improved load balancing algorithm.

[l2tpns.git] / grpsess.c

diff --git a/grpsess.c b/grpsess.c
index bfc7d8d..f38ab1a 100644 (file)
--- a/grpsess.c
+++ b/grpsess.c
@@ -12,6 +12,7 @@
 
 #include "l2tpns.h"
 #include "util.h"
+#include "cluster.h"
 
 #ifdef BGP
 #include "bgp.h"
@@ -41,6 +42,14 @@ static groupidt grp_lookup_ipmap(in_addr_t ip)
        return h[*a].grp;
 }
 
+typedef struct
+{
+       sessionidt sid_loaddist[0x10000];
+}
+local_group;
+
+local_group *grp_local = NULL;         // Array of local_group structures.
+
 //
 // Take an IP address in HOST byte order and
 // add it to the grouid by IP cache.
@@ -84,7 +93,7 @@ groupidt grp_groupbyip(in_addr_t ip)
 //
 // This adds it to the routing table, advertises it
 // via BGP if enabled, and stuffs it into the
-// 'sessionbyip' cache.
+// 'groupbyip' cache.
 //
 // 'ip' must be in _host_ order.
 //
@@ -128,7 +137,7 @@ static void grp_routeset(groupidt g, in_addr_t ip, int prefixlen, int add)
        n_ip = htonl(ip);
        netlink_addattr(&req.nh, RTA_DST, &n_ip, sizeof(n_ip));
 
-       LOG(1, 0, 0, "Route (Group) %s %s/%d\n", add ? "add" : "del", fmtaddr(htonl(ip), 0), prefixlen);
+       LOG(3, 0, 0, "Route (Group) %s %s/%d\n", add ? "add" : "del", fmtaddr(htonl(ip), 0), prefixlen);
 
        if (netlink_send(&req.nh) < 0)
                LOG(0, 0, 0, "grp_routeset() error in sending netlink message: %s\n", strerror(errno));
@@ -153,7 +162,10 @@ static void grp_routeset(groupidt g, in_addr_t ip, int prefixlen, int add)
                        g = 0;  // Caching the session as '0' is the same as uncaching.
 
                for (i = ip; i < ip+(1<<(32-prefixlen)) ; ++i)
+               {
                        grp_cache_ipmap(i, g);
+                       if (!g) cache_ipmap(i, 0);
+               }
        }
 }
 
@@ -212,7 +224,7 @@ void grp_removesession(groupidt g, sessionidt s)
 
                                if (gnextgrpid == g)
                                {
-                                       gnextgrpid = 0;
+                                       gnextgrpid = grpsession[g].prev;
                                }
                                else
                                {
@@ -228,6 +240,7 @@ void grp_removesession(groupidt g, sessionidt s)
                                }
 
                                memset(&grpsession[g], 0, sizeof(grpsession[0]));
+                               grpsession[g].state = GROUPEFREE;
                        }
                        else
                        {
@@ -236,6 +249,8 @@ void grp_removesession(groupidt g, sessionidt s)
                                                &grpsession[g].sesslist[i+1],
                                                (grpsession[g].nbsession - i) * sizeof(grpsession[g].sesslist[i]));
                        }
+
+                       cluster_send_groupe(g);
                        return;
                }
        }
@@ -270,6 +285,7 @@ static int grp_addsession(groupidt g, sessionidt s, uint8_t weight)
                        // it's the first session of the group, set to next group
                        grpsession[g].prev = gnextgrpid;
                        gnextgrpid = g;
+                       grpsession[g].state = GROUPEOPEN;
                }
                grpsession[g].sesslist[i].sid = s;
                grpsession[g].sesslist[i].weight = weight;
@@ -356,6 +372,9 @@ void grp_processvendorspecific(sessionidt s, uint8_t *pvs)
                        return;
                }
 
+               if (grpid > config->cluster_highest_groupeid)
+                       config->cluster_highest_groupeid = grpid;
+
                n++;
        }
 
@@ -420,6 +439,8 @@ void grp_processvendorspecific(sessionidt s, uint8_t *pvs)
 // Init data structures
 void grp_initdata()
 {
+       int i;
+
        // Set default value (10s)
        config->grp_txrate_average_time = 10;
 
@@ -430,6 +451,18 @@ void grp_initdata()
        }
 
        memset(grpsession, 0, sizeof(grpsession[0]) * MAXGROUPE);
+       for (i = 1; i < MAXGROUPE; i++)
+       {
+               grpsession[i].state = GROUPEUNDEF;
+       }
+
+       if (!(grp_local = shared_malloc(sizeof(local_group) * MAXGROUPE)))
+       {
+               LOG(0, 0, 0, "Error doing malloc for grp_local: %s\n", strerror(errno));
+               exit(1);
+       }
+       memset(grp_local, 0, sizeof(grp_local[0]) * MAXGROUPE);
+
 }
 
 // Update time_changed of the group
@@ -481,7 +514,7 @@ static void grp_uncache_ipsession(groupidt g, sessionidt s)
 }
 
 // return the next session can be used on the group
-sessionidt grp_getnextsession(groupidt g, in_addr_t ip)
+sessionidt grp_getnextsession(groupidt g, in_addr_t ip, in_addr_t ip_src)
 {
        sessionidt s = 0, s2 = 0, s3 = 0;
        int i;
@@ -500,8 +533,13 @@ sessionidt grp_getnextsession(groupidt g, in_addr_t ip)
                {
                        if ((s2 = grpsession[g].sesslist[i].sid))
                        {
-                               grpsession[g].sesslist[i].tx_rate = session[s2].coutgrp_delta/ltime_changed;
-                               session[s2].coutgrp_delta = 0;
+                               uint32_t coutgrp_delta = 0;
+
+                               if (session[s2].cout >= grpsession[g].sesslist[i].prev_coutgrp)
+                                       coutgrp_delta = session[s2].cout - grpsession[g].sesslist[i].prev_coutgrp;
+                               grpsession[g].sesslist[i].prev_coutgrp = session[s2].cout;
+
+                               grpsession[g].sesslist[i].tx_rate = coutgrp_delta/ltime_changed;
 
                                txrate = grpsession[g].sesslist[i].tx_rate/grpsession[g].sesslist[i].weight;
                                if (txrate < mintxrate)
@@ -529,9 +567,30 @@ sessionidt grp_getnextsession(groupidt g, in_addr_t ip)
 
        if ((s = sessionbyip(ip)))
        {
-               if (s == grpsession[g].smax)
+               uint8_t *as = (uint8_t *) &ip_src;
+               uint8_t *ad = (uint8_t *) &ip;
+               uint16_t ai = ad[3];
+               ai <<= 8;
+               ai &= as[3];
+
+               s = grp_local[g].sid_loaddist[ai];
+               if (!s)
                {
                        s = grpsession[g].smin;
+                       grp_local[g].sid_loaddist[ai] = s;
+               }
+
+               if (g != grp_groupbysession(s))
+               {
+                       // This session does not belong to this group
+                       LOG(2, s, session[s].tunnel, "Warning, the session does not belong to group %d\n", g);
+                       s = 0;
+                       grp_local[g].sid_loaddist[ai] = 0;
+               }
+               else if (s == grpsession[g].smax)
+               {
+                       s = grpsession[g].smin;
+                       grp_local[g].sid_loaddist[ai] = s;
                        grpsession[g].smax = 0;
                }
                else if ( (session[s].ppp.phase > Establish) &&
@@ -542,6 +601,7 @@ sessionidt grp_getnextsession(groupidt g, in_addr_t ip)
                else
                {
                        s = 0;
+                       grp_local[g].sid_loaddist[ai] = 0;
                }
        }
 
@@ -587,3 +647,66 @@ sessionidt grp_getnextsession(groupidt g, in_addr_t ip)
 
        return s;
 }
+
+// load a groupe receive from master
+int grp_cluster_load_groupe(groupidt g, groupsesst *new)
+{
+       int i;
+       int updategroup = 0;
+
+       if (g >= MAXGROUPE)
+       {
+               LOG(0, 0, 0, "ERROR: Received a group id > MAXGROUPE!\n");
+               return 0;
+       }
+
+       if ((grpsession[g].nbroutesgrp != new->nbroutesgrp) ||
+               (grpsession[g].nbsession != new->nbsession))
+       {
+               updategroup = 1;
+       }
+
+       if (!updategroup)
+       {
+               // Check session list
+               for (i = 0; i < grpsession[g].nbsession; i++)
+               {
+                       if (grpsession[g].sesslist[i].sid != new->sesslist[i].sid)
+                       {
+                               updategroup = 1;
+                               break;
+                       }
+               }
+       }
+
+       if (!updategroup)
+       {
+               // Check routes list
+               for (i = 0; i < grpsession[g].nbroutesgrp; i++)
+               {
+                       if (grpsession[g].route[i].ip != new->route[i].ip)
+                       {
+                               updategroup = 1;
+                               break;
+                       }
+               }
+       }
+
+       // needs update
+       if (updategroup)
+       {
+               // Del all routes
+               grp_setgrouproute(g, 0);
+       }
+
+       memcpy(&grpsession[g], new, sizeof(grpsession[g]));     // Copy over..
+
+       // needs update
+       if (updategroup)
+       {
+               // Add all routes
+               grp_setgrouproute(g, 1);
+       }
+
+       return 1;
+}