// L2TPNS Clustering Stuff
-char const *cvs_id_cluster = "$Id: cluster.c,v 1.19 2004/11/29 02:17:17 bodea Exp $";
+char const *cvs_id_cluster = "$Id: cluster.c,v 1.25 2004/12/16 08:49:53 bodea Exp $";
#include <stdio.h>
#include <sys/file.h>
#include <unistd.h>
#include <stdio.h>
#include <libcli.h>
+#include <inttypes.h>
#include "l2tpns.h"
#include "cluster.h"
// Module variables.
int cluster_sockfd = 0; // The filedescriptor for the cluster communications port.
-ipt my_address = 0; // The network address of my ethernet port.
+in_addr_t my_address = 0; // The network address of my ethernet port.
static int walk_session_number = 0; // The next session to send when doing the slow table walk.
static int walk_tunnel_number = 0; // The next tunnel to send when doing the slow table walk.
// we can re-transmit if needed.
static struct {
- u32 peer;
- time_t basetime;
- clockt timestamp;
- int uptodate;
+ in_addr_t peer;
+ time_t basetime;
+ clockt timestamp;
+ int uptodate;
} peers[CLUSTER_MAX_SIZE]; // List of all the peers we've heard from.
static int num_peers; // Number of peers in list.
-static int rle_decompress(u8 ** src_p, int ssize, u8 *dst, int dsize);
-static int rle_compress(u8 ** src_p, int ssize, u8 *dst, int dsize);
+static int rle_decompress(uint8_t **src_p, int ssize, uint8_t *dst, int dsize);
+static int rle_compress(uint8_t **src_p, int ssize, uint8_t *dst, int dsize);
//
// Create a listening socket
struct sockaddr_in interface_addr;
struct ip_mreq mreq;
struct ifreq ifr;
- int opt = 0;
+ int opt;
config->cluster_undefined_sessions = MAXSESSION-1;
config->cluster_undefined_tunnels = MAXTUNNEL-1;
addr.sin_addr.s_addr = INADDR_ANY;
setsockopt(cluster_sockfd, SOL_SOCKET, SO_REUSEADDR, &addr, sizeof(addr));
+ opt = fcntl(cluster_sockfd, F_GETFL, 0);
+ fcntl(cluster_sockfd, F_SETFL, opt | O_NONBLOCK);
+
if (bind(cluster_sockfd, (void *) &addr, sizeof(addr)) < 0)
{
LOG(0, 0, 0, "Failed to bind cluster socket: %s\n", strerror(errno));
return -1;
}
- if (setsockopt (cluster_sockfd, IPPROTO_IP, IP_MULTICAST_IF, &interface_addr, sizeof(interface_addr)) < 0)
+ if (setsockopt(cluster_sockfd, IPPROTO_IP, IP_MULTICAST_IF, &interface_addr, sizeof(interface_addr)) < 0)
{
LOG(0, 0, 0, "Failed to setsockopt (set mcast interface): %s\n", strerror(errno));
return -1;
// Maintains the format. Assumes that the caller
// has passed in a big enough buffer!
//
-static void add_type(char ** p, int type, int more, char * data, int size)
+static void add_type(char **p, int type, int more, char *data, int size)
{
- * ( (u32*)(*p) ) = type;
- *p += sizeof(u32);
+ *((uint32_t *) (*p)) = type;
+ *p += sizeof(uint32_t);
- * ( (u32*)(*p) ) = more;
- *p += sizeof(u32);
+ *((uint32_t *)(*p)) = more;
+ *p += sizeof(uint32_t);
if (data && size > 0) {
memcpy(*p, data, size);
- (*p) += size;
+ *p += size;
}
}
// Send a unicast UDP packet to a peer with 'data' as the
// contents.
//
-static int peer_send_data(u32 peer, char * data, int size)
+static int peer_send_data(in_addr_t peer, char *data, int size)
{
struct sockaddr_in addr = {0};
//
// Send a structured message to a peer with a single element of type 'type'.
//
-static int peer_send_message(u32 peer, int type, int more, char * data, int size)
+static int peer_send_message(in_addr_t peer, int type, int more, char *data, int size)
{
char buf[65536]; // Vast overkill.
- char * p = buf;
+ char *p = buf;
LOG(4, 0, 0, "Sending message to peer (type %d, more %d, size %d)\n", type, more, size);
add_type(&p, type, more, data, size);
// The master just processes the payload as if it had
// received it off the tun device.
//
-int master_forward_packet(char *data, int size, u32 addr, int port)
+int master_forward_packet(char *data, int size, in_addr_t addr, int port)
{
char buf[65536]; // Vast overkill.
char *p = buf;
LOG(4, 0, 0, "Forwarding packet from %s to master (size %d)\n", fmtaddr(addr, 0), size);
STAT(c_forwarded);
- add_type(&p, C_FORWARD, addr, (char*) &port, sizeof(port) );
+ add_type(&p, C_FORWARD, addr, (char *) &port, sizeof(port));
memcpy(p, data, size);
p += size;
- return peer_send_data(config->cluster_master_address, buf, (p-buf) );
-
+ return peer_send_data(config->cluster_master_address, buf, (p - buf));
}
//
// Send a chunk of data as a heartbeat..
// We save it in the history buffer as we do so.
//
-static void send_heartbeat(int seq, char * data, int size)
+static void send_heartbeat(int seq, char *data, int size)
{
int i;
// Forward the data to the master.
LOG(4, 0, 0, "Sending byte counters to master (%d elements)\n", c);
- peer_send_message(config->cluster_master_address, C_BYTES, c, (char*) &b, sizeof(b[0]) * c);
+ peer_send_message(config->cluster_master_address, C_BYTES, c, (char *) &b, sizeof(b[0]) * c);
return;
}
{
switch (type) {
case C_CSESSION: { // Compressed C_SESSION.
- u8 c[sizeof(sessiont) * 2]; // Bigger than worst case.
- u8 *d = (u8 *) &session[id];
- u8 *orig = d;
+ uint8_t c[sizeof(sessiont) * 2]; // Bigger than worst case.
+ uint8_t *d = (uint8_t *) &session[id];
+ uint8_t *orig = d;
int size;
size = rle_compress( &d, sizeof(sessiont), c, sizeof(c) );
// Did we compress the full structure, and is the size actually
// reduced??
if ( (d - orig) == sizeof(sessiont) && size < sizeof(sessiont) ) {
- add_type(p, C_CSESSION, id, (char*) c, size);
+ add_type(p, C_CSESSION, id, (char *) c, size);
break;
}
// Failed to compress : Fall through.
}
case C_SESSION: add_type(p, C_SESSION, id,
- (char*) &session[id], sizeof(sessiont));
+ (char *) &session[id], sizeof(sessiont));
break;
case C_CTUNNEL: { // Compressed C_TUNNEL
- u8 c[sizeof(tunnelt) * 2]; // Bigger than worst case.
- u8 *d = (u8 *) &tunnel[id];
- u8 *orig = d;
+ uint8_t c[sizeof(tunnelt) * 2]; // Bigger than worst case.
+ uint8_t *d = (uint8_t *) &tunnel[id];
+ uint8_t *orig = d;
int size;
size = rle_compress( &d, sizeof(tunnelt), c, sizeof(c) );
// Failed to compress : Fall through.
}
case C_TUNNEL: add_type(p, C_TUNNEL, id,
- (char*) &tunnel[id], sizeof(tunnelt));
+ (char *) &tunnel[id], sizeof(tunnelt));
break;
default:
LOG(0, 0, 0, "Found an invalid type in heart queue! (%d)\n", type);
if (!config->cluster_iam_master) // Only the master does this.
return;
+ config->cluster_table_version += config->cluster_num_changes;
+
// Fill out the heartbeat header.
memset(&h, 0, sizeof(h));
h.size_tunn = sizeof(tunnelt);
h.interval = config->cluster_hb_interval;
h.timeout = config->cluster_hb_timeout;
+ h.table_version = config->cluster_table_version;
- add_type(&p, C_HEARTBEAT, HB_VERSION, (char*) &h, sizeof(h));
+ add_type(&p, C_HEARTBEAT, HB_VERSION, (char *) &h, sizeof(h));
for (i = 0; i < config->cluster_num_changes; ++i) {
hb_add_type(&p, cluster_changes[i].type, cluster_changes[i].id);
}
if (p > (buff + sizeof(buff))) { // Did we somehow manage to overun the buffer?
- LOG(0, 0, 0, "FATAL: Overran the heartbeat buffer! This is fatal. Exiting. (size %d)\n", p - buff);
+ LOG(0, 0, 0, "FATAL: Overran the heartbeat buffer! This is fatal. Exiting. (size %d)\n", (int) (p - buff));
kill(0, SIGTERM);
exit(1);
}
//
// Fill out the packet with sessions from the session table...
// (not forgetting to leave space so we can get some tunnels in too )
- while ( (p + sizeof(u32) * 2 + sizeof(sessiont) * 2 ) < (buff + MAX_HEART_SIZE) ) {
+ while ( (p + sizeof(uint32_t) * 2 + sizeof(sessiont) * 2 ) < (buff + MAX_HEART_SIZE) ) {
if (!walk_session_number) // session #0 isn't valid.
++walk_session_number;
// than the session table. This is good because stuffing up a
// tunnel is a much bigger deal than stuffing up a session.
//
- while ( (p + sizeof(u32) * 2 + sizeof(tunnelt) ) < (buff + MAX_HEART_SIZE) ) {
+ while ( (p + sizeof(uint32_t) * 2 + sizeof(tunnelt) ) < (buff + MAX_HEART_SIZE) ) {
if (!walk_tunnel_number) // tunnel #0 isn't valid.
++walk_tunnel_number;
//
// Did we do something wrong?
if (p > (buff + sizeof(buff))) { // Did we somehow manage to overun the buffer?
- LOG(0, 0, 0, "Overran the heartbeat buffer now! This is fatal. Exiting. (size %d)\n", p - buff);
+ LOG(0, 0, 0, "Overran the heartbeat buffer now! This is fatal. Exiting. (size %d)\n", (int) (p - buff));
kill(0, SIGTERM);
exit(1);
}
- LOG(3, 0, 0, "Sending heartbeat #%d with %d changes (%d x-sess, %d x-tunnels, %d highsess, %d hightun, size %d)\n",
- h.seq, config->cluster_num_changes, count, tcount, config->cluster_highest_sessionid,
- config->cluster_highest_tunnelid, (p-buff));
+ LOG(3, 0, 0, "Sending v%d heartbeat #%d, change #%" PRIu64 " with %d changes "
+ "(%d x-sess, %d x-tunnels, %d highsess, %d hightun, size %d)\n",
+ HB_VERSION, h.seq, h.table_version, config->cluster_num_changes,
+ count, tcount, config->cluster_highest_sessionid,
+ config->cluster_highest_tunnelid, (int) (p - buff));
config->cluster_num_changes = 0;
// missed a packet. We'll resend it every packet since
// the last one it's seen.
//
-static int cluster_catchup_slave(int seq, u32 slave)
+static int cluster_catchup_slave(int seq, in_addr_t slave)
{
int s;
int diff;
diff += HB_MAX_SEQ;
if (diff >= HB_HISTORY_SIZE) { // Ouch. We don't have the packet to send it!
- LOG(0, 0, 0, "A slaved asked for message %d when our seq number is %d. Killing it.\n",
+ LOG(0, 0, 0, "A slave asked for message %d when our seq number is %d. Killing it.\n",
seq, config->cluster_seq_number);
return peer_send_message(slave, C_KILL, seq, NULL, 0);// Kill the slave. Nothing else to do.
}
// We've heard from another peer! Add it to the list
// that we select from at election time.
//
-static int cluster_add_peer(u32 peer, time_t basetime, pingt *pp, int size)
+static int cluster_add_peer(in_addr_t peer, time_t basetime, pingt *pp, int size)
{
int i;
- u32 clusterid;
+ in_addr_t clusterid;
pingt p;
// Allow for backward compatability.
if (size > sizeof(p))
size = sizeof(p);
- memset( (void*) &p, 0, sizeof(p) );
- memcpy( (void*) &p, (void*) pp, size);
+ memset( (void *) &p, 0, sizeof(p) );
+ memcpy( (void *) &p, (void *) pp, size);
clusterid = p.addr;
if (clusterid != config->bind_address)
// Note that we don't mark the session as dirty; We rely on
// the slow table walk to propogate this back out to the slaves.
//
-static int cluster_handle_bytes(char * data, int size)
+static int cluster_handle_bytes(char *data, int size)
{
- bytest * b;
+ bytest *b;
- b = (bytest*) data;
+ b = (bytest *) data;
LOG(3, 0, 0, "Got byte counter update (size %d)\n", size);
//
// Handle receiving a session structure in a heartbeat packet.
//
-static int cluster_recv_session(int more , u8 * p)
+static int cluster_recv_session(int more, uint8_t *p)
{
if (more >= MAXSESSION) {
LOG(0, 0, 0, "DANGER: Received a heartbeat session id > MAXSESSION!\n");
}
}
- load_session(more, (sessiont*) p); // Copy session into session table..
+ load_session(more, (sessiont *) p); // Copy session into session table..
LOG(5, more, 0, "Received session update (%d undef)\n", config->cluster_undefined_sessions);
return 0;
}
-static int cluster_recv_tunnel(int more, u8 *p)
+static int cluster_recv_tunnel(int more, uint8_t *p)
{
if (more >= MAXTUNNEL) {
LOG(0, 0, 0, "DANGER: Received a tunnel session id > MAXTUNNEL!\n");
//
// Process a heartbeat..
//
-static int cluster_process_heartbeat(u8 * data, int size, int more, u8 * p, u32 addr)
+// v3: added interval, timeout
+// v4: added table_version
+static int cluster_process_heartbeat(uint8_t *data, int size, int more, uint8_t *p, in_addr_t addr)
{
- heartt * h;
+ heartt *h;
int s = size - (p-data);
int i, type;
-#if HB_VERSION != 3
+#if HB_VERSION != 4
# error "need to update cluster_process_heartbeat()"
#endif
- // we handle version 2+
- if (more < 2 || more > HB_VERSION) {
+ // we handle versions 3 through 4
+ if (more < 3 || more > HB_VERSION) {
LOG(0, 0, 0, "Received a heartbeat version that I don't support (%d)!\n", more);
return -1; // Ignore it??
}
if (s < sizeof(*h))
goto shortpacket;
- h = (heartt*) p;
+ h = (heartt *) p;
p += sizeof(*h);
s -= sizeof(*h);
if (config->cluster_iam_master) { // Sanity...
// Note that this MUST match the election process above!
- LOG(0, 0, 0, "I just got a packet claiming to be from a master but _I_ am the master!\n");
+ LOG(0, 0, 0, "I just got a heartbeat from master %s, but _I_ am the master!\n", fmtaddr(addr, 0));
if (!h->basetime) {
- LOG(0, 0, 0, "Heartbeat from addr %s with zero basetime!\n", fmtaddr(addr, 0));
+ LOG(0, 0, 0, "Heartbeat with zero basetime! Ignoring\n");
return -1; // Skip it.
}
+
+ if (more >= 4) {
+ if (h->table_version > config->cluster_table_version) {
+ LOG(0, 0, 0, "They've seen more state changes (%" PRIu64 " vs my %" PRIu64 ") so I'm gone!\n",
+ h->table_version, config->cluster_table_version);
+
+ kill(0, SIGTERM);
+ exit(1);
+ }
+ if (h->table_version < config->cluster_table_version)
+ return -1;
+ }
+
if (basetime > h->basetime) {
- LOG(0, 0, 0, "They're (%s) an older master than me so I'm gone!\n", fmtaddr(addr, 0));
+ LOG(0, 0, 0, "They're an older master than me so I'm gone!\n");
kill(0, SIGTERM);
exit(1);
}
- if (basetime == h->basetime && my_address < addr) { // Tie breaker.
+
+ if (basetime < h->basetime)
+ return -1;
+
+ if (my_address < addr) { // Tie breaker.
LOG(0, 0, 0, "They're a higher IP address than me, so I'm gone!\n");
kill(0, SIGTERM);
exit(1);
}
+
return -1; // Skip it.
}
// that the free session pointer is correct.
cluster_check_sessions(h->highsession, h->freesession, h->hightunnel);
- if (more > 2) // reserved section of heartt was not initialized prior to v3
+ if (h->interval != config->cluster_hb_interval)
{
- if (h->interval != config->cluster_hb_interval)
- {
- LOG(2, 0, 0, "Master set ping/heartbeat interval to %u (was %u)\n",
- h->interval, config->cluster_hb_interval);
+ LOG(2, 0, 0, "Master set ping/heartbeat interval to %u (was %u)\n",
+ h->interval, config->cluster_hb_interval);
- config->cluster_hb_interval = h->interval;
- }
+ config->cluster_hb_interval = h->interval;
+ }
- if (h->timeout != config->cluster_hb_timeout)
- {
- LOG(2, 0, 0, "Master set heartbeat timeout to %u (was %u)\n",
- h->timeout, config->cluster_hb_timeout);
+ if (h->timeout != config->cluster_hb_timeout)
+ {
+ LOG(2, 0, 0, "Master set heartbeat timeout to %u (was %u)\n",
+ h->timeout, config->cluster_hb_timeout);
- config->cluster_hb_timeout = h->timeout;
- }
+ config->cluster_hb_timeout = h->timeout;
}
// Ok. process the packet...
while ( s > 0) {
- type = * ((u32*) p);
- p += sizeof(u32);
- s -= sizeof(u32);
+ type = *((uint32_t *) p);
+ p += sizeof(uint32_t);
+ s -= sizeof(uint32_t);
- more = * ((u32*) p);
- p += sizeof(u32);
- s -= sizeof(u32);
+ more = *((uint32_t *) p);
+ p += sizeof(uint32_t);
+ s -= sizeof(uint32_t);
switch (type) {
case C_CSESSION: { // Compressed session structure.
- u8 c [ sizeof(sessiont) + 2];
+ uint8_t c[ sizeof(sessiont) + 2];
int size;
- u8 * orig_p = p;
+ uint8_t *orig_p = p;
- size = rle_decompress((u8 **) &p, s, c, sizeof(c) );
+ size = rle_decompress((uint8_t **) &p, s, c, sizeof(c) );
s -= (p - orig_p);
if (size != sizeof(sessiont) ) { // Ouch! Very very bad!
break;
case C_CTUNNEL: { // Compressed tunnel structure.
- u8 c [ sizeof(tunnelt) + 2];
+ uint8_t c[ sizeof(tunnelt) + 2];
int size;
- u8 * orig_p = p;
+ uint8_t *orig_p = p;
- size = rle_decompress( (u8 **) &p, s, c, sizeof(c) );
+ size = rle_decompress((uint8_t **) &p, s, c, sizeof(c));
s -= (p - orig_p);
if (size != sizeof(tunnelt) ) { // Ouch! Very very bad!
}
config->cluster_last_hb = TIME; // Successfully received a heartbeat!
+ config->cluster_table_version = h->table_version;
return 0;
shortpacket:
// We got a packet on the cluster port!
// Handle pings, lastseens, and heartbeats!
//
-int processcluster(char * data, int size, u32 addr)
+int processcluster(char *data, int size, in_addr_t addr)
{
int type, more;
- char * p = data;
+ char *p = data;
int s = size;
if (addr == my_address)
if (s < 8)
goto shortpacket;
- type = * ((u32*) p);
- p += sizeof(u32);
- s -= sizeof(u32);
+ type = *((uint32_t *) p);
+ p += sizeof(uint32_t);
+ s -= sizeof(uint32_t);
- more = * ((u32*) p);
- p += sizeof(u32);
- s -= sizeof(u32);
+ more = *((uint32_t *) p);
+ p += sizeof(uint32_t);
+ s -= sizeof(uint32_t);
switch (type) {
case C_PING: // Update the peers table.
- return cluster_add_peer(addr, more, (pingt*)p, s);
+ return cluster_add_peer(addr, more, (pingt *) p, s);
case C_LASTSEEN: // Catch up a slave (slave missed a packet).
return cluster_catchup_slave(more, addr);
struct sockaddr_in a;
a.sin_addr.s_addr = more;
- a.sin_port = * (int*) p;
+ a.sin_port = *(int *) p;
s -= sizeof(int);
p += sizeof(int);
: "Not defined",
0.1 * (TIME - config->cluster_last_hb));
cli_print(cli, "Uptodate : %s", config->cluster_iam_uptodate ? "Yes" : "No");
+ cli_print(cli, "Table version # : %llu", config->cluster_table_version);
cli_print(cli, "Next sequence number expected: %d", config->cluster_seq_number);
cli_print(cli, "%d sessions undefined of %d", config->cluster_undefined_sessions, config->cluster_highest_sessionid);
cli_print(cli, "%d tunnels undefined of %d", config->cluster_undefined_tunnels, config->cluster_highest_tunnelid);
} else {
+ cli_print(cli, "Table version # : %llu", config->cluster_table_version);
cli_print(cli, "Next heartbeat # : %d", config->cluster_seq_number);
cli_print(cli, "Highest session : %d", config->cluster_highest_sessionid);
cli_print(cli, "Highest tunnel : %d", config->cluster_highest_tunnelid);
//
// Worst case is a 50% expansion in space required (trying to
// compress { 0x00, 0x01 } * N )
-static int rle_compress(u8 ** src_p, int ssize, u8 *dst, int dsize)
+static int rle_compress(uint8_t **src_p, int ssize, uint8_t *dst, int dsize)
{
int count;
int orig_dsize = dsize;
- u8 * x,*src;
+ uint8_t *x, *src;
src = *src_p;
while (ssize > 0 && dsize > 2) {
// Return the number of dst bytes used.
// Updates the 'src_p' pointer to point to the
// first un-used byte.
-static int rle_decompress(u8 ** src_p, int ssize, u8 *dst, int dsize)
+static int rle_decompress(uint8_t **src_p, int ssize, uint8_t *dst, int dsize)
{
int count;
int orig_dsize = dsize;
- char * src = *src_p;
+ char *src = *src_p;
while (ssize >0 && dsize > 0) { // While there's more to decompress, and there's room in the decompress buffer...
count = *src++; --ssize; // get the count byte from the source.
projects / l2tpns.git / blobdiff