#include <linux/sched.h>
#include <asm/msr.h>
+#include <linux/net.h>
+#include <linux/socket.h>
+#include <net/sock.h>
+
#include <vnet/vnet.h>
#include <vnet/vnet_hashtable.h>
#include "palacios-vnet.h"
#include "palacios.h"
+
#define VNET_SERVER_PORT 9000
-#define VNET_YIELD_TIME_USEC 1000
+#define VNET_NOPROGRESS_LIMIT 1000
+
+#define VNET_YIELD_TIME_USEC 1000
struct vnet_link {
uint32_t dst_ip;
return -1;
}
+ if (link->sock_proto == UDP) {
+ // no UDP checksumming
+ lock_sock(link->sock->sk);
+ link->sock->sk->sk_no_check = 1;
+ release_sock(link->sock->sk);
+ }
+
memset(&link->sock_addr, 0, sizeof(struct sockaddr));
link->sock_addr.sin_family = AF_INET;
link->sock_addr.sin_addr.s_addr = link->dst_ip;
link->sock_addr.sin_port = htons(link->dst_port);
+
if ((err = link->sock->ops->connect(link->sock, (struct sockaddr *)&(link->sock_addr), sizeof(struct sockaddr), 0) < 0)) {
WARNING("Could not connect to remote VNET Server, error %d\n", err);
return -1;
}
+
spin_lock_irqsave(&(vnet_brg_s.lock), flags);
list_add(&(link->node), &(vnet_brg_s.link_list));
vnet_brg_s.num_links ++;
iov.iov_base = buf;
iov.iov_len = len;
- msg.msg_flags = 0;
+ msg.msg_flags = MSG_NOSIGNAL;
msg.msg_name = addr;
msg.msg_namelen = sizeof(struct sockaddr_in);
msg.msg_control = NULL;
return -1;
}
+ if (vnet_brg_s.serv_proto == UDP) {
+ // No UDP checksumming is done
+ lock_sock(vnet_brg_s.serv_sock->sk);
+ vnet_brg_s.serv_sock->sk->sk_no_check = 1;
+ release_sock(vnet_brg_s.serv_sock->sk);
+ }
+
memset(&vnet_brg_s.serv_addr, 0, sizeof(struct sockaddr));
vnet_brg_s.serv_addr.sin_family = AF_INET;
}
}
+
+
return 0;
}
struct sockaddr_in pkt_addr;
struct vnet_link * link = NULL;
int len;
+ uint64_t noprogress_count;
INFO("Palacios VNET Bridge: UDP receiving server ..... \n");
return -1;
}
+
+ noprogress_count=0;
while (!kthread_should_stop()) {
// If it would have blocked, we have no packet, and so
// we will give other threads on this core a chance
if (len==-EAGAIN || len==-EWOULDBLOCK || len==-EINTR) {
- palacios_yield_cpu_timed(VNET_YIELD_TIME_USEC);
+
+ // avoid rollover in the counter out of paranoia
+ if (! ((noprogress_count + 1) < noprogress_count)) {
+ noprogress_count++;
+ }
+
+ // adaptively select yielding strategy depending on
+ // whether we are making progress
+ if (noprogress_count < VNET_NOPROGRESS_LIMIT) {
+ // Likely making progress, do fast yield so we
+ // come back immediately if there is no other action
+ palacios_yield_cpu();
+ } else {
+ // Likely not making progress, do potentially slow
+ // yield - we won't come back for until VNET_YIELD_TIME_USEC has passed
+ palacios_yield_cpu_timed(VNET_YIELD_TIME_USEC);
+ }
+
continue;
}
+
+ // Something interesting has happened, therefore progress!
+ noprogress_count=0;
+
+
if(len < 0) {
WARNING("Receive error: Could not get packet, error %d\n", len);
continue;
#define PrintDebug(fmt, args...)
#endif
-#define VNET_YIELD_USEC 1000
+#define VNET_NOPROGRESS_LIMIT 1000
+#define VNET_YIELD_USEC 1000
int net_debug = 0;
Vnet_Free(route);
}
}
-
+
vnet_unlock_irqrestore(vnet_state.lock, flags);
}
+// Match classes, must be in order
+#define NUM_MATCH_CLASSES 4
+#define NUM_MATCH_CLASSES_BOUND 3
+#define NONE 0
+#define NOT 1
+#define ANY 2
+#define DIRECT 3
-/* At the end allocate a route_list
- * This list will be inserted into the cache so we don't need to free it
- */
-static struct route_list * match_route(const struct v3_vnet_pkt * pkt) {
+static inline uint8_t match_mac(uint8_t test_mac[ETH_ALEN],
+ uint8_t route_mac[ETH_ALEN],
+ uint8_t route_qual)
+{
+ switch (route_qual) {
+ case MAC_NOSET:
+ return NONE;
+ break;
+ case MAC_NONE:
+ return NONE;
+ break;
+ case MAC_ANY:
+ return ANY;
+ break;
+ case MAC_NOT:
+ if (memcmp(test_mac,route_mac,ETH_ALEN)) {
+ return NOT;
+ } else {
+ return NONE;
+ }
+ break;
+ case MAC_ADDR:
+ if (memcmp(test_mac,route_mac,ETH_ALEN)) {
+ return NONE;
+ } else {
+ return DIRECT;
+ }
+ break;
+ default:
+ PrintError("Unknown qualifier %u\n",route_qual);
+ return NONE;
+ break;
+ }
+
+}
+
+#define QUAL_TO_STR(q) ( \
+(q)==MAC_NOSET ? "MAC_NOSET" : \
+(q)==MAC_NONE ? "MAC_NONE" : \
+(q)==MAC_ANY ? "MAC_ANY" : \
+(q)==MAC_NOT ? "MAC_NOT" : \
+(q)==MAC_ADDR ? "MAC_ADDR" : \
+"***UNDEFINED****" \
+ ) \
+
+#define MATCH_CLASS_TO_STR(c) ( \
+(c)==NONE ? "NONE" : \
+(c)==NOT ? "NOT" : \
+(c)==ANY ? "ANY" : \
+(c)==DIRECT ? "DIRECT" : \
+"***UNDEFINED****" \
+ ) \
+
+
+
+/*
+
+Original priority behavior...
+
+priority src srcqual dst dstqual
+3 ANY ANY
+4 X NONE
+5 ANY X NOT
+5 X NOT ANY
+6 X ~NOT ANY
+6 ANY X ~NOT
+7 X ~NOT X NOT
+7 X NOT X ~NOT
+8 X ~NOT X ~NOT
+8 X ~NOT X ~NOT
+
+*/
+
+/*
+ Current priority order is given in the following table
+*/
+
+// [src][dst] => priority
+static int priority_map[NUM_MATCH_CLASSES][NUM_MATCH_CLASSES] =
+{
+ [NONE] = { [ 0 ... NUM_MATCH_CLASSES_BOUND ] = -1}, // ignore if it's not a source match
+ [NOT][NONE] = -1, // ignore it if there is no destination match
+ [NOT][NOT] = 3,
+ [NOT][ANY] = 5,
+ [NOT][DIRECT] = 7,
+ [ANY][NONE] = -1, // ignore if there is no destination match
+ [ANY][NOT] = 5,
+ [ANY][ANY] = 6,
+ [ANY][DIRECT] = 6,
+ [DIRECT][NONE] = -1, // ignore if there is no destination match
+ [DIRECT][NOT] = 7,
+ [DIRECT][ANY] = 8,
+ [DIRECT][DIRECT] = 8,
+};
+
+
+
+
+static inline int match_priority(uint8_t src_mac[ETH_ALEN],
+ uint8_t dst_mac[ETH_ALEN],
+ uint8_t route_src_mac[ETH_ALEN],
+ uint8_t route_src_qual,
+ uint8_t route_dst_mac[ETH_ALEN],
+ uint8_t route_dst_qual)
+
+{
+
+ return priority_map[match_mac(src_mac,route_src_mac,route_src_qual)][match_mac(dst_mac,route_dst_mac,route_dst_qual)];
+}
+
+
+/*
+ Route matching will return the list of the highest priority routes that
+ match. It's a list because it's possible to have multiple high priority routes
+ */
+static struct route_list * match_route(const struct v3_vnet_pkt * pkt)
+{
+ int i;
struct vnet_route_info * route = NULL;
struct route_list * matches = NULL;
int num_matches = 0;
- int max_rank = 0;
+ int max_priority = -1;
struct list_head match_list;
struct eth_hdr * hdr = (struct eth_hdr *)(pkt->data);
- // uint8_t src_type = pkt->src_type;
- // uint32_t src_link = pkt->src_id;
+ //
+ //
+ // NOTE: USING THE MATCH_NODE in the route list to record a match list
+ // IS A DISASTER WAITING TO HAPPEN
+ //
+
#ifdef V3_CONFIG_DEBUG_VNET
{
- char dst_str[100];
- char src_str[100];
-
+ char dst_str[32], src_str[32];
mac2str(hdr->src_mac, src_str);
mac2str(hdr->dst_mac, dst_str);
PrintDebug("VNET/P Core: match_route. pkt: SRC(%s), DEST(%s)\n", src_str, dst_str);
}
#endif
-
- INIT_LIST_HEAD(&match_list);
-#define UPDATE_MATCHES(rank) do { \
- if (max_rank < (rank)) { \
- max_rank = (rank); \
- INIT_LIST_HEAD(&match_list); \
- \
- list_add(&(route->match_node), &match_list); \
- num_matches = 1; \
- } else if (max_rank == (rank)) { \
- list_add(&(route->match_node), &match_list); \
- num_matches++; \
- } \
- } while (0)
+ INIT_LIST_HEAD(&match_list);
+
-
list_for_each_entry(route, &(vnet_state.routes), node) {
+
struct v3_vnet_route * route_def = &(route->route_def);
+
+ int priority;
+
+ priority = match_priority(hdr->src_mac,
+ hdr->dst_mac,
+ route_def->src_mac,
+ route_def->src_mac_qual,
+ route_def->dst_mac,
+ route_def->dst_mac_qual);
-/*
- // CHECK SOURCE TYPE HERE
- if ( (route_def->src_type != LINK_ANY) &&
- ( (route_def->src_type != src_type) ||
- ( (route_def->src_id != src_link) &&
- (route_def->src_id != -1)))) {
- continue;
+
+
+#ifdef V3_CONFIG_DEBUG_VNET
+ {
+ char dst_str[32];
+ char src_str[32];
+
+ mac2str(route_def->src_mac, src_str);
+ mac2str(route_def->dst_mac, dst_str);
+
+ PrintDebug("Tested match against SRC(%s) SRC_QUAL(%s), DEST(%s) DST_QUAL(%s): "
+ "SRC_MATCH=%s DEST_MATCH=%s PRIORITY=%d\n",
+ src_str, QUAL_TO_STR(route_def->src_mac_qual),
+ dst_str, QUAL_TO_STR(route_def->dst_mac_qual),
+ MATCH_CLASS_TO_STR(match_mac(hdr->src_mac,route_def->src_mac,route_def->src_mac_qual)),
+ MATCH_CLASS_TO_STR(match_mac(hdr->dst_mac,route_def->dst_mac,route_def->dst_mac_qual)),
+ priority);
}
-*/
+#endif
- if ((route_def->dst_mac_qual == MAC_ANY) &&
- (route_def->src_mac_qual == MAC_ANY)) {
- UPDATE_MATCHES(3);
+ if (priority<0) {
+ PrintDebug("No match to this rule\n");
+ continue;
}
-
- if (memcmp(route_def->src_mac, hdr->src_mac, 6) == 0) {
- if (route_def->src_mac_qual != MAC_NOT) {
- if (route_def->dst_mac_qual == MAC_ANY) {
- UPDATE_MATCHES(6);
- } else if (route_def->dst_mac_qual != MAC_NOT &&
- memcmp(route_def->dst_mac, hdr->dst_mac, 6) == 0) {
- UPDATE_MATCHES(8);
- }
+
+ if (priority > max_priority) {
+ PrintDebug("New highest priority match, reseting list\n");
+ max_priority = priority;
+
+ struct vnet_route_info *my_route, *tmp_route;
+
+ list_for_each_entry_safe(my_route, tmp_route, &match_list,match_node) {
+ list_del(&(my_route->match_node));
}
- }
+
+ list_add(&(route->match_node), &match_list);
+ num_matches = 1;
- if (memcmp(route_def->dst_mac, hdr->dst_mac, 6) == 0) {
- if (route_def->dst_mac_qual != MAC_NOT) {
- if (route_def->src_mac_qual == MAC_ANY) {
- UPDATE_MATCHES(6);
- } else if ((route_def->src_mac_qual != MAC_NOT) &&
- (memcmp(route_def->src_mac, hdr->src_mac, 6) == 0)) {
- UPDATE_MATCHES(8);
- }
- }
- }
+ } else if (priority == max_priority) {
+ PrintDebug("Equal priority match, adding to list\n");
- if ((route_def->dst_mac_qual == MAC_NOT) &&
- (memcmp(route_def->dst_mac, hdr->dst_mac, 6) != 0)) {
- if (route_def->src_mac_qual == MAC_ANY) {
- UPDATE_MATCHES(5);
- } else if ((route_def->src_mac_qual != MAC_NOT) &&
- (memcmp(route_def->src_mac, hdr->src_mac, 6) == 0)) {
- UPDATE_MATCHES(7);
- }
- }
+ list_add(&(route->match_node), &match_list);
+ num_matches++;
+ }
- if ((route_def->src_mac_qual == MAC_NOT) &&
- (memcmp(route_def->src_mac, hdr->src_mac, 6) != 0)) {
- if (route_def->dst_mac_qual == MAC_ANY) {
- UPDATE_MATCHES(5);
- } else if ((route_def->dst_mac_qual != MAC_NOT) &&
- (memcmp(route_def->dst_mac, hdr->dst_mac, 6) == 0)) {
- UPDATE_MATCHES(7);
- }
- }
-
- // Default route
- if ( (memcmp(route_def->src_mac, hdr->src_mac, 6) == 0) &&
- (route_def->dst_mac_qual == MAC_NONE)) {
- UPDATE_MATCHES(4);
- }
}
PrintDebug("VNET/P Core: match_route: Matches=%d\n", num_matches);
if (num_matches <= 0) {
return NULL;
}
-
+
matches = (struct route_list *)Vnet_Malloc(sizeof(struct route_list) +
(sizeof(struct vnet_route_info *) * num_matches));
matches->num_routes = num_matches;
- {
- int i = 0;
- list_for_each_entry(route, &match_list, match_node) {
+ i=0;
+ list_for_each_entry(route, &match_list, match_node) {
+ if (i==num_matches) {
+ // the list should never have more than num_matches on it...
+ PrintError("Weird list behavior\n");
+ break;
+ } else {
matches->routes[i++] = route;
}
+
}
return matches;
}
-int v3_vnet_query_header(uint8_t src_mac[6],
- uint8_t dest_mac[6],
+int v3_vnet_query_header(uint8_t src_mac[ETH_ALEN],
+ uint8_t dest_mac[ETH_ALEN],
int recv, // 0 = send, 1=recv
struct v3_vnet_header *header)
{
struct route_list *routes;
struct vnet_route_info *r;
struct v3_vnet_pkt p;
+ void *flags;
p.size=14;
p.data=p.header;
memcpy(header->src_mac,src_mac,6);
memcpy(header->dst_mac,dest_mac,6);
+
+ flags = vnet_lock_irqsave(vnet_state.lock);
look_into_cache(&p,&routes);
if (!routes) {
routes = match_route(&p);
if (!routes) {
+ vnet_unlock_irqrestore(vnet_state.lock,flags);
PrintError("Cannot match route\n");
header->header_type=VNET_HEADER_NOMATCH;
header->header_len=0;
add_route_to_cache(&p,routes);
}
}
+
+ vnet_unlock_irqrestore(vnet_state.lock,flags);
if (routes->num_routes<1) {
PrintError("Less than one route\n");
struct vnet_dev * dev = NULL;
int more;
int rc;
+ uint64_t noprogress_count;
Vnet_Print(0, "VNET/P Polling Thread Starting ....\n");
return -1;
}
+ noprogress_count=0;
while (!vnet_thread_should_stop()) {
v3_enqueue(vnet_state.poll_devs, (addr_t)dev);
}
- // Yield regardless of whether we handled any devices - need
- // to allow other threads to run
+
if (more) {
- // we have more to do, so we want to get back asap
+ noprogress_count=0;
+ } else {
+ if ( ! ((noprogress_count+1) < noprogress_count)) {
+ noprogress_count++;
+ }
+ }
+
+ // adaptively yield
+ if (noprogress_count < VNET_NOPROGRESS_LIMIT) {
V3_Yield();
} else {
- // put ourselves briefly to sleep if we we don't have more
V3_Yield_Timed(VNET_YIELD_USEC);
}
