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 nic_poll_type_t mode; /*vmm_drivern or guest_drivern */
51 uint64_t bytes_tx, bytes_rx;
52 uint32_t pkts_tx, pkt_rx;
54 struct list_head node;
55 } __attribute__((packed));
59 struct v3_vm_info * vm;
60 struct v3_vnet_bridge_ops brg_ops;
66 } __attribute__((packed));
70 struct vnet_route_info {
71 struct v3_vnet_route route_def;
73 struct vnet_dev * dst_dev;
74 struct vnet_dev * src_dev;
76 struct list_head node;
77 struct list_head match_node; // used for route matching
82 uint8_t hash_buf[VNET_HASH_SIZE];
85 struct vnet_route_info * routes[0];
86 } __attribute__((packed));
90 struct list_head routes;
91 struct list_head devs;
96 struct vnet_brg_dev * bridge;
99 struct vnet_stat stats;
101 struct hashtable * route_cache;
106 #ifdef CONFIG_DEBUG_VNET
107 static inline void mac_to_string(uint8_t * mac, char * buf) {
108 snprintf(buf, 100, "%2x:%2x:%2x:%2x:%2x:%2x",
109 mac[0], mac[1], mac[2],
110 mac[3], mac[4], mac[5]);
113 static void print_route(struct v3_vnet_route * route){
116 mac_to_string(route->src_mac, str);
117 PrintDebug("Src Mac (%s), src_qual (%d)\n",
118 str, route->src_mac_qual);
119 mac_to_string(route->dst_mac, str);
120 PrintDebug("Dst Mac (%s), dst_qual (%d)\n",
121 str, route->dst_mac_qual);
122 PrintDebug("Src dev id (%d), src type (%d)",
125 PrintDebug("Dst dev id (%d), dst type (%d)\n",
130 static void dump_routes(){
131 struct vnet_route_info *route;
134 PrintDebug("\n========Dump routes starts ============\n");
135 list_for_each_entry(route, &(vnet_state.routes), node) {
136 PrintDebug("\nroute %d:\n", i++);
138 print_route(&(route->route_def));
139 if (route->route_def.dst_type == LINK_INTERFACE) {
140 PrintDebug("dst_dev (%p), dst_dev_id (%d), dst_dev_ops(%p), dst_dev_data (%p)\n",
142 route->dst_dev->dev_id,
143 (void *)&(route->dst_dev->dev_ops),
144 route->dst_dev->private_data);
148 PrintDebug("\n========Dump routes end ============\n");
155 * A VNET packet is a packed struct with the hashed fields grouped together.
156 * This means we can generate the hash from an offset into the pkt struct
158 static inline uint_t hash_fn(addr_t hdr_ptr) {
159 uint8_t * hdr_buf = (uint8_t *)hdr_ptr;
161 return v3_hash_buffer(hdr_buf, VNET_HASH_SIZE);
164 static inline int hash_eq(addr_t key1, addr_t key2) {
165 return (memcmp((uint8_t *)key1, (uint8_t *)key2, VNET_HASH_SIZE) == 0);
168 static int add_route_to_cache(const struct v3_vnet_pkt * pkt, struct route_list * routes) {
169 memcpy(routes->hash_buf, pkt->hash_buf, VNET_HASH_SIZE);
171 if (v3_htable_insert(vnet_state.route_cache, (addr_t)routes->hash_buf, (addr_t)routes) == 0) {
172 PrintError("VNET/P Core: Failed to insert new route entry to the cache\n");
179 static int clear_hash_cache() {
180 v3_free_htable(vnet_state.route_cache, 1, 1);
181 vnet_state.route_cache = v3_create_htable(0, &hash_fn, &hash_eq);
186 static int look_into_cache(const struct v3_vnet_pkt * pkt, struct route_list ** routes) {
187 *routes = (struct route_list *)v3_htable_search(vnet_state.route_cache, (addr_t)(pkt->hash_buf));
193 static struct vnet_dev * dev_by_id(int idx) {
194 struct vnet_dev * dev = NULL;
196 list_for_each_entry(dev, &(vnet_state.devs), node) {
197 int dev_id = dev->dev_id;
206 static struct vnet_dev * dev_by_mac(uint8_t * mac) {
207 struct vnet_dev * dev = NULL;
209 list_for_each_entry(dev, &(vnet_state.devs), node) {
210 if (!compare_ethaddr(dev->mac_addr, mac)){
219 int v3_vnet_find_dev(uint8_t * mac) {
220 struct vnet_dev * dev = NULL;
222 dev = dev_by_mac(mac);
232 int v3_vnet_add_route(struct v3_vnet_route route) {
233 struct vnet_route_info * new_route = NULL;
236 new_route = (struct vnet_route_info *)V3_Malloc(sizeof(struct vnet_route_info));
237 memset(new_route, 0, sizeof(struct vnet_route_info));
239 #ifdef CONFIG_DEBUG_VNET
240 PrintDebug("VNET/P Core: add_route_entry:\n");
244 memcpy(new_route->route_def.src_mac, route.src_mac, ETH_ALEN);
245 memcpy(new_route->route_def.dst_mac, route.dst_mac, ETH_ALEN);
246 new_route->route_def.src_mac_qual = route.src_mac_qual;
247 new_route->route_def.dst_mac_qual = route.dst_mac_qual;
248 new_route->route_def.dst_type = route.dst_type;
249 new_route->route_def.src_type = route.src_type;
250 new_route->route_def.src_id = route.src_id;
251 new_route->route_def.dst_id = route.dst_id;
253 if (new_route->route_def.dst_type == LINK_INTERFACE) {
254 new_route->dst_dev = dev_by_id(new_route->route_def.dst_id);
257 if (new_route->route_def.src_type == LINK_INTERFACE) {
258 new_route->src_dev = dev_by_id(new_route->route_def.src_id);
262 flags = v3_lock_irqsave(vnet_state.lock);
264 list_add(&(new_route->node), &(vnet_state.routes));
267 v3_unlock_irqrestore(vnet_state.lock, flags);
270 #ifdef CONFIG_DEBUG_VNET
278 /* delete all route entries with specfied src or dst device id */
279 static void inline del_routes_by_dev(int dev_id){
280 struct vnet_route_info * route = NULL;
283 flags = v3_lock_irqsave(vnet_state.lock);
285 list_for_each_entry(route, &(vnet_state.routes), node) {
286 if((route->route_def.dst_type == LINK_INTERFACE &&
287 route->route_def.dst_id == dev_id) ||
288 (route->route_def.src_type == LINK_INTERFACE &&
289 route->route_def.src_id == dev_id)){
291 list_del(&(route->node));
292 list_del(&(route->match_node));
297 v3_unlock_irqrestore(vnet_state.lock, flags);
300 /* At the end allocate a route_list
301 * This list will be inserted into the cache so we don't need to free it
303 static struct route_list * match_route(const struct v3_vnet_pkt * pkt) {
304 struct vnet_route_info * route = NULL;
305 struct route_list * matches = NULL;
308 struct list_head match_list;
309 struct eth_hdr * hdr = (struct eth_hdr *)(pkt->data);
310 // uint8_t src_type = pkt->src_type;
311 // uint32_t src_link = pkt->src_id;
313 #ifdef CONFIG_DEBUG_VNET
318 mac_to_string(hdr->src_mac, src_str);
319 mac_to_string(hdr->dst_mac, dst_str);
320 PrintDebug("VNET/P Core: match_route. pkt: SRC(%s), DEST(%s)\n", src_str, dst_str);
324 INIT_LIST_HEAD(&match_list);
326 #define UPDATE_MATCHES(rank) do { \
327 if (max_rank < (rank)) { \
329 INIT_LIST_HEAD(&match_list); \
331 list_add(&(route->match_node), &match_list); \
333 } else if (max_rank == (rank)) { \
334 list_add(&(route->match_node), &match_list); \
340 list_for_each_entry(route, &(vnet_state.routes), node) {
341 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 != -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 vnet_state.stats.rx_bytes += pkt->size;
447 vnet_state.stats.rx_pkts++;
449 look_into_cache(pkt, &matched_routes);
450 if (matched_routes == NULL) {
451 PrintDebug("VNET/P Core: send pkt Looking into routing table\n");
453 matched_routes = match_route(pkt);
455 if (matched_routes) {
456 add_route_to_cache(pkt, matched_routes);
458 PrintDebug("VNET/P Core: Could not find route for packet... discards packet\n");
459 v3_unlock_irqrestore(vnet_state.lock, flags);
460 return 0; /* do we return -1 here?*/
464 v3_unlock_irqrestore(vnet_state.lock, flags);
466 PrintDebug("VNET/P Core: send pkt route matches %d\n", matched_routes->num_routes);
468 for (i = 0; i < matched_routes->num_routes; i++) {
469 struct vnet_route_info * route = matched_routes->routes[i];
471 if (route->route_def.dst_type == LINK_EDGE) {
472 struct vnet_brg_dev *bridge = vnet_state.bridge;
473 pkt->dst_type = LINK_EDGE;
474 pkt->dst_id = route->route_def.dst_id;
476 if (bridge == NULL || (bridge->active == 0)) {
477 PrintDebug("VNET/P Core: No active bridge to sent data to\n");
481 if(bridge->brg_ops.input(bridge->vm, pkt, bridge->private_data) < 0){
482 PrintDebug("VNET/P Core: Packet not sent properly to bridge\n");
485 vnet_state.stats.tx_bytes += pkt->size;
486 vnet_state.stats.tx_pkts ++;
487 } else if (route->route_def.dst_type == LINK_INTERFACE) {
488 if (route->dst_dev == NULL || route->dst_dev->active == 0){
489 PrintDebug("VNET/P Core: No active device to sent data to\n");
493 if(route->dst_dev->dev_ops.input(route->dst_dev->vm, pkt, route->dst_dev->private_data) < 0) {
494 PrintDebug("VNET/P Core: Packet not sent properly\n");
497 vnet_state.stats.tx_bytes += pkt->size;
498 vnet_state.stats.tx_pkts ++;
500 PrintError("VNET/P Core: Wrong dst type\n");
507 int v3_vnet_add_dev(struct v3_vm_info * vm, uint8_t * mac,
508 struct v3_vnet_dev_ops *ops,
510 struct vnet_dev * new_dev = NULL;
513 new_dev = (struct vnet_dev *)V3_Malloc(sizeof(struct vnet_dev));
515 if (new_dev == NULL) {
516 PrintError("Malloc fails\n");
520 memcpy(new_dev->mac_addr, mac, 6);
521 new_dev->dev_ops.input = ops->input;
522 new_dev->dev_ops.poll = ops->poll;
523 new_dev->private_data = priv_data;
527 new_dev->mode = GUEST_DRIVERN;
529 flags = v3_lock_irqsave(vnet_state.lock);
531 if (dev_by_mac(mac) == NULL) {
532 list_add(&(new_dev->node), &(vnet_state.devs));
533 new_dev->dev_id = ++vnet_state.num_devs;
536 v3_unlock_irqrestore(vnet_state.lock, flags);
538 /* if the device was found previosly the id should still be 0 */
539 if (new_dev->dev_id == 0) {
540 PrintError("VNET/P Core: Device Already exists\n");
544 PrintDebug("VNET/P Core: Add Device: dev_id %d\n", new_dev->dev_id);
546 return new_dev->dev_id;
551 int v3_vnet_del_dev(int dev_id){
552 struct vnet_dev * dev = NULL;
555 flags = v3_lock_irqsave(vnet_state.lock);
557 dev = dev_by_id(dev_id);
559 list_del(&(dev->node));
560 del_routes_by_dev(dev_id);
563 v3_unlock_irqrestore(vnet_state.lock, flags);
567 PrintDebug("VNET/P Core: Remove Device: dev_id %d\n", dev_id);
572 int v3_vnet_stat(struct vnet_stat * stats){
574 stats->rx_bytes = vnet_state.stats.rx_bytes;
575 stats->rx_pkts = vnet_state.stats.rx_pkts;
576 stats->tx_bytes = vnet_state.stats.tx_bytes;
577 stats->tx_pkts = vnet_state.stats.tx_pkts;
582 static void free_devices(){
583 struct vnet_dev * dev = NULL;
585 list_for_each_entry(dev, &(vnet_state.devs), node) {
586 list_del(&(dev->node));
591 static void free_routes(){
592 struct vnet_route_info * route = NULL;
594 list_for_each_entry(route, &(vnet_state.routes), node) {
595 list_del(&(route->node));
596 list_del(&(route->match_node));
601 int v3_vnet_add_bridge(struct v3_vm_info * vm,
602 struct v3_vnet_bridge_ops * ops,
607 struct vnet_brg_dev * tmp_bridge = NULL;
609 flags = v3_lock_irqsave(vnet_state.lock);
611 if (vnet_state.bridge == NULL) {
613 vnet_state.bridge = (void *)1;
616 v3_unlock_irqrestore(vnet_state.lock, flags);
618 if (bridge_free == 0) {
619 PrintError("VNET/P Core: Bridge already set\n");
623 tmp_bridge = (struct vnet_brg_dev *)V3_Malloc(sizeof(struct vnet_brg_dev));
625 if (tmp_bridge == NULL) {
626 PrintError("Malloc Fails\n");
627 vnet_state.bridge = NULL;
632 tmp_bridge->brg_ops.input = ops->input;
633 tmp_bridge->brg_ops.poll = ops->poll;
634 tmp_bridge->private_data = priv_data;
635 tmp_bridge->active = 1;
636 tmp_bridge->mode = GUEST_DRIVERN;
637 tmp_bridge->type = type;
639 /* make this atomic to avoid possible race conditions */
640 flags = v3_lock_irqsave(vnet_state.lock);
641 vnet_state.bridge = tmp_bridge;
642 v3_unlock_irqrestore(vnet_state.lock, flags);
648 void v3_vnet_do_poll(struct v3_vm_info * vm){
649 struct vnet_dev * dev = NULL;
651 /* TODO: run this on separate threads
652 * round-robin schedule, with maximal budget for each poll
654 list_for_each_entry(dev, &(vnet_state.devs), node) {
655 if(dev->mode == VMM_DRIVERN){
656 dev->dev_ops.poll(vm, -1, dev->private_data);
663 memset(&vnet_state, 0, sizeof(vnet_state));
665 INIT_LIST_HEAD(&(vnet_state.routes));
666 INIT_LIST_HEAD(&(vnet_state.devs));
668 vnet_state.num_devs = 0;
669 vnet_state.num_routes = 0;
671 if (v3_lock_init(&(vnet_state.lock)) == -1){
672 PrintError("VNET/P Core: Fails to initiate lock\n");
675 vnet_state.route_cache = v3_create_htable(0, &hash_fn, &hash_eq);
677 if (vnet_state.route_cache == NULL) {
678 PrintError("VNET/P Core: Fails to initiate route cache\n");
682 PrintDebug("VNET/P Core is initiated\n");
688 void v3_deinit_vnet(){
690 v3_lock_deinit(&(vnet_state.lock));
695 v3_free_htable(vnet_state.route_cache, 1, 1);
696 V3_Free(vnet_state.bridge);