2 * Copyright (c) 2016 HUAWEI TECHNOLOGIES CO., LTD.
3 * Copyright (c) 2016 FUJITSU LIMITED
4 * Copyright (c) 2016 Intel Corporation
6 * Author: Zhang Chen <zhangchen.fnst@cn.fujitsu.com>
8 * This work is licensed under the terms of the GNU GPL, version 2 or
9 * later. See the COPYING file in the top-level directory.
12 #include "qemu/osdep.h"
15 #include "net/filter.h"
17 #include "qemu-common.h"
18 #include "qapi/error.h"
19 #include "qapi/qmp/qerror.h"
20 #include "qapi-visit.h"
21 #include "qom/object.h"
22 #include "qemu/main-loop.h"
24 #include "net/checksum.h"
26 #define FILTER_COLO_REWRITER(obj) \
27 OBJECT_CHECK(RewriterState, (obj), TYPE_FILTER_REWRITER)
29 #define TYPE_FILTER_REWRITER "filter-rewriter"
31 typedef struct RewriterState
{
32 NetFilterState parent_obj
;
33 NetQueue
*incoming_queue
;
34 /* hashtable to save connection */
35 GHashTable
*connection_track_table
;
38 static void filter_rewriter_flush(NetFilterState
*nf
)
40 RewriterState
*s
= FILTER_COLO_REWRITER(nf
);
42 if (!qemu_net_queue_flush(s
->incoming_queue
)) {
43 /* Unable to empty the queue, purge remaining packets */
44 qemu_net_queue_purge(s
->incoming_queue
, nf
->netdev
);
49 * Return 1 on success, if return 0 means the pkt
52 static int is_tcp_packet(Packet
*pkt
)
54 if (!parse_packet_early(pkt
) &&
55 pkt
->ip
->ip_p
== IPPROTO_TCP
) {
62 /* handle tcp packet from primary guest */
63 static int handle_primary_tcp_pkt(NetFilterState
*nf
,
67 struct tcphdr
*tcp_pkt
;
69 tcp_pkt
= (struct tcphdr
*)pkt
->transport_header
;
70 if (trace_event_get_state(TRACE_COLO_FILTER_REWRITER_DEBUG
)) {
71 char *sdebug
, *ddebug
;
72 sdebug
= strdup(inet_ntoa(pkt
->ip
->ip_src
));
73 ddebug
= strdup(inet_ntoa(pkt
->ip
->ip_dst
));
74 trace_colo_filter_rewriter_pkt_info(__func__
, sdebug
, ddebug
,
75 ntohl(tcp_pkt
->th_seq
), ntohl(tcp_pkt
->th_ack
),
77 trace_colo_filter_rewriter_conn_offset(conn
->offset
);
82 if (((tcp_pkt
->th_flags
& (TH_ACK
| TH_SYN
)) == TH_SYN
)) {
84 * we use this flag update offset func
85 * run once in independent tcp connection
90 if (((tcp_pkt
->th_flags
& (TH_ACK
| TH_SYN
)) == TH_ACK
)) {
93 * offset = secondary_seq - primary seq
94 * ack packet sent by guest from primary node,
95 * so we use th_ack - 1 get primary_seq
97 conn
->offset
-= (ntohl(tcp_pkt
->th_ack
) - 1);
100 /* handle packets to the secondary from the primary */
101 tcp_pkt
->th_ack
= htonl(ntohl(tcp_pkt
->th_ack
) + conn
->offset
);
103 net_checksum_calculate((uint8_t *)pkt
->data
, pkt
->size
);
109 /* handle tcp packet from secondary guest */
110 static int handle_secondary_tcp_pkt(NetFilterState
*nf
,
114 struct tcphdr
*tcp_pkt
;
116 tcp_pkt
= (struct tcphdr
*)pkt
->transport_header
;
118 if (trace_event_get_state(TRACE_COLO_FILTER_REWRITER_DEBUG
)) {
119 char *sdebug
, *ddebug
;
120 sdebug
= strdup(inet_ntoa(pkt
->ip
->ip_src
));
121 ddebug
= strdup(inet_ntoa(pkt
->ip
->ip_dst
));
122 trace_colo_filter_rewriter_pkt_info(__func__
, sdebug
, ddebug
,
123 ntohl(tcp_pkt
->th_seq
), ntohl(tcp_pkt
->th_ack
),
125 trace_colo_filter_rewriter_conn_offset(conn
->offset
);
130 if (((tcp_pkt
->th_flags
& (TH_ACK
| TH_SYN
)) == (TH_ACK
| TH_SYN
))) {
132 * save offset = secondary_seq and then
133 * in handle_primary_tcp_pkt make offset
134 * = secondary_seq - primary_seq
136 conn
->offset
= ntohl(tcp_pkt
->th_seq
);
139 if ((tcp_pkt
->th_flags
& (TH_ACK
| TH_SYN
)) == TH_ACK
) {
140 /* handle packets to the primary from the secondary*/
141 tcp_pkt
->th_seq
= htonl(ntohl(tcp_pkt
->th_seq
) - conn
->offset
);
143 net_checksum_calculate((uint8_t *)pkt
->data
, pkt
->size
);
149 static ssize_t
colo_rewriter_receive_iov(NetFilterState
*nf
,
150 NetClientState
*sender
,
152 const struct iovec
*iov
,
154 NetPacketSent
*sent_cb
)
156 RewriterState
*s
= FILTER_COLO_REWRITER(nf
);
160 ssize_t size
= iov_size(iov
, iovcnt
);
161 char *buf
= g_malloc0(size
);
163 iov_to_buf(iov
, iovcnt
, 0, buf
, size
);
164 pkt
= packet_new(buf
, size
);
167 * if we get tcp packet
168 * we will rewrite it to make secondary guest's
169 * connection established successfully
171 if (pkt
&& is_tcp_packet(pkt
)) {
173 fill_connection_key(pkt
, &key
);
175 if (sender
== nf
->netdev
) {
177 * We need make tcp TX and RX packet
178 * into one connection.
180 reverse_connection_key(&key
);
182 conn
= connection_get(s
->connection_track_table
,
186 if (sender
== nf
->netdev
) {
187 /* NET_FILTER_DIRECTION_TX */
188 if (!handle_primary_tcp_pkt(nf
, conn
, pkt
)) {
189 qemu_net_queue_send(s
->incoming_queue
, sender
, 0,
190 (const uint8_t *)pkt
->data
, pkt
->size
, NULL
);
191 packet_destroy(pkt
, NULL
);
194 * We block the packet here,after rewrite pkt
200 /* NET_FILTER_DIRECTION_RX */
201 if (!handle_secondary_tcp_pkt(nf
, conn
, pkt
)) {
202 qemu_net_queue_send(s
->incoming_queue
, sender
, 0,
203 (const uint8_t *)pkt
->data
, pkt
->size
, NULL
);
204 packet_destroy(pkt
, NULL
);
207 * We block the packet here,after rewrite pkt
215 packet_destroy(pkt
, NULL
);
220 static void colo_rewriter_cleanup(NetFilterState
*nf
)
222 RewriterState
*s
= FILTER_COLO_REWRITER(nf
);
225 if (s
->incoming_queue
) {
226 filter_rewriter_flush(nf
);
227 g_free(s
->incoming_queue
);
231 static void colo_rewriter_setup(NetFilterState
*nf
, Error
**errp
)
233 RewriterState
*s
= FILTER_COLO_REWRITER(nf
);
235 s
->connection_track_table
= g_hash_table_new_full(connection_key_hash
,
236 connection_key_equal
,
239 s
->incoming_queue
= qemu_new_net_queue(qemu_netfilter_pass_to_next
, nf
);
242 static void colo_rewriter_class_init(ObjectClass
*oc
, void *data
)
244 NetFilterClass
*nfc
= NETFILTER_CLASS(oc
);
246 nfc
->setup
= colo_rewriter_setup
;
247 nfc
->cleanup
= colo_rewriter_cleanup
;
248 nfc
->receive_iov
= colo_rewriter_receive_iov
;
251 static const TypeInfo colo_rewriter_info
= {
252 .name
= TYPE_FILTER_REWRITER
,
253 .parent
= TYPE_NETFILTER
,
254 .class_init
= colo_rewriter_class_init
,
255 .instance_size
= sizeof(RewriterState
),
258 static void register_types(void)
260 type_register_static(&colo_rewriter_info
);
263 type_init(register_types
);