nd->used && !nd->instantiated) {
1324             fprintf(stderr, "Warning: requested NIC (%s, model %s) "
1325                     "was not created (not supported by this machine?)\n",
1326                     nd->name ? nd->name : "anonymous",
1327                     nd->model ? nd->model : "unspecified");
1328         }
1329     }
1330 }
1331 
1332 static int net_init_client(QemuOpts *opts, void *dummy)
1333 {
1334     Error *local_err = NULL;
1335 
1336     net_client_init(opts, 0, &local_err);
1337     if (local_err) {
1338         error_report_err(local_err);
1339         return -1;
1340     }
1341 
1342     return 0;
1343 }
1344 
1345 static int net_init_netdev(QemuOpts *opts, void *dummy)
1346 {
1347     Error *local_err = NULL;
1348     int ret;
1349 
1350     ret = net_client_init(opts, 1, &local_err);
1351     if (local_err) {
1352         error_report_err(local_err);
1353         return -1;
1354     }
1355 
1356     return ret;
1357 }
1358 
1359 int net_init_clients(void)
1360 {
1361     QemuOptsList *net = qemu_find_opts("net");
1362 
1363     if (default_net) {
1364         /* if no clients, we use a default config */
1365         qemu_opts_set(net, NULL, "type", "nic", &error_abort);
1366 #ifdef CONFIG_SLIRP
1367         qemu_opts_set(net, NULL, "type", "user", &error_abort);
1368 #endif
1369     }
1370 
1371     net_change_state_entry =
1372         qemu_add_vm_change_state_handler(net_vm_change_state_handler, NULL);
1373 
1374     QTAILQ_INIT(&net_clients);
1375 
1376     if (qemu_opts_foreach(qemu_find_opts("netdev"), net_init_netdev, NULL, 1) == -1)
1377         return -1;
1378 
1379     if (qemu_opts_foreach(net, net_init_client, NULL, 1) == -1) {
1380         return -1;
1381     }
1382 
1383     return 0;
1384 }
1385 
1386 int net_client_parse(QemuOptsList *opts_list, const char *optarg)
1387 {
1388 #if defined(CONFIG_SLIRP)
1389     int ret;
1390     if (net_slirp_parse_legacy(opts_list, optarg, &ret)) {
1391         return ret;
1392     }
1393 #endif
1394 
1395     if (!qemu_opts_parse(opts_list, optarg, 1)) {
1396         return -1;
1397     }
1398 
1399     default_net = 0;
1400     return 0;
1401 }
1402 
1403 /* From FreeBSD */
1404 /* XXX: optimize */
1405 unsigned compute_mcast_idx(const uint8_t *ep)
1406 {
1407     uint32_t crc;
1408     int carry, i, j;
1409     uint8_t b;
1410 
1411     crc = 0xffffffff;
1412     for (i = 0; i < 6; i++) {
1413         b = *ep++;
1414         for (j = 0; j < 8; j++) {
1415             carry = ((crc & 0x80000000L) ? 1 : 0) ^ (b & 0x01);
1416             crc <<= 1;
1417             b >>= 1;
1418             if (carry) {
1419                 crc = ((crc ^ POLYNOMIAL) | carry);
1420             }
1421         }
1422     }
1423     return crc >> 26;
1424 }
1425 
1426 QemuOptsList qemu_netdev_opts = {
1427     .name = "netdev",
1428     .implied_opt_name = "type",
1429     .head = QTAILQ_HEAD_INITIALIZER(qemu_netdev_opts.head),
1430     .desc = {
1431         /*
1432          * no elements => accept any params
1433          * validation will happen later
1434          */
1435         { /* end of list */ }
1436     },
1437 };
1438 
1439 QemuOptsList qemu_net_opts = {
1440     .name = "net",
1441     .implied_opt_name = "type",
1442     .head = QTAILQ_HEAD_INITIALIZER(qemu_net_opts.head),
1443     .desc = {
1444         /*
1445          * no elements => accept any params
1446          * validation will happen later
1447          */
1448         { /* end of list */ }
1449     },
1450 };
1451