qemu-e2k/net/colo.c

232 lines
5.8 KiB
C
Raw Normal View History

/*
* COarse-grain LOck-stepping Virtual Machines for Non-stop Service (COLO)
* (a.k.a. Fault Tolerance or Continuous Replication)
*
* Copyright (c) 2016 HUAWEI TECHNOLOGIES CO., LTD.
* Copyright (c) 2016 FUJITSU LIMITED
* Copyright (c) 2016 Intel Corporation
*
* Author: Zhang Chen <zhangchen.fnst@cn.fujitsu.com>
*
* This work is licensed under the terms of the GNU GPL, version 2 or
* later. See the COPYING file in the top-level directory.
*/
#include "qemu/osdep.h"
#include "trace.h"
#include "colo.h"
uint32_t connection_key_hash(const void *opaque)
{
const ConnectionKey *key = opaque;
uint32_t a, b, c;
/* Jenkins hash */
a = b = c = JHASH_INITVAL + sizeof(*key);
a += key->src.s_addr;
b += key->dst.s_addr;
c += (key->src_port | key->dst_port << 16);
__jhash_mix(a, b, c);
a += key->ip_proto;
__jhash_final(a, b, c);
return c;
}
int connection_key_equal(const void *key1, const void *key2)
{
return memcmp(key1, key2, sizeof(ConnectionKey)) == 0;
}
int parse_packet_early(Packet *pkt)
{
int network_length;
static const uint8_t vlan[] = {0x81, 0x00};
uint8_t *data = pkt->data + pkt->vnet_hdr_len;
uint16_t l3_proto;
ssize_t l2hdr_len = eth_get_l2_hdr_length(data);
if (pkt->size < ETH_HLEN + pkt->vnet_hdr_len) {
trace_colo_proxy_main("pkt->size < ETH_HLEN");
return 1;
}
/*
* TODO: support vlan.
*/
if (!memcmp(&data[12], vlan, sizeof(vlan))) {
trace_colo_proxy_main("COLO-proxy don't support vlan");
return 1;
}
pkt->network_header = data + l2hdr_len;
const struct iovec l2vec = {
.iov_base = (void *) data,
.iov_len = l2hdr_len
};
l3_proto = eth_get_l3_proto(&l2vec, 1, l2hdr_len);
if (l3_proto != ETH_P_IP) {
return 1;
}
network_length = pkt->ip->ip_hl * 4;
if (pkt->size < l2hdr_len + network_length + pkt->vnet_hdr_len) {
trace_colo_proxy_main("pkt->size < network_header + network_length");
return 1;
}
pkt->transport_header = pkt->network_header + network_length;
return 0;
}
void extract_ip_and_port(uint32_t tmp_ports, ConnectionKey *key, Packet *pkt)
{
key->src = pkt->ip->ip_src;
key->dst = pkt->ip->ip_dst;
key->src_port = ntohs(tmp_ports >> 16);
key->dst_port = ntohs(tmp_ports & 0xffff);
}
void fill_connection_key(Packet *pkt, ConnectionKey *key)
{
uint32_t tmp_ports;
memset(key, 0, sizeof(*key));
key->ip_proto = pkt->ip->ip_p;
switch (key->ip_proto) {
case IPPROTO_TCP:
case IPPROTO_UDP:
case IPPROTO_DCCP:
case IPPROTO_ESP:
case IPPROTO_SCTP:
case IPPROTO_UDPLITE:
tmp_ports = *(uint32_t *)(pkt->transport_header);
extract_ip_and_port(tmp_ports, key, pkt);
break;
case IPPROTO_AH:
tmp_ports = *(uint32_t *)(pkt->transport_header + 4);
extract_ip_and_port(tmp_ports, key, pkt);
break;
default:
break;
}
}
void reverse_connection_key(ConnectionKey *key)
{
struct in_addr tmp_ip;
uint16_t tmp_port;
tmp_ip = key->src;
key->src = key->dst;
key->dst = tmp_ip;
tmp_port = key->src_port;
key->src_port = key->dst_port;
key->dst_port = tmp_port;
}
Connection *connection_new(ConnectionKey *key)
{
Connection *conn = g_slice_new(Connection);
conn->ip_proto = key->ip_proto;
conn->processing = false;
filter-rewriter: rewrite tcp packet to keep secondary connection We will rewrite tcp packet secondary received and sent. When colo guest is a tcp server. Firstly, client start a tcp handshake. the packet's seq=client_seq, ack=0,flag=SYN. COLO primary guest get this pkt and mirror(filter-mirror) to secondary guest, secondary get it use filter-redirector. Then,primary guest response pkt (seq=primary_seq,ack=client_seq+1,flag=ACK|SYN). secondary guest response pkt (seq=secondary_seq,ack=client_seq+1,flag=ACK|SYN). In here,we use filter-rewriter save the secondary_seq to it's tcp connection. Finally handshake,client send pkt (seq=client_seq+1,ack=primary_seq+1,flag=ACK). Here,filter-rewriter can get primary_seq, and rewrite ack from primary_seq+1 to secondary_seq+1, recalculate checksum. So the secondary tcp connection kept good. When we send/recv packet. client send pkt(seq=client_seq+1+data_len,ack=primary_seq+1,flag=ACK|PSH). filter-rewriter rewrite ack and send to secondary guest. primary guest response pkt (seq=primary_seq+1,ack=client_seq+1+data_len,flag=ACK) secondary guest response pkt (seq=secondary_seq+1,ack=client_seq+1+data_len,flag=ACK) we rewrite secondary guest seq from secondary_seq+1 to primary_seq+1. So tcp connection kept good. In code We use offset( = secondary_seq - primary_seq ) to rewrite seq or ack. handle_primary_tcp_pkt: tcp_pkt->th_ack += offset; handle_secondary_tcp_pkt: tcp_pkt->th_seq -= offset; Signed-off-by: Zhang Chen <zhangchen.fnst@cn.fujitsu.com> Signed-off-by: Li Zhijian <lizhijian@cn.fujitsu.com> Signed-off-by: Wen Congyang <wency@cn.fujitsu.com> Signed-off-by: Jason Wang <jasowang@redhat.com>
2016-09-27 04:22:34 +02:00
conn->offset = 0;
conn->tcp_state = TCPS_CLOSED;
colo: compare the packet based on the tcp sequence number Packet size some time different or when network is busy. Based on same payload size, but TCP protocol can not guarantee send the same one packet in the same way, like that: We send this payload: ------------------------------ | header |1|2|3|4|5|6|7|8|9|0| ------------------------------ primary: ppkt1: ---------------- | header |1|2|3| ---------------- ppkt2: ------------------------ | header |4|5|6|7|8|9|0| ------------------------ secondary: spkt1: ------------------------------ | header |1|2|3|4|5|6|7|8|9|0| ------------------------------ In the original method, ppkt1 and ppkt2 are different in size and spkt1, so they can't compare and trigger the checkpoint. I have tested FTP get 200M and 1G file many times, I found that the performance was less than 1% of the native. Now I reconstructed the comparison of TCP packets based on the TCP sequence number. first of all, ppkt1 and spkt1 have the same starting sequence number, so they can compare, even though their length is different. And then ppkt1 with a smaller payload length is used as the comparison length, if the payload is same, send out the ppkt1 and record the offset(the length of ppkt1 payload) in spkt1. The next comparison, ppkt2 and spkt1 can be compared from the recorded position of spkt1. like that: ---------------- | header |1|2|3| ppkt1 ---------|-----| | | ---------v-----v-------------- | header |1|2|3|4|5|6|7|8|9|0| spkt1 ---------------|\------------| | \offset | ---------v-------------v | header |4|5|6|7|8|9|0| ppkt2 ------------------------ In this way, the performance can reach native 20% in my multiple tests. Cc: Zhang Chen <zhangckid@gmail.com> Cc: Li Zhijian <lizhijian@cn.fujitsu.com> Cc: Jason Wang <jasowang@redhat.com> Signed-off-by: Mao Zhongyi <maozy.fnst@cn.fujitsu.com> Signed-off-by: Li Zhijian <lizhijian@cn.fujitsu.com> Signed-off-by: Zhang Chen <zhangckid@gmail.com> Reviewed-by: Zhang Chen <zhangckid@gmail.com> Tested-by: Zhang Chen <zhangckid@gmail.com> Signed-off-by: Jason Wang <jasowang@redhat.com>
2017-12-25 03:54:12 +01:00
conn->pack = 0;
conn->sack = 0;
g_queue_init(&conn->primary_list);
g_queue_init(&conn->secondary_list);
return conn;
}
void connection_destroy(void *opaque)
{
Connection *conn = opaque;
g_queue_foreach(&conn->primary_list, packet_destroy, NULL);
g_queue_clear(&conn->primary_list);
g_queue_foreach(&conn->secondary_list, packet_destroy, NULL);
g_queue_clear(&conn->secondary_list);
g_slice_free(Connection, conn);
}
Packet *packet_new(const void *data, int size, int vnet_hdr_len)
{
Packet *pkt = g_slice_new(Packet);
pkt->data = g_memdup(data, size);
pkt->size = size;
pkt->creation_ms = qemu_clock_get_ms(QEMU_CLOCK_HOST);
pkt->vnet_hdr_len = vnet_hdr_len;
colo: compare the packet based on the tcp sequence number Packet size some time different or when network is busy. Based on same payload size, but TCP protocol can not guarantee send the same one packet in the same way, like that: We send this payload: ------------------------------ | header |1|2|3|4|5|6|7|8|9|0| ------------------------------ primary: ppkt1: ---------------- | header |1|2|3| ---------------- ppkt2: ------------------------ | header |4|5|6|7|8|9|0| ------------------------ secondary: spkt1: ------------------------------ | header |1|2|3|4|5|6|7|8|9|0| ------------------------------ In the original method, ppkt1 and ppkt2 are different in size and spkt1, so they can't compare and trigger the checkpoint. I have tested FTP get 200M and 1G file many times, I found that the performance was less than 1% of the native. Now I reconstructed the comparison of TCP packets based on the TCP sequence number. first of all, ppkt1 and spkt1 have the same starting sequence number, so they can compare, even though their length is different. And then ppkt1 with a smaller payload length is used as the comparison length, if the payload is same, send out the ppkt1 and record the offset(the length of ppkt1 payload) in spkt1. The next comparison, ppkt2 and spkt1 can be compared from the recorded position of spkt1. like that: ---------------- | header |1|2|3| ppkt1 ---------|-----| | | ---------v-----v-------------- | header |1|2|3|4|5|6|7|8|9|0| spkt1 ---------------|\------------| | \offset | ---------v-------------v | header |4|5|6|7|8|9|0| ppkt2 ------------------------ In this way, the performance can reach native 20% in my multiple tests. Cc: Zhang Chen <zhangckid@gmail.com> Cc: Li Zhijian <lizhijian@cn.fujitsu.com> Cc: Jason Wang <jasowang@redhat.com> Signed-off-by: Mao Zhongyi <maozy.fnst@cn.fujitsu.com> Signed-off-by: Li Zhijian <lizhijian@cn.fujitsu.com> Signed-off-by: Zhang Chen <zhangckid@gmail.com> Reviewed-by: Zhang Chen <zhangckid@gmail.com> Tested-by: Zhang Chen <zhangckid@gmail.com> Signed-off-by: Jason Wang <jasowang@redhat.com>
2017-12-25 03:54:12 +01:00
pkt->tcp_seq = 0;
pkt->tcp_ack = 0;
pkt->seq_end = 0;
pkt->header_size = 0;
pkt->payload_size = 0;
pkt->offset = 0;
pkt->flags = 0;
return pkt;
}
void packet_destroy(void *opaque, void *user_data)
{
Packet *pkt = opaque;
g_free(pkt->data);
g_slice_free(Packet, pkt);
}
/*
* Clear hashtable, stop this hash growing really huge
*/
void connection_hashtable_reset(GHashTable *connection_track_table)
{
g_hash_table_remove_all(connection_track_table);
}
/* if not found, create a new connection and add to hash table */
Connection *connection_get(GHashTable *connection_track_table,
ConnectionKey *key,
GQueue *conn_list)
{
Connection *conn = g_hash_table_lookup(connection_track_table, key);
if (conn == NULL) {
ConnectionKey *new_key = g_memdup(key, sizeof(*key));
conn = connection_new(key);
if (g_hash_table_size(connection_track_table) > HASHTABLE_MAX_SIZE) {
trace_colo_proxy_main("colo proxy connection hashtable full,"
" clear it");
connection_hashtable_reset(connection_track_table);
/*
* clear the conn_list
*/
while (!g_queue_is_empty(conn_list)) {
connection_destroy(g_queue_pop_head(conn_list));
}
}
g_hash_table_insert(connection_track_table, new_key, conn);
}
return conn;
}
bool connection_has_tracked(GHashTable *connection_track_table,
ConnectionKey *key)
{
Connection *conn = g_hash_table_lookup(connection_track_table, key);
return conn ? true : false;
}