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(uint8_t * 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))
213 int v3_vnet_add_route(struct v3_vnet_route route) {
214 struct vnet_route_info * new_route = NULL;
217 new_route = (struct vnet_route_info *)V3_Malloc(sizeof(struct vnet_route_info));
218 memset(new_route, 0, sizeof(struct vnet_route_info));
220 PrintDebug("VNET/P Core: add_route_entry: dst_id: %d, dst_type: %d\n",
221 route.dst_id, route.dst_type);
223 memcpy(new_route->route_def.src_mac, route.src_mac, ETH_ALEN);
224 memcpy(new_route->route_def.dst_mac, route.dst_mac, ETH_ALEN);
225 new_route->route_def.src_mac_qual = route.src_mac_qual;
226 new_route->route_def.dst_mac_qual = route.dst_mac_qual;
227 new_route->route_def.dst_type = route.dst_type;
228 new_route->route_def.src_type = route.src_type;
230 if(route.dst_id == -1){
231 if (new_route->route_def.dst_type == LINK_INTERFACE) {
232 new_route->dst_dev = dev_by_mac(route.dst_mac);
234 new_route->route_def.dst_id = new_route->dst_dev->dev_id;
236 new_route->route_def.dst_id = route.dst_id;
237 if (new_route->route_def.dst_type == LINK_INTERFACE) {
238 new_route->dst_dev = dev_by_id(new_route->route_def.dst_id);
242 if(route.src_id == -1){
243 if (new_route->route_def.src_type == LINK_INTERFACE) {
244 new_route->src_dev = dev_by_mac(route.src_mac);
246 new_route->route_def.src_id = new_route->src_dev->dev_id;
248 new_route->route_def.src_id = route.src_id;
249 if (new_route->route_def.src_type == LINK_INTERFACE) {
250 new_route->src_dev = dev_by_id(new_route->route_def.src_id);
254 flags = v3_lock_irqsave(vnet_state.lock);
256 list_add(&(new_route->node), &(vnet_state.routes));
259 v3_unlock_irqrestore(vnet_state.lock, flags);
262 #ifdef CONFIG_DEBUG_VNET
270 /* delete all route entries with specfied src or dst device id */
271 static void inline del_routes_by_dev(int dev_id){
272 struct vnet_route_info * route = NULL;
275 flags = v3_lock_irqsave(vnet_state.lock);
277 list_for_each_entry(route, &(vnet_state.routes), node) {
278 if((route->route_def.dst_type == LINK_INTERFACE &&
279 route->route_def.dst_id == dev_id) ||
280 (route->route_def.src_type == LINK_INTERFACE &&
281 route->route_def.src_id == dev_id)){
283 list_del(&(route->node));
284 list_del(&(route->match_node));
289 v3_unlock_irqrestore(vnet_state.lock, flags);
292 /* At the end allocate a route_list
293 * This list will be inserted into the cache so we don't need to free it
295 static struct route_list * match_route(const struct v3_vnet_pkt * pkt) {
296 struct vnet_route_info * route = NULL;
297 struct route_list * matches = NULL;
300 struct list_head match_list;
301 struct eth_hdr * hdr = (struct eth_hdr *)(pkt->data);
302 uint8_t src_type = pkt->src_type;
303 uint32_t src_link = pkt->src_id;
305 #ifdef CONFIG_DEBUG_VNET
310 mac_to_string(hdr->src_mac, src_str);
311 mac_to_string(hdr->dst_mac, dst_str);
312 PrintDebug("VNET/P Core: match_route. pkt: SRC(%s), DEST(%s)\n", src_str, dst_str);
316 INIT_LIST_HEAD(&match_list);
318 #define UPDATE_MATCHES(rank) do { \
319 if (max_rank < (rank)) { \
321 INIT_LIST_HEAD(&match_list); \
323 list_add(&(route->match_node), &match_list); \
325 } else if (max_rank == (rank)) { \
326 list_add(&(route->match_node), &match_list); \
332 list_for_each_entry(route, &(vnet_state.routes), node) {
333 struct v3_vnet_route * route_def = &(route->route_def);
335 // CHECK SOURCE TYPE HERE
336 if ( (route_def->src_type != LINK_ANY) &&
337 ( (route_def->src_type != src_type) ||
338 ( (route_def->src_id != src_link) &&
339 (route_def->src_id != (uint32_t)-1)))) {
344 if ((route_def->dst_mac_qual == MAC_ANY) &&
345 (route_def->src_mac_qual == MAC_ANY)) {
349 if (memcmp(route_def->src_mac, hdr->src_mac, 6) == 0) {
350 if (route_def->src_mac_qual != MAC_NOT) {
351 if (route_def->dst_mac_qual == MAC_ANY) {
353 } else if (route_def->dst_mac_qual != MAC_NOT &&
354 memcmp(route_def->dst_mac, hdr->dst_mac, 6) == 0) {
360 if (memcmp(route_def->dst_mac, hdr->dst_mac, 6) == 0) {
361 if (route_def->dst_mac_qual != MAC_NOT) {
362 if (route_def->src_mac_qual == MAC_ANY) {
364 } else if ((route_def->src_mac_qual != MAC_NOT) &&
365 (memcmp(route_def->src_mac, hdr->src_mac, 6) == 0)) {
371 if ((route_def->dst_mac_qual == MAC_NOT) &&
372 (memcmp(route_def->dst_mac, hdr->dst_mac, 6) != 0)) {
373 if (route_def->src_mac_qual == MAC_ANY) {
375 } else if ((route_def->src_mac_qual != MAC_NOT) &&
376 (memcmp(route_def->src_mac, hdr->src_mac, 6) == 0)) {
381 if ((route_def->src_mac_qual == MAC_NOT) &&
382 (memcmp(route_def->src_mac, hdr->src_mac, 6) != 0)) {
383 if (route_def->dst_mac_qual == MAC_ANY) {
385 } else if ((route_def->dst_mac_qual != MAC_NOT) &&
386 (memcmp(route_def->dst_mac, hdr->dst_mac, 6) == 0)) {
392 if ( (memcmp(route_def->src_mac, hdr->src_mac, 6) == 0) &&
393 (route_def->dst_mac_qual == MAC_NONE)) {
398 PrintDebug("VNET/P Core: match_route: Matches=%d\n", num_matches);
400 if (num_matches == 0) {
404 matches = (struct route_list *)V3_Malloc(sizeof(struct route_list) +
405 (sizeof(struct vnet_route_info *) * num_matches));
407 matches->num_routes = num_matches;
411 list_for_each_entry(route, &match_list, match_node) {
412 matches->routes[i++] = route;
420 int v3_vnet_send_pkt(struct v3_vnet_pkt * pkt, void * private_data) {
421 struct route_list * matched_routes = NULL;
425 #ifdef CONFIG_DEBUG_VNET
427 int cpu = V3_Get_CPU();
428 PrintDebug("VNET/P Core: cpu %d: pkt (size %d, src_id:%d, src_type: %d, dst_id: %d, dst_type: %d)\n",
429 cpu, pkt->size, pkt->src_id,
430 pkt->src_type, pkt->dst_id, pkt->dst_type);
431 //v3_hexdump(pkt->data, pkt->size, NULL, 0);
435 flags = v3_lock_irqsave(vnet_state.lock);
437 look_into_cache(pkt, &matched_routes);
438 if (matched_routes == NULL) {
439 PrintDebug("VNET/P Core: send pkt Looking into routing table\n");
441 matched_routes = match_route(pkt);
443 if (matched_routes) {
444 add_route_to_cache(pkt, matched_routes);
446 PrintDebug("VNET/P Core: Could not find route for packet... discards packet\n");
447 v3_unlock_irqrestore(vnet_state.lock, flags);
448 return 0; /* do we return -1 here?*/
452 v3_unlock_irqrestore(vnet_state.lock, flags);
454 PrintDebug("VNET/P Core: send pkt route matches %d\n", matched_routes->num_routes);
456 for (i = 0; i < matched_routes->num_routes; i++) {
457 struct vnet_route_info * route = matched_routes->routes[i];
459 if (route->route_def.dst_type == LINK_EDGE) {
460 struct vnet_brg_dev *bridge = vnet_state.bridge;
461 pkt->dst_type = LINK_EDGE;
462 pkt->dst_id = route->route_def.dst_id;
464 if (bridge == NULL || (bridge->active == 0)) {
465 PrintDebug("VNET/P Core: No active bridge to sent data to\n");
469 if(bridge->brg_ops.input(bridge->vm, pkt, bridge->private_data) < 0){
470 PrintDebug("VNET/P Core: Packet not sent properly to bridge\n");
473 } else if (route->route_def.dst_type == LINK_INTERFACE) {
474 if (route->dst_dev == NULL || route->dst_dev->active == 0){
475 PrintDebug("VNET/P Core: No active device to sent data to\n");
479 if(route->dst_dev->dev_ops.input(route->dst_dev->vm, pkt, route->dst_dev->private_data) < 0) {
480 PrintDebug("VNET/P Core: Packet not sent properly\n");
484 PrintError("VNET/P Core: Wrong dst type\n");
491 int v3_vnet_add_dev(struct v3_vm_info * vm, uint8_t * mac,
492 struct v3_vnet_dev_ops *ops,
494 struct vnet_dev * new_dev = NULL;
497 new_dev = (struct vnet_dev *)V3_Malloc(sizeof(struct vnet_dev));
499 if (new_dev == NULL) {
500 PrintError("Malloc fails\n");
504 memcpy(new_dev->mac_addr, mac, 6);
505 new_dev->dev_ops.input = ops->input;
506 new_dev->dev_ops.poll = ops->poll;
507 new_dev->private_data = priv_data;
511 new_dev->mode = GUEST_DRIVERN;
513 flags = v3_lock_irqsave(vnet_state.lock);
515 if (dev_by_mac(mac) == NULL) {
516 list_add(&(new_dev->node), &(vnet_state.devs));
517 new_dev->dev_id = ++vnet_state.num_devs;
520 v3_unlock_irqrestore(vnet_state.lock, flags);
522 /* if the device was found previosly the id should still be 0 */
523 if (new_dev->dev_id == 0) {
524 PrintError("VNET/P Core: Device Already exists\n");
528 PrintDebug("VNET/P Core: Add Device: dev_id %d\n", new_dev->dev_id);
530 return new_dev->dev_id;
535 int v3_vnet_del_dev(int dev_id){
536 struct vnet_dev * dev = NULL;
539 flags = v3_lock_irqsave(vnet_state.lock);
541 dev = dev_by_id(dev_id);
543 list_del(&(dev->node));
544 del_routes_by_dev(dev_id);
547 v3_unlock_irqrestore(vnet_state.lock, flags);
551 PrintDebug("VNET/P Core: Remove Device: dev_id %d\n", dev_id);
557 static void free_devices(){
558 struct vnet_dev * dev = NULL;
560 list_for_each_entry(dev, &(vnet_state.devs), node) {
561 list_del(&(dev->node));
566 static void free_routes(){
567 struct vnet_route_info * route = NULL;
569 list_for_each_entry(route, &(vnet_state.routes), node) {
570 list_del(&(route->node));
571 list_del(&(route->match_node));
576 /* TODO: Round-bin or ?? */
577 void v3_vnet_poll(struct v3_vm_info * vm){
578 struct vnet_dev * dev = NULL;
579 struct vnet_brg_dev *bridge = vnet_state.bridge;
581 list_for_each_entry(dev, &(vnet_state.devs), node) {
582 if(dev->mode == VMM_DRIVERN &&
586 dev->dev_ops.poll(vm, dev->private_data);
590 if (bridge != NULL &&
592 bridge->mode == VMM_DRIVERN) {
594 bridge->brg_ops.poll(bridge->vm, bridge->private_data);
599 int v3_vnet_add_bridge(struct v3_vm_info * vm,
600 struct v3_vnet_bridge_ops * ops,
605 struct vnet_brg_dev * tmp_bridge = NULL;
607 flags = v3_lock_irqsave(vnet_state.lock);
609 if (vnet_state.bridge == NULL) {
611 vnet_state.bridge = (void *)1;
614 v3_unlock_irqrestore(vnet_state.lock, flags);
616 if (bridge_free == 0) {
617 PrintError("VNET/P Core: Bridge already set\n");
621 tmp_bridge = (struct vnet_brg_dev *)V3_Malloc(sizeof(struct vnet_brg_dev));
623 if (tmp_bridge == NULL) {
624 PrintError("Malloc Fails\n");
625 vnet_state.bridge = NULL;
630 tmp_bridge->brg_ops.input = ops->input;
631 tmp_bridge->brg_ops.poll = ops->poll;
632 tmp_bridge->private_data = priv_data;
633 tmp_bridge->active = 1;
634 tmp_bridge->mode = GUEST_DRIVERN;
635 tmp_bridge->type = type;
637 /* make this atomic to avoid possible race conditions */
638 flags = v3_lock_irqsave(vnet_state.lock);
639 vnet_state.bridge = tmp_bridge;
640 v3_unlock_irqrestore(vnet_state.lock, flags);
647 memset(&vnet_state, 0, sizeof(vnet_state));
649 INIT_LIST_HEAD(&(vnet_state.routes));
650 INIT_LIST_HEAD(&(vnet_state.devs));
652 vnet_state.num_devs = 0;
653 vnet_state.num_routes = 0;
655 if (v3_lock_init(&(vnet_state.lock)) == -1){
656 PrintError("VNET/P Core: Fails to initiate lock\n");
659 vnet_state.route_cache = v3_create_htable(0, &hash_fn, &hash_eq);
661 if (vnet_state.route_cache == NULL) {
662 PrintError("VNET/P Core: Fails to initiate route cache\n");
666 PrintDebug("VNET/P Core is initiated\n");
672 void v3_deinit_vnet(){
674 v3_lock_deinit(&(vnet_state.lock));
679 v3_free_htable(vnet_state.route_cache, 1, 1);
680 V3_Free(vnet_state.bridge);