2 * This file is part of the Palacios Virtual Machine Monitor developed
3 * by the V3VEE Project with funding from the United States National
4 * Science Foundation and the Department of Energy.
6 * The V3VEE Project is a joint project between Northwestern University
7 * and the University of New Mexico. You can find out more at
10 * Copyright (c) 2010, Lei Xia <lxia@northwestern.edu>
11 * Copyright (c) 2009, Yuan Tang <ytang@northwestern.edu>
12 * Copyright (c) 2009, The V3VEE Project <http://www.v3vee.org>
13 * All rights reserved.
15 * Author: Lei Xia <lxia@northwestern.edu>
16 * Yuan Tang <ytang@northwestern.edu>
18 * This is free software. You are permitted to use,
19 * redistribute, and modify it as specified in the file "V3VEE_LICENSE".
22 #include <palacios/vmm_vnet.h>
23 #include <palacios/vm_guest_mem.h>
24 #include <palacios/vmm_lock.h>
25 #include <palacios/vmm_queue.h>
26 #include <palacios/vmm_sprintf.h>
27 #include <palacios/vmm_ethernet.h>
29 #ifndef CONFIG_DEBUG_VNET
31 #define PrintDebug(fmt, args...)
35 uint8_t dst_mac[ETH_ALEN];
36 uint8_t src_mac[ETH_ALEN];
37 uint16_t type; /* indicates layer 3 protocol type */
38 } __attribute__((packed));
43 uint8_t mac_addr[ETH_ALEN];
44 struct v3_vm_info * vm;
45 struct v3_vnet_dev_ops dev_ops;
49 uint8_t mode; //vmm_drivern or guest_drivern
51 struct list_head node;
52 } __attribute__((packed));
56 struct v3_vm_info * vm;
57 struct v3_vnet_bridge_ops brg_ops;
63 } __attribute__((packed));
67 struct vnet_route_info {
68 struct v3_vnet_route route_def;
70 struct vnet_dev * dst_dev;
71 struct vnet_dev * src_dev;
73 struct list_head node;
74 struct list_head match_node; // used for route matching
79 uint8_t hash_buf[VNET_HASH_SIZE];
82 struct vnet_route_info * routes[0];
83 } __attribute__((packed));
87 struct list_head routes;
88 struct list_head devs;
93 struct vnet_brg_dev *bridge;
97 struct hashtable * route_cache;
102 #ifdef CONFIG_DEBUG_VNET
103 static inline void mac_to_string(char * mac, char * buf) {
104 snprintf(buf, 100, "%d:%d:%d:%d:%d:%d",
105 mac[0], mac[1], mac[2],
106 mac[3], mac[4], mac[5]);
109 static void print_route(struct vnet_route_info * route){
112 mac_to_string(route->route_def.src_mac, str);
113 PrintDebug("Src Mac (%s), src_qual (%d)\n",
114 str, route->route_def.src_mac_qual);
115 mac_to_string(route->route_def.dst_mac, str);
116 PrintDebug("Dst Mac (%s), dst_qual (%d)\n",
117 str, route->route_def.dst_mac_qual);
118 PrintDebug("Src dev id (%d), src type (%d)",
119 route->route_def.src_id,
120 route->route_def.src_type);
121 PrintDebug("Dst dev id (%d), dst type (%d)\n",
122 route->route_def.dst_id,
123 route->route_def.dst_type);
124 if (route->route_def.dst_type == LINK_INTERFACE) {
125 PrintDebug("dst_dev (%p), dst_dev_id (%d), dst_dev_ops(%p), dst_dev_data (%p)\n",
127 route->dst_dev->dev_id,
128 (void *)&(route->dst_dev->dev_ops),
129 route->dst_dev->private_data);
133 static void dump_routes(){
134 struct vnet_route_info *route;
137 PrintDebug("\n========Dump routes starts ============\n");
138 list_for_each_entry(route, &(vnet_state.routes), node) {
139 PrintDebug("\nroute %d:\n", i++);
143 PrintDebug("\n========Dump routes end ============\n");
150 * A VNET packet is a packed struct with the hashed fields grouped together.
151 * This means we can generate the hash from an offset into the pkt struct
153 static inline uint_t hash_fn(addr_t hdr_ptr) {
154 uint8_t * hdr_buf = (uint8_t *)hdr_ptr;
156 return v3_hash_buffer(hdr_buf, VNET_HASH_SIZE);
159 static inline int hash_eq(addr_t key1, addr_t key2) {
160 return (memcmp((uint8_t *)key1, (uint8_t *)key2, VNET_HASH_SIZE) == 0);
163 static int add_route_to_cache(const struct v3_vnet_pkt * pkt, struct route_list * routes) {
164 memcpy(routes->hash_buf, pkt->hash_buf, VNET_HASH_SIZE);
166 if (v3_htable_insert(vnet_state.route_cache, (addr_t)routes->hash_buf, (addr_t)routes) == 0) {
167 PrintError("VNET/P Core: Failed to insert new route entry to the cache\n");
174 static int clear_hash_cache() {
175 v3_free_htable(vnet_state.route_cache, 1, 1);
176 vnet_state.route_cache = v3_create_htable(0, &hash_fn, &hash_eq);
181 static int look_into_cache(const struct v3_vnet_pkt * pkt, struct route_list ** routes) {
182 *routes = (struct route_list *)v3_htable_search(vnet_state.route_cache, (addr_t)(pkt->hash_buf));
188 static struct vnet_dev * dev_by_id(int idx) {
189 struct vnet_dev * dev = NULL;
191 list_for_each_entry(dev, &(vnet_state.devs), node) {
192 int dev_id = dev->dev_id;
201 static struct vnet_dev * dev_by_mac(uint8_t * mac) {
202 struct vnet_dev * dev = NULL;
204 list_for_each_entry(dev, &(vnet_state.devs), node) {
205 if (!memcmp(dev->mac_addr, mac, ETH_ALEN))
212 int v3_vnet_id_by_mac(char * mac){
213 struct vnet_dev *dev = dev_by_mac(mac);
222 int v3_vnet_add_route(struct v3_vnet_route route) {
223 struct vnet_route_info * new_route = NULL;
226 new_route = (struct vnet_route_info *)V3_Malloc(sizeof(struct vnet_route_info));
227 memset(new_route, 0, sizeof(struct vnet_route_info));
229 PrintDebug("VNET/P Core: add_route_entry: dst_id: %d, dst_type: %d\n",
230 route.dst_id, route.dst_type);
232 memcpy(new_route->route_def.src_mac, route.src_mac, ETH_ALEN);
233 memcpy(new_route->route_def.dst_mac, route.dst_mac, ETH_ALEN);
234 new_route->route_def.src_mac_qual = route.src_mac_qual;
235 new_route->route_def.dst_mac_qual = route.dst_mac_qual;
236 new_route->route_def.dst_type = route.dst_type;
237 new_route->route_def.src_type = route.src_type;
239 if(route.dst_id == -1){
240 if (new_route->route_def.dst_type == LINK_INTERFACE) {
241 new_route->dst_dev = dev_by_mac(route.dst_mac);
243 new_route->route_def.dst_id = new_route->dst_dev->dev_id;
245 new_route->route_def.dst_id = route.dst_id;
246 if (new_route->route_def.dst_type == LINK_INTERFACE) {
247 new_route->dst_dev = dev_by_id(new_route->route_def.dst_id);
251 if(route.src_id == -1){
252 if (new_route->route_def.src_type == LINK_INTERFACE) {
253 new_route->src_dev = dev_by_mac(route.src_mac);
255 new_route->route_def.src_id = new_route->src_dev->dev_id;
257 new_route->route_def.src_id = route.src_id;
258 if (new_route->route_def.src_type == LINK_INTERFACE) {
259 new_route->src_dev = dev_by_id(new_route->route_def.src_id);
263 flags = v3_lock_irqsave(vnet_state.lock);
265 list_add(&(new_route->node), &(vnet_state.routes));
268 v3_unlock_irqrestore(vnet_state.lock, flags);
271 #ifdef CONFIG_DEBUG_VNET
279 /* delete all route entries with specfied src or dst device id */
280 static void inline del_routes_by_dev(int dev_id){
281 struct vnet_route_info * route = NULL;
284 flags = v3_lock_irqsave(vnet_state.lock);
286 list_for_each_entry(route, &(vnet_state.routes), node) {
287 if((route->route_def.dst_type == LINK_INTERFACE &&
288 route->route_def.dst_id == dev_id) ||
289 (route->route_def.src_type == LINK_INTERFACE &&
290 route->route_def.src_id == dev_id)){
292 list_del(&(route->node));
293 list_del(&(route->match_node));
298 v3_unlock_irqrestore(vnet_state.lock, flags);
301 /* At the end allocate a route_list
302 * This list will be inserted into the cache so we don't need to free it
304 static struct route_list * match_route(const struct v3_vnet_pkt * pkt) {
305 struct vnet_route_info * route = NULL;
306 struct route_list * matches = NULL;
309 struct list_head match_list;
310 struct eth_hdr * hdr = (struct eth_hdr *)(pkt->data);
311 uint8_t src_type = pkt->src_type;
312 uint32_t src_link = pkt->src_id;
314 #ifdef CONFIG_DEBUG_VNET
319 mac_to_string(hdr->src_mac, src_str);
320 mac_to_string(hdr->dst_mac, dst_str);
321 PrintDebug("VNET/P Core: match_route. pkt: SRC(%s), DEST(%s)\n", src_str, dst_str);
325 INIT_LIST_HEAD(&match_list);
327 #define UPDATE_MATCHES(rank) do { \
328 if (max_rank < (rank)) { \
330 INIT_LIST_HEAD(&match_list); \
332 list_add(&(route->match_node), &match_list); \
334 } else if (max_rank == (rank)) { \
335 list_add(&(route->match_node), &match_list); \
341 list_for_each_entry(route, &(vnet_state.routes), node) {
342 struct v3_vnet_route * route_def = &(route->route_def);
344 // CHECK SOURCE TYPE HERE
345 if ( (route_def->src_type != LINK_ANY) &&
346 ( (route_def->src_type != src_type) ||
347 ( (route_def->src_id != src_link) &&
348 (route_def->src_id != (uint32_t)-1)))) {
353 if ((route_def->dst_mac_qual == MAC_ANY) &&
354 (route_def->src_mac_qual == MAC_ANY)) {
358 if (memcmp(route_def->src_mac, hdr->src_mac, 6) == 0) {
359 if (route_def->src_mac_qual != MAC_NOT) {
360 if (route_def->dst_mac_qual == MAC_ANY) {
362 } else if (route_def->dst_mac_qual != MAC_NOT &&
363 memcmp(route_def->dst_mac, hdr->dst_mac, 6) == 0) {
369 if (memcmp(route_def->dst_mac, hdr->dst_mac, 6) == 0) {
370 if (route_def->dst_mac_qual != MAC_NOT) {
371 if (route_def->src_mac_qual == MAC_ANY) {
373 } else if ((route_def->src_mac_qual != MAC_NOT) &&
374 (memcmp(route_def->src_mac, hdr->src_mac, 6) == 0)) {
380 if ((route_def->dst_mac_qual == MAC_NOT) &&
381 (memcmp(route_def->dst_mac, hdr->dst_mac, 6) != 0)) {
382 if (route_def->src_mac_qual == MAC_ANY) {
384 } else if ((route_def->src_mac_qual != MAC_NOT) &&
385 (memcmp(route_def->src_mac, hdr->src_mac, 6) == 0)) {
390 if ((route_def->src_mac_qual == MAC_NOT) &&
391 (memcmp(route_def->src_mac, hdr->src_mac, 6) != 0)) {
392 if (route_def->dst_mac_qual == MAC_ANY) {
394 } else if ((route_def->dst_mac_qual != MAC_NOT) &&
395 (memcmp(route_def->dst_mac, hdr->dst_mac, 6) == 0)) {
401 if ( (memcmp(route_def->src_mac, hdr->src_mac, 6) == 0) &&
402 (route_def->dst_mac_qual == MAC_NONE)) {
407 PrintDebug("VNET/P Core: match_route: Matches=%d\n", num_matches);
409 if (num_matches == 0) {
413 matches = (struct route_list *)V3_Malloc(sizeof(struct route_list) +
414 (sizeof(struct vnet_route_info *) * num_matches));
416 matches->num_routes = num_matches;
420 list_for_each_entry(route, &match_list, match_node) {
421 matches->routes[i++] = route;
429 int v3_vnet_send_pkt(struct v3_vnet_pkt * pkt, void * private_data) {
430 struct route_list * matched_routes = NULL;
434 #ifdef CONFIG_DEBUG_VNET
436 int cpu = V3_Get_CPU();
437 PrintDebug("VNET/P Core: cpu %d: pkt (size %d, src_id:%d, src_type: %d, dst_id: %d, dst_type: %d)\n",
438 cpu, pkt->size, pkt->src_id,
439 pkt->src_type, pkt->dst_id, pkt->dst_type);
440 //v3_hexdump(pkt->data, pkt->size, NULL, 0);
444 flags = v3_lock_irqsave(vnet_state.lock);
446 look_into_cache(pkt, &matched_routes);
447 if (matched_routes == NULL) {
448 PrintDebug("VNET/P Core: send pkt Looking into routing table\n");
450 matched_routes = match_route(pkt);
452 if (matched_routes) {
453 add_route_to_cache(pkt, matched_routes);
455 PrintDebug("VNET/P Core: Could not find route for packet... discards packet\n");
456 v3_unlock_irqrestore(vnet_state.lock, flags);
457 return 0; /* do we return -1 here?*/
461 v3_unlock_irqrestore(vnet_state.lock, flags);
463 PrintDebug("VNET/P Core: send pkt route matches %d\n", matched_routes->num_routes);
465 for (i = 0; i < matched_routes->num_routes; i++) {
466 struct vnet_route_info * route = matched_routes->routes[i];
468 if (route->route_def.dst_type == LINK_EDGE) {
469 struct vnet_brg_dev *bridge = vnet_state.bridge;
470 pkt->dst_type = LINK_EDGE;
471 pkt->dst_id = route->route_def.dst_id;
473 if (bridge == NULL || (bridge->active == 0)) {
474 PrintDebug("VNET/P Core: No active bridge to sent data to\n");
478 if(bridge->brg_ops.input(bridge->vm, pkt, bridge->private_data) < 0){
479 PrintDebug("VNET/P Core: Packet not sent properly to bridge\n");
482 } else if (route->route_def.dst_type == LINK_INTERFACE) {
483 if (route->dst_dev == NULL || route->dst_dev->active == 0){
484 PrintDebug("VNET/P Core: No active device to sent data to\n");
488 if(route->dst_dev->dev_ops.input(route->dst_dev->vm, pkt, route->dst_dev->private_data) < 0) {
489 PrintDebug("VNET/P Core: Packet not sent properly\n");
493 PrintError("VNET/P Core: Wrong dst type\n");
500 int v3_vnet_add_dev(struct v3_vm_info * vm, uint8_t * mac,
501 struct v3_vnet_dev_ops *ops,
503 struct vnet_dev * new_dev = NULL;
506 new_dev = (struct vnet_dev *)V3_Malloc(sizeof(struct vnet_dev));
508 if (new_dev == NULL) {
509 PrintError("Malloc fails\n");
513 memcpy(new_dev->mac_addr, mac, 6);
514 new_dev->dev_ops.input = ops->input;
515 new_dev->dev_ops.poll = ops->poll;
516 new_dev->private_data = priv_data;
520 new_dev->mode = GUEST_DRIVERN;
522 flags = v3_lock_irqsave(vnet_state.lock);
524 if (dev_by_mac(mac) == NULL) {
525 list_add(&(new_dev->node), &(vnet_state.devs));
526 new_dev->dev_id = ++vnet_state.num_devs;
529 v3_unlock_irqrestore(vnet_state.lock, flags);
531 /* if the device was found previosly the id should still be 0 */
532 if (new_dev->dev_id == 0) {
533 PrintError("VNET/P Core: Device Already exists\n");
537 PrintDebug("VNET/P Core: Add Device: dev_id %d\n", new_dev->dev_id);
539 return new_dev->dev_id;
544 int v3_vnet_del_dev(int dev_id){
545 struct vnet_dev * dev = NULL;
548 flags = v3_lock_irqsave(vnet_state.lock);
550 dev = dev_by_id(dev_id);
552 list_del(&(dev->node));
553 del_routes_by_dev(dev_id);
556 v3_unlock_irqrestore(vnet_state.lock, flags);
560 PrintDebug("VNET/P Core: Remove Device: dev_id %d\n", dev_id);
566 static void free_devices(){
567 struct vnet_dev * dev = NULL;
569 list_for_each_entry(dev, &(vnet_state.devs), node) {
570 list_del(&(dev->node));
575 static void free_routes(){
576 struct vnet_route_info * route = NULL;
578 list_for_each_entry(route, &(vnet_state.routes), node) {
579 list_del(&(route->node));
580 list_del(&(route->match_node));
586 void v3_vnet_poll(struct v3_vm_info * vm){
587 struct vnet_dev * dev = NULL;
588 struct vnet_brg_dev *bridge = vnet_state.bridge;
590 list_for_each_entry(dev, &(vnet_state.devs), node) {
591 if(dev->mode == VMM_DRIVERN &&
595 dev->dev_ops.poll(vm, dev->private_data);
599 if (bridge != NULL &&
601 bridge->mode == VMM_DRIVERN) {
603 bridge->brg_ops.poll(bridge->vm, bridge->private_data);
608 int v3_vnet_add_bridge(struct v3_vm_info * vm,
609 struct v3_vnet_bridge_ops * ops,
614 struct vnet_brg_dev * tmp_bridge = NULL;
616 flags = v3_lock_irqsave(vnet_state.lock);
618 if (vnet_state.bridge == NULL) {
620 vnet_state.bridge = (void *)1;
623 v3_unlock_irqrestore(vnet_state.lock, flags);
625 if (bridge_free == 0) {
626 PrintError("VNET/P Core: Bridge already set\n");
630 tmp_bridge = (struct vnet_brg_dev *)V3_Malloc(sizeof(struct vnet_brg_dev));
632 if (tmp_bridge == NULL) {
633 PrintError("Malloc Fails\n");
634 vnet_state.bridge = NULL;
639 tmp_bridge->brg_ops.input = ops->input;
640 tmp_bridge->brg_ops.poll = ops->poll;
641 tmp_bridge->private_data = priv_data;
642 tmp_bridge->active = 1;
643 tmp_bridge->mode = GUEST_DRIVERN;
644 tmp_bridge->type = type;
646 /* make this atomic to avoid possible race conditions */
647 flags = v3_lock_irqsave(vnet_state.lock);
648 vnet_state.bridge = tmp_bridge;
649 v3_unlock_irqrestore(vnet_state.lock, flags);
656 memset(&vnet_state, 0, sizeof(vnet_state));
658 INIT_LIST_HEAD(&(vnet_state.routes));
659 INIT_LIST_HEAD(&(vnet_state.devs));
661 vnet_state.num_devs = 0;
662 vnet_state.num_routes = 0;
664 if (v3_lock_init(&(vnet_state.lock)) == -1){
665 PrintError("VNET/P Core: Fails to initiate lock\n");
668 vnet_state.route_cache = v3_create_htable(0, &hash_fn, &hash_eq);
670 if (vnet_state.route_cache == NULL) {
671 PrintError("VNET/P Core: Fails to initiate route cache\n");
675 PrintDebug("VNET/P Core is initiated\n");
681 void v3_deinit_vnet(){
683 v3_lock_deinit(&(vnet_state.lock));
688 v3_free_htable(vnet_state.route_cache, 1, 1);
689 V3_Free(vnet_state.bridge);