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>
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 <vnet/vnet.h>
23 #include <vnet/vnet_hashtable.h>
24 #include <vnet/vnet_host.h>
25 #include <vnet/vnet_vmm.h>
27 #include <palacios/vmm_queue.h>
29 #ifndef V3_CONFIG_DEBUG_VNET
31 #define PrintDebug(fmt, args...)
34 #define VNET_YIELD_USEC 1000
39 uint8_t dst_mac[ETH_ALEN];
40 uint8_t src_mac[ETH_ALEN];
41 uint16_t type; /* indicates layer 3 protocol type */
42 } __attribute__((packed));
47 uint8_t mac_addr[ETH_ALEN];
48 struct v3_vm_info * vm;
49 struct v3_vnet_dev_ops dev_ops;
53 #define VNET_MAX_QUOTE 64
58 struct list_head node;
59 } __attribute__((packed));
63 struct v3_vm_info * vm;
64 struct v3_vnet_bridge_ops brg_ops;
69 } __attribute__((packed));
73 struct vnet_route_info {
74 struct v3_vnet_route route_def;
76 struct vnet_dev * dst_dev;
77 struct vnet_dev * src_dev;
81 struct list_head node;
82 struct list_head match_node; // used for route matching
87 uint8_t hash_buf[VNET_HASH_SIZE];
90 struct vnet_route_info * routes[0];
91 } __attribute__((packed));
96 struct v3_vnet_pkt pkt;
103 struct list_head routes;
104 struct list_head devs;
113 struct vnet_brg_dev * bridge;
116 struct vnet_stat stats;
118 /* device queue that are waiting to be polled */
119 struct v3_queue * poll_devs;
121 struct vnet_thread * pkt_flush_thread;
123 struct hashtable * route_cache;
127 #ifdef V3_CONFIG_DEBUG_VNET
128 static inline void mac2str(uint8_t * mac, char * buf) {
129 snprintf(buf, 100, "%2x:%2x:%2x:%2x:%2x:%2x",
130 mac[0], mac[1], mac[2],
131 mac[3], mac[4], mac[5]);
134 static void print_route(struct v3_vnet_route * route){
137 mac2str(route->src_mac, str);
138 PrintDebug("Src Mac (%s), src_qual (%d)\n",
139 str, route->src_mac_qual);
140 mac2str(route->dst_mac, str);
141 PrintDebug("Dst Mac (%s), dst_qual (%d)\n",
142 str, route->dst_mac_qual);
143 PrintDebug("Src dev id (%d), src type (%d)",
146 PrintDebug("Dst dev id (%d), dst type (%d)\n",
151 static void dump_routes(){
152 struct vnet_route_info *route;
154 PrintDebug("\n========Dump routes starts ============\n");
155 list_for_each_entry(route, &(vnet_state.routes), node) {
156 PrintDebug("\nroute %d:\n", route->idx);
158 print_route(&(route->route_def));
159 if (route->route_def.dst_type == LINK_INTERFACE) {
160 PrintDebug("dst_dev (%p), dst_dev_id (%d), dst_dev_ops(%p), dst_dev_data (%p)\n",
162 route->dst_dev->dev_id,
163 (void *)&(route->dst_dev->dev_ops),
164 route->dst_dev->private_data);
168 PrintDebug("\n========Dump routes end ============\n");
175 * A VNET packet is a packed struct with the hashed fields grouped together.
176 * This means we can generate the hash from an offset into the pkt struct
178 static inline uint_t hash_fn(addr_t hdr_ptr) {
179 uint8_t * hdr_buf = (uint8_t *)hdr_ptr;
181 return vnet_hash_buffer(hdr_buf, VNET_HASH_SIZE);
184 static inline int hash_eq(addr_t key1, addr_t key2) {
185 return (memcmp((uint8_t *)key1, (uint8_t *)key2, VNET_HASH_SIZE) == 0);
188 static int add_route_to_cache(const struct v3_vnet_pkt * pkt, struct route_list * routes) {
189 memcpy(routes->hash_buf, pkt->hash_buf, VNET_HASH_SIZE);
191 if (vnet_htable_insert(vnet_state.route_cache, (addr_t)routes->hash_buf, (addr_t)routes) == 0) {
192 PrintError("VNET/P Core: Failed to insert new route entry to the cache\n");
199 static int clear_hash_cache() {
200 vnet_free_htable(vnet_state.route_cache, 1, 1);
201 vnet_state.route_cache = vnet_create_htable(0, &hash_fn, &hash_eq);
206 static int look_into_cache(const struct v3_vnet_pkt * pkt,
207 struct route_list ** routes) {
208 *routes = (struct route_list *)vnet_htable_search(vnet_state.route_cache, (addr_t)(pkt->hash_buf));
214 static struct vnet_dev * dev_by_id(int idx) {
215 struct vnet_dev * dev = NULL;
217 list_for_each_entry(dev, &(vnet_state.devs), node) {
218 if (dev->dev_id == idx) {
226 static struct vnet_dev * dev_by_mac(uint8_t * mac) {
227 struct vnet_dev * dev = NULL;
229 list_for_each_entry(dev, &(vnet_state.devs), node) {
230 if (!compare_ethaddr(dev->mac_addr, mac)){
239 int v3_vnet_find_dev(uint8_t * mac) {
240 struct vnet_dev * dev = NULL;
242 dev = dev_by_mac(mac);
252 int v3_vnet_add_route(struct v3_vnet_route route) {
253 struct vnet_route_info * new_route = NULL;
256 new_route = (struct vnet_route_info *)Vnet_Malloc(sizeof(struct vnet_route_info));
257 memset(new_route, 0, sizeof(struct vnet_route_info));
259 #ifdef V3_CONFIG_DEBUG_VNET
260 PrintDebug("VNET/P Core: add_route_entry:\n");
264 memcpy(new_route->route_def.src_mac, route.src_mac, ETH_ALEN);
265 memcpy(new_route->route_def.dst_mac, route.dst_mac, ETH_ALEN);
266 new_route->route_def.src_mac_qual = route.src_mac_qual;
267 new_route->route_def.dst_mac_qual = route.dst_mac_qual;
268 new_route->route_def.dst_type = route.dst_type;
269 new_route->route_def.src_type = route.src_type;
270 new_route->route_def.src_id = route.src_id;
271 new_route->route_def.dst_id = route.dst_id;
273 if (new_route->route_def.dst_type == LINK_INTERFACE) {
274 new_route->dst_dev = dev_by_id(new_route->route_def.dst_id);
277 if (new_route->route_def.src_type == LINK_INTERFACE) {
278 new_route->src_dev = dev_by_id(new_route->route_def.src_id);
282 flags = vnet_lock_irqsave(vnet_state.lock);
284 list_add(&(new_route->node), &(vnet_state.routes));
285 new_route->idx = ++ vnet_state.route_idx;
286 vnet_state.num_routes ++;
288 vnet_unlock_irqrestore(vnet_state.lock, flags);
292 #ifdef V3_CONFIG_DEBUG_VNET
296 return new_route->idx;
300 void v3_vnet_del_route(uint32_t route_idx){
301 struct vnet_route_info * route = NULL;
304 flags = vnet_lock_irqsave(vnet_state.lock);
306 list_for_each_entry(route, &(vnet_state.routes), node) {
307 Vnet_Print(0, "v3_vnet_del_route, route idx: %d\n", route->idx);
308 if(route->idx == route_idx){
309 list_del(&(route->node));
315 vnet_unlock_irqrestore(vnet_state.lock, flags);
318 #ifdef V3_CONFIG_DEBUG_VNET
324 /* delete all route entries with specfied src or dst device id */
325 static void inline del_routes_by_dev(int dev_id){
326 struct vnet_route_info * route, *tmp_route;
329 flags = vnet_lock_irqsave(vnet_state.lock);
331 list_for_each_entry_safe(route, tmp_route, &(vnet_state.routes), node) {
332 if((route->route_def.dst_type == LINK_INTERFACE &&
333 route->route_def.dst_id == dev_id) ||
334 (route->route_def.src_type == LINK_INTERFACE &&
335 route->route_def.src_id == dev_id)){
337 list_del(&(route->node));
338 list_del(&(route->match_node));
343 vnet_unlock_irqrestore(vnet_state.lock, flags);
346 /* At the end allocate a route_list
347 * This list will be inserted into the cache so we don't need to free it
349 static struct route_list * match_route(const struct v3_vnet_pkt * pkt) {
350 struct vnet_route_info * route = NULL;
351 struct route_list * matches = NULL;
354 struct list_head match_list;
355 struct eth_hdr * hdr = (struct eth_hdr *)(pkt->data);
356 // uint8_t src_type = pkt->src_type;
357 // uint32_t src_link = pkt->src_id;
359 #ifdef V3_CONFIG_DEBUG_VNET
364 mac2str(hdr->src_mac, src_str);
365 mac2str(hdr->dst_mac, dst_str);
366 PrintDebug("VNET/P Core: match_route. pkt: SRC(%s), DEST(%s)\n", src_str, dst_str);
370 INIT_LIST_HEAD(&match_list);
372 #define UPDATE_MATCHES(rank) do { \
373 if (max_rank < (rank)) { \
375 INIT_LIST_HEAD(&match_list); \
377 list_add(&(route->match_node), &match_list); \
379 } else if (max_rank == (rank)) { \
380 list_add(&(route->match_node), &match_list); \
386 list_for_each_entry(route, &(vnet_state.routes), node) {
387 struct v3_vnet_route * route_def = &(route->route_def);
390 // CHECK SOURCE TYPE HERE
391 if ( (route_def->src_type != LINK_ANY) &&
392 ( (route_def->src_type != src_type) ||
393 ( (route_def->src_id != src_link) &&
394 (route_def->src_id != -1)))) {
399 if ((route_def->dst_mac_qual == MAC_ANY) &&
400 (route_def->src_mac_qual == MAC_ANY)) {
404 if (memcmp(route_def->src_mac, hdr->src_mac, 6) == 0) {
405 if (route_def->src_mac_qual != MAC_NOT) {
406 if (route_def->dst_mac_qual == MAC_ANY) {
408 } else if (route_def->dst_mac_qual != MAC_NOT &&
409 memcmp(route_def->dst_mac, hdr->dst_mac, 6) == 0) {
415 if (memcmp(route_def->dst_mac, hdr->dst_mac, 6) == 0) {
416 if (route_def->dst_mac_qual != MAC_NOT) {
417 if (route_def->src_mac_qual == MAC_ANY) {
419 } else if ((route_def->src_mac_qual != MAC_NOT) &&
420 (memcmp(route_def->src_mac, hdr->src_mac, 6) == 0)) {
426 if ((route_def->dst_mac_qual == MAC_NOT) &&
427 (memcmp(route_def->dst_mac, hdr->dst_mac, 6) != 0)) {
428 if (route_def->src_mac_qual == MAC_ANY) {
430 } else if ((route_def->src_mac_qual != MAC_NOT) &&
431 (memcmp(route_def->src_mac, hdr->src_mac, 6) == 0)) {
436 if ((route_def->src_mac_qual == MAC_NOT) &&
437 (memcmp(route_def->src_mac, hdr->src_mac, 6) != 0)) {
438 if (route_def->dst_mac_qual == MAC_ANY) {
440 } else if ((route_def->dst_mac_qual != MAC_NOT) &&
441 (memcmp(route_def->dst_mac, hdr->dst_mac, 6) == 0)) {
447 if ( (memcmp(route_def->src_mac, hdr->src_mac, 6) == 0) &&
448 (route_def->dst_mac_qual == MAC_NONE)) {
453 PrintDebug("VNET/P Core: match_route: Matches=%d\n", num_matches);
455 if (num_matches == 0) {
459 matches = (struct route_list *)Vnet_Malloc(sizeof(struct route_list) +
460 (sizeof(struct vnet_route_info *) * num_matches));
462 matches->num_routes = num_matches;
466 list_for_each_entry(route, &match_list, match_node) {
467 matches->routes[i++] = route;
475 int v3_vnet_send_pkt(struct v3_vnet_pkt * pkt, void * private_data) {
476 struct route_list * matched_routes = NULL;
480 int cpu = V3_Get_CPU();
481 Vnet_Print(2, "VNET/P Core: cpu %d: pkt (size %d, src_id:%d, src_type: %d, dst_id: %d, dst_type: %d)\n",
482 cpu, pkt->size, pkt->src_id,
483 pkt->src_type, pkt->dst_id, pkt->dst_type);
485 v3_hexdump(pkt->data, pkt->size, NULL, 0);
488 flags = vnet_lock_irqsave(vnet_state.lock);
490 vnet_state.stats.rx_bytes += pkt->size;
491 vnet_state.stats.rx_pkts++;
493 look_into_cache(pkt, &matched_routes);
494 if (matched_routes == NULL) {
495 PrintDebug("VNET/P Core: send pkt Looking into routing table\n");
497 matched_routes = match_route(pkt);
499 if (matched_routes) {
500 add_route_to_cache(pkt, matched_routes);
502 PrintDebug("VNET/P Core: Could not find route for packet... discards packet\n");
503 vnet_unlock_irqrestore(vnet_state.lock, flags);
504 return 0; /* do we return -1 here?*/
508 vnet_unlock_irqrestore(vnet_state.lock, flags);
510 PrintDebug("VNET/P Core: send pkt route matches %d\n", matched_routes->num_routes);
512 for (i = 0; i < matched_routes->num_routes; i++) {
513 struct vnet_route_info * route = matched_routes->routes[i];
515 if (route->route_def.dst_type == LINK_EDGE) {
516 struct vnet_brg_dev * bridge = vnet_state.bridge;
517 pkt->dst_type = LINK_EDGE;
518 pkt->dst_id = route->route_def.dst_id;
520 if (bridge == NULL) {
521 Vnet_Print(2, "VNET/P Core: No active bridge to sent data to\n");
525 if(bridge->brg_ops.input(bridge->vm, pkt, bridge->private_data) < 0){
526 Vnet_Print(2, "VNET/P Core: Packet not sent properly to bridge\n");
529 vnet_state.stats.tx_bytes += pkt->size;
530 vnet_state.stats.tx_pkts ++;
531 } else if (route->route_def.dst_type == LINK_INTERFACE) {
532 if (route->dst_dev == NULL){
533 Vnet_Print(2, "VNET/P Core: No active device to sent data to\n");
537 if(route->dst_dev->dev_ops.input(route->dst_dev->vm, pkt, route->dst_dev->private_data) < 0) {
538 Vnet_Print(2, "VNET/P Core: Packet not sent properly\n");
541 vnet_state.stats.tx_bytes += pkt->size;
542 vnet_state.stats.tx_pkts ++;
544 Vnet_Print(0, "VNET/P Core: Wrong dst type\n");
552 int v3_vnet_add_dev(struct v3_vm_info * vm, uint8_t * mac,
553 struct v3_vnet_dev_ops * ops, int quote, int poll_state,
555 struct vnet_dev * new_dev = NULL;
558 new_dev = (struct vnet_dev *)Vnet_Malloc(sizeof(struct vnet_dev));
560 if (new_dev == NULL) {
561 Vnet_Print(0, "Malloc fails\n");
565 memcpy(new_dev->mac_addr, mac, 6);
566 new_dev->dev_ops.input = ops->input;
567 new_dev->dev_ops.poll = ops->poll;
568 new_dev->private_data = priv_data;
571 new_dev->quote = quote<VNET_MAX_QUOTE?quote:VNET_MAX_QUOTE;
572 new_dev->poll = poll_state;
574 flags = vnet_lock_irqsave(vnet_state.lock);
576 if (dev_by_mac(mac) == NULL) {
577 list_add(&(new_dev->node), &(vnet_state.devs));
578 new_dev->dev_id = ++ vnet_state.dev_idx;
579 vnet_state.num_devs ++;
582 v3_enqueue(vnet_state.poll_devs, (addr_t)new_dev);
585 PrintError("VNET/P: Device with the same MAC is already there\n");
588 vnet_unlock_irqrestore(vnet_state.lock, flags);
590 /* if the device was found previosly the id should still be 0 */
591 if (new_dev->dev_id == 0) {
592 Vnet_Print(0, "VNET/P Core: Device Already exists\n");
596 PrintDebug("VNET/P Core: Add Device: dev_id %d\n", new_dev->dev_id);
598 return new_dev->dev_id;
602 int v3_vnet_del_dev(int dev_id){
603 struct vnet_dev * dev = NULL;
606 flags = vnet_lock_irqsave(vnet_state.lock);
608 dev = dev_by_id(dev_id);
610 list_del(&(dev->node));
611 //del_routes_by_dev(dev_id);
612 vnet_state.num_devs --;
615 vnet_unlock_irqrestore(vnet_state.lock, flags);
619 PrintDebug("VNET/P Core: Remove Device: dev_id %d\n", dev_id);
625 int v3_vnet_stat(struct vnet_stat * stats){
626 stats->rx_bytes = vnet_state.stats.rx_bytes;
627 stats->rx_pkts = vnet_state.stats.rx_pkts;
628 stats->tx_bytes = vnet_state.stats.tx_bytes;
629 stats->tx_pkts = vnet_state.stats.tx_pkts;
634 static void deinit_devices_list(){
635 struct vnet_dev * dev, * tmp;
637 list_for_each_entry_safe(dev, tmp, &(vnet_state.devs), node) {
638 list_del(&(dev->node));
643 static void deinit_routes_list(){
644 struct vnet_route_info * route, * tmp;
646 list_for_each_entry_safe(route, tmp, &(vnet_state.routes), node) {
647 list_del(&(route->node));
648 list_del(&(route->match_node));
653 int v3_vnet_add_bridge(struct v3_vm_info * vm,
654 struct v3_vnet_bridge_ops * ops,
659 struct vnet_brg_dev * tmp_bridge = NULL;
661 flags = vnet_lock_irqsave(vnet_state.lock);
662 if (vnet_state.bridge == NULL) {
664 vnet_state.bridge = (void *)1;
666 vnet_unlock_irqrestore(vnet_state.lock, flags);
668 if (bridge_free == 0) {
669 PrintError("VNET/P Core: Bridge already set\n");
673 tmp_bridge = (struct vnet_brg_dev *)Vnet_Malloc(sizeof(struct vnet_brg_dev));
675 if (tmp_bridge == NULL) {
676 PrintError("Malloc Fails\n");
677 vnet_state.bridge = NULL;
682 tmp_bridge->brg_ops.input = ops->input;
683 tmp_bridge->brg_ops.poll = ops->poll;
684 tmp_bridge->private_data = priv_data;
685 tmp_bridge->type = type;
687 /* make this atomic to avoid possible race conditions */
688 flags = vnet_lock_irqsave(vnet_state.lock);
689 vnet_state.bridge = tmp_bridge;
690 vnet_unlock_irqrestore(vnet_state.lock, flags);
696 void v3_vnet_del_bridge(uint8_t type) {
698 struct vnet_brg_dev * tmp_bridge = NULL;
700 flags = vnet_lock_irqsave(vnet_state.lock);
702 if (vnet_state.bridge != NULL && vnet_state.bridge->type == type) {
703 tmp_bridge = vnet_state.bridge;
704 vnet_state.bridge = NULL;
707 vnet_unlock_irqrestore(vnet_state.lock, flags);
710 Vnet_Free(tmp_bridge);
715 /* can be instanieoued to multiple threads
716 * that runs on multiple cores
717 * or it could be running on a dedicated side core
719 static int vnet_tx_flush(void * args){
720 struct vnet_dev * dev = NULL;
724 Vnet_Print(0, "VNET/P Polling Thread Starting ....\n");
726 // since there are multiple instances of this thread, and only
727 // one queue of pollable devices, our model here will be to synchronize
728 // on that queue, removing devices as we go, and keeping them
729 // then putting them back on the queue when we are done
730 // in this way, multiple instances of this function will never
731 // be polling the same device at the same time
733 struct v3_queue * tq = v3_create_queue();
736 PrintError("VNET/P polling thread cannot allocate queue\n");
741 while (!vnet_thread_should_stop()) {
743 more=0; // will indicate if any device has more work for us to do
745 while ((dev = (struct vnet_dev *)v3_dequeue(vnet_state.poll_devs))) {
746 // we are handling this device
747 v3_enqueue(tq,(addr_t)dev);
749 if (dev->poll && dev->dev_ops.poll) {
750 // The device's poll function MUST NOT BLOCK
751 rc = dev->dev_ops.poll(dev->vm, dev->quote, dev->private_data);
754 Vnet_Print(0, "VNET/P: poll from device %p error (ignoring) !\n", dev);
761 while ((dev = (struct vnet_dev *)v3_dequeue(tq))) {
762 // now someone else can handle it
763 v3_enqueue(vnet_state.poll_devs, (addr_t)dev);
766 // Yield regardless of whether we handled any devices - need
767 // to allow other threads to run
769 // we have more to do, so we want to get back asap
772 // put ourselves briefly to sleep if we we don't have more
773 V3_Yield_Timed(VNET_YIELD_USEC);
778 Vnet_Print(0, "VNET/P Polling Thread Done.\n");
784 memset(&vnet_state, 0, sizeof(vnet_state));
786 INIT_LIST_HEAD(&(vnet_state.routes));
787 INIT_LIST_HEAD(&(vnet_state.devs));
789 vnet_state.num_devs = 0;
790 vnet_state.num_routes = 0;
792 if (vnet_lock_init(&(vnet_state.lock)) == -1){
793 PrintError("VNET/P: Fails to initiate lock\n");
796 vnet_state.route_cache = vnet_create_htable(0, &hash_fn, &hash_eq);
797 if (vnet_state.route_cache == NULL) {
798 PrintError("VNET/P: Fails to initiate route cache\n");
802 vnet_state.poll_devs = v3_create_queue();
804 vnet_state.pkt_flush_thread = vnet_start_thread(vnet_tx_flush, NULL, "vnetd-1");
806 PrintDebug("VNET/P is initiated\n");
812 void v3_deinit_vnet(){
814 PrintDebug("Stopping flush thread\n");
815 // This will pause until the flush thread is gone
816 vnet_thread_stop(vnet_state.pkt_flush_thread);
817 // At this point there should be no lock-holder
819 PrintDebug("Deiniting Device List\n");
820 // close any devices we have open
821 deinit_devices_list();
823 PrintDebug("Deiniting Route List\n");
824 // remove any routes we have
825 deinit_routes_list();
827 PrintDebug("Freeing hash table\n");
828 // remove the hash table
829 vnet_free_htable(vnet_state.route_cache, 1, 1);
832 PrintDebug("Removing Bridge\n");
833 // remove bridge if it was added
834 if (vnet_state.bridge) {
835 Vnet_Free(vnet_state.bridge);
838 PrintDebug("Deleting lock\n");
839 // eliminate the lock
840 vnet_lock_deinit(&(vnet_state.lock));