hw/rdma: PVRDMA commands and data-path ops
[qemu.git] / net / filter-rewriter.c
blob6201494ceb83d067a5d236ea7d03ec3e6b65469e
1 /*
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"
13 #include "trace.h"
14 #include "net/colo.h"
15 #include "net/filter.h"
16 #include "net/net.h"
17 #include "qemu-common.h"
18 #include "qemu/error-report.h"
19 #include "qapi-visit.h"
20 #include "qom/object.h"
21 #include "qemu/main-loop.h"
22 #include "qemu/iov.h"
23 #include "net/checksum.h"
25 #define FILTER_COLO_REWRITER(obj) \
26 OBJECT_CHECK(RewriterState, (obj), TYPE_FILTER_REWRITER)
28 #define TYPE_FILTER_REWRITER "filter-rewriter"
30 typedef struct RewriterState {
31 NetFilterState parent_obj;
32 NetQueue *incoming_queue;
33 /* hashtable to save connection */
34 GHashTable *connection_track_table;
35 bool vnet_hdr;
36 } RewriterState;
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
50 * is not TCP packet
52 static int is_tcp_packet(Packet *pkt)
54 if (!parse_packet_early(pkt) &&
55 pkt->ip->ip_p == IPPROTO_TCP) {
56 return 1;
57 } else {
58 return 0;
62 /* handle tcp packet from primary guest */
63 static int handle_primary_tcp_pkt(NetFilterState *nf,
64 Connection *conn,
65 Packet *pkt)
67 struct tcphdr *tcp_pkt;
69 tcp_pkt = (struct tcphdr *)pkt->transport_header;
70 if (trace_event_get_state_backends(TRACE_COLO_FILTER_REWRITER_DEBUG)) {
71 trace_colo_filter_rewriter_pkt_info(__func__,
72 inet_ntoa(pkt->ip->ip_src), inet_ntoa(pkt->ip->ip_dst),
73 ntohl(tcp_pkt->th_seq), ntohl(tcp_pkt->th_ack),
74 tcp_pkt->th_flags);
75 trace_colo_filter_rewriter_conn_offset(conn->offset);
78 if (((tcp_pkt->th_flags & (TH_ACK | TH_SYN)) == TH_SYN)) {
80 * we use this flag update offset func
81 * run once in independent tcp connection
83 conn->syn_flag = 1;
86 if (((tcp_pkt->th_flags & (TH_ACK | TH_SYN)) == TH_ACK)) {
87 if (conn->syn_flag) {
89 * offset = secondary_seq - primary seq
90 * ack packet sent by guest from primary node,
91 * so we use th_ack - 1 get primary_seq
93 conn->offset -= (ntohl(tcp_pkt->th_ack) - 1);
94 conn->syn_flag = 0;
96 if (conn->offset) {
97 /* handle packets to the secondary from the primary */
98 tcp_pkt->th_ack = htonl(ntohl(tcp_pkt->th_ack) + conn->offset);
100 net_checksum_calculate((uint8_t *)pkt->data + pkt->vnet_hdr_len,
101 pkt->size - pkt->vnet_hdr_len);
105 return 0;
108 /* handle tcp packet from secondary guest */
109 static int handle_secondary_tcp_pkt(NetFilterState *nf,
110 Connection *conn,
111 Packet *pkt)
113 struct tcphdr *tcp_pkt;
115 tcp_pkt = (struct tcphdr *)pkt->transport_header;
117 if (trace_event_get_state_backends(TRACE_COLO_FILTER_REWRITER_DEBUG)) {
118 trace_colo_filter_rewriter_pkt_info(__func__,
119 inet_ntoa(pkt->ip->ip_src), inet_ntoa(pkt->ip->ip_dst),
120 ntohl(tcp_pkt->th_seq), ntohl(tcp_pkt->th_ack),
121 tcp_pkt->th_flags);
122 trace_colo_filter_rewriter_conn_offset(conn->offset);
125 if (((tcp_pkt->th_flags & (TH_ACK | TH_SYN)) == (TH_ACK | TH_SYN))) {
127 * save offset = secondary_seq and then
128 * in handle_primary_tcp_pkt make offset
129 * = secondary_seq - primary_seq
131 conn->offset = ntohl(tcp_pkt->th_seq);
134 if ((tcp_pkt->th_flags & (TH_ACK | TH_SYN)) == TH_ACK) {
135 /* Only need to adjust seq while offset is Non-zero */
136 if (conn->offset) {
137 /* handle packets to the primary from the secondary*/
138 tcp_pkt->th_seq = htonl(ntohl(tcp_pkt->th_seq) - conn->offset);
140 net_checksum_calculate((uint8_t *)pkt->data + pkt->vnet_hdr_len,
141 pkt->size - pkt->vnet_hdr_len);
145 return 0;
148 static ssize_t colo_rewriter_receive_iov(NetFilterState *nf,
149 NetClientState *sender,
150 unsigned flags,
151 const struct iovec *iov,
152 int iovcnt,
153 NetPacketSent *sent_cb)
155 RewriterState *s = FILTER_COLO_REWRITER(nf);
156 Connection *conn;
157 ConnectionKey key;
158 Packet *pkt;
159 ssize_t size = iov_size(iov, iovcnt);
160 ssize_t vnet_hdr_len = 0;
161 char *buf = g_malloc0(size);
163 iov_to_buf(iov, iovcnt, 0, buf, size);
165 if (s->vnet_hdr) {
166 vnet_hdr_len = nf->netdev->vnet_hdr_len;
169 pkt = packet_new(buf, size, vnet_hdr_len);
170 g_free(buf);
173 * if we get tcp packet
174 * we will rewrite it to make secondary guest's
175 * connection established successfully
177 if (pkt && is_tcp_packet(pkt)) {
179 fill_connection_key(pkt, &key);
181 if (sender == nf->netdev) {
183 * We need make tcp TX and RX packet
184 * into one connection.
186 reverse_connection_key(&key);
188 conn = connection_get(s->connection_track_table,
189 &key,
190 NULL);
192 if (sender == nf->netdev) {
193 /* NET_FILTER_DIRECTION_TX */
194 if (!handle_primary_tcp_pkt(nf, conn, pkt)) {
195 qemu_net_queue_send(s->incoming_queue, sender, 0,
196 (const uint8_t *)pkt->data, pkt->size, NULL);
197 packet_destroy(pkt, NULL);
198 pkt = NULL;
200 * We block the packet here,after rewrite pkt
201 * and will send it
203 return 1;
205 } else {
206 /* NET_FILTER_DIRECTION_RX */
207 if (!handle_secondary_tcp_pkt(nf, conn, pkt)) {
208 qemu_net_queue_send(s->incoming_queue, sender, 0,
209 (const uint8_t *)pkt->data, pkt->size, NULL);
210 packet_destroy(pkt, NULL);
211 pkt = NULL;
213 * We block the packet here,after rewrite pkt
214 * and will send it
216 return 1;
221 packet_destroy(pkt, NULL);
222 pkt = NULL;
223 return 0;
226 static void colo_rewriter_cleanup(NetFilterState *nf)
228 RewriterState *s = FILTER_COLO_REWRITER(nf);
230 /* flush packets */
231 if (s->incoming_queue) {
232 filter_rewriter_flush(nf);
233 g_free(s->incoming_queue);
237 static void colo_rewriter_setup(NetFilterState *nf, Error **errp)
239 RewriterState *s = FILTER_COLO_REWRITER(nf);
241 s->connection_track_table = g_hash_table_new_full(connection_key_hash,
242 connection_key_equal,
243 g_free,
244 connection_destroy);
245 s->incoming_queue = qemu_new_net_queue(qemu_netfilter_pass_to_next, nf);
248 static bool filter_rewriter_get_vnet_hdr(Object *obj, Error **errp)
250 RewriterState *s = FILTER_COLO_REWRITER(obj);
252 return s->vnet_hdr;
255 static void filter_rewriter_set_vnet_hdr(Object *obj,
256 bool value,
257 Error **errp)
259 RewriterState *s = FILTER_COLO_REWRITER(obj);
261 s->vnet_hdr = value;
264 static void filter_rewriter_init(Object *obj)
266 RewriterState *s = FILTER_COLO_REWRITER(obj);
268 s->vnet_hdr = false;
269 object_property_add_bool(obj, "vnet_hdr_support",
270 filter_rewriter_get_vnet_hdr,
271 filter_rewriter_set_vnet_hdr, NULL);
274 static void colo_rewriter_class_init(ObjectClass *oc, void *data)
276 NetFilterClass *nfc = NETFILTER_CLASS(oc);
278 nfc->setup = colo_rewriter_setup;
279 nfc->cleanup = colo_rewriter_cleanup;
280 nfc->receive_iov = colo_rewriter_receive_iov;
283 static const TypeInfo colo_rewriter_info = {
284 .name = TYPE_FILTER_REWRITER,
285 .parent = TYPE_NETFILTER,
286 .class_init = colo_rewriter_class_init,
287 .instance_init = filter_rewriter_init,
288 .instance_size = sizeof(RewriterState),
291 static void register_types(void)
293 type_register_static(&colo_rewriter_info);
296 type_init(register_types);