xref: /openbmc/qemu/net/filter-rewriter.c (revision ab938ae4) 
1 /*
2  * Copyright (c) 2016 HUAWEI TECHNOLOGIES CO., LTD.
3  * Copyright (c) 2016 FUJITSU LIMITED
4  * Copyright (c) 2016 Intel Corporation
5  *
6  * Author: Zhang Chen <zhangchen.fnst@cn.fujitsu.com>
7  *
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.
10  */
11 
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 "qapi/error.h"
19 #include "qapi/qmp/qerror.h"
20 #include "qemu/error-report.h"
21 #include "qapi-visit.h"
22 #include "qom/object.h"
23 #include "qemu/main-loop.h"
24 #include "qemu/iov.h"
25 #include "net/checksum.h"
26 
27 #define FILTER_COLO_REWRITER(obj) \
28     OBJECT_CHECK(RewriterState, (obj), TYPE_FILTER_REWRITER)
29 
30 #define TYPE_FILTER_REWRITER "filter-rewriter"
31 
32 typedef struct RewriterState {
33     NetFilterState parent_obj;
34     NetQueue *incoming_queue;
35     /* hashtable to save connection */
36     GHashTable *connection_track_table;
37     bool vnet_hdr;
38 } RewriterState;
39 
40 static void filter_rewriter_flush(NetFilterState *nf)
41 {
42     RewriterState *s = FILTER_COLO_REWRITER(nf);
43 
44     if (!qemu_net_queue_flush(s->incoming_queue)) {
45         /* Unable to empty the queue, purge remaining packets */
46         qemu_net_queue_purge(s->incoming_queue, nf->netdev);
47     }
48 }
49 
50 /*
51  * Return 1 on success, if return 0 means the pkt
52  * is not TCP packet
53  */
54 static int is_tcp_packet(Packet *pkt)
55 {
56     if (!parse_packet_early(pkt) &&
57         pkt->ip->ip_p == IPPROTO_TCP) {
58         return 1;
59     } else {
60         return 0;
61     }
62 }
63 
64 /* handle tcp packet from primary guest */
65 static int handle_primary_tcp_pkt(NetFilterState *nf,
66                                   Connection *conn,
67                                   Packet *pkt)
68 {
69     struct tcphdr *tcp_pkt;
70 
71     tcp_pkt = (struct tcphdr *)pkt->transport_header;
72     if (trace_event_get_state_backends(TRACE_COLO_FILTER_REWRITER_DEBUG)) {
73         trace_colo_filter_rewriter_pkt_info(__func__,
74                     inet_ntoa(pkt->ip->ip_src), inet_ntoa(pkt->ip->ip_dst),
75                     ntohl(tcp_pkt->th_seq), ntohl(tcp_pkt->th_ack),
76                     tcp_pkt->th_flags);
77         trace_colo_filter_rewriter_conn_offset(conn->offset);
78     }
79 
80     if (((tcp_pkt->th_flags & (TH_ACK | TH_SYN)) == TH_SYN)) {
81         /*
82          * we use this flag update offset func
83          * run once in independent tcp connection
84          */
85         conn->syn_flag = 1;
86     }
87 
88     if (((tcp_pkt->th_flags & (TH_ACK | TH_SYN)) == TH_ACK)) {
89         if (conn->syn_flag) {
90             /*
91              * offset = secondary_seq - primary seq
92              * ack packet sent by guest from primary node,
93              * so we use th_ack - 1 get primary_seq
94              */
95             conn->offset -= (ntohl(tcp_pkt->th_ack) - 1);
96             conn->syn_flag = 0;
97         }
98         if (conn->offset) {
99             /* handle packets to the secondary from the primary */
100             tcp_pkt->th_ack = htonl(ntohl(tcp_pkt->th_ack) + conn->offset);
101 
102             net_checksum_calculate((uint8_t *)pkt->data + pkt->vnet_hdr_len,
103                                    pkt->size - pkt->vnet_hdr_len);
104         }
105     }
106 
107     return 0;
108 }
109 
110 /* handle tcp packet from secondary guest */
111 static int handle_secondary_tcp_pkt(NetFilterState *nf,
112                                     Connection *conn,
113                                     Packet *pkt)
114 {
115     struct tcphdr *tcp_pkt;
116 
117     tcp_pkt = (struct tcphdr *)pkt->transport_header;
118 
119     if (trace_event_get_state_backends(TRACE_COLO_FILTER_REWRITER_DEBUG)) {
120         trace_colo_filter_rewriter_pkt_info(__func__,
121                     inet_ntoa(pkt->ip->ip_src), inet_ntoa(pkt->ip->ip_dst),
122                     ntohl(tcp_pkt->th_seq), ntohl(tcp_pkt->th_ack),
123                     tcp_pkt->th_flags);
124         trace_colo_filter_rewriter_conn_offset(conn->offset);
125     }
126 
127     if (((tcp_pkt->th_flags & (TH_ACK | TH_SYN)) == (TH_ACK | TH_SYN))) {
128         /*
129          * save offset = secondary_seq and then
130          * in handle_primary_tcp_pkt make offset
131          * = secondary_seq - primary_seq
132          */
133         conn->offset = ntohl(tcp_pkt->th_seq);
134     }
135 
136     if ((tcp_pkt->th_flags & (TH_ACK | TH_SYN)) == TH_ACK) {
137         /* Only need to adjust seq while offset is Non-zero */
138         if (conn->offset) {
139             /* handle packets to the primary from the secondary*/
140             tcp_pkt->th_seq = htonl(ntohl(tcp_pkt->th_seq) - conn->offset);
141 
142             net_checksum_calculate((uint8_t *)pkt->data + pkt->vnet_hdr_len,
143                                    pkt->size - pkt->vnet_hdr_len);
144         }
145     }
146 
147     return 0;
148 }
149 
150 static ssize_t colo_rewriter_receive_iov(NetFilterState *nf,
151                                          NetClientState *sender,
152                                          unsigned flags,
153                                          const struct iovec *iov,
154                                          int iovcnt,
155                                          NetPacketSent *sent_cb)
156 {
157     RewriterState *s = FILTER_COLO_REWRITER(nf);
158     Connection *conn;
159     ConnectionKey key;
160     Packet *pkt;
161     ssize_t size = iov_size(iov, iovcnt);
162     ssize_t vnet_hdr_len = 0;
163     char *buf = g_malloc0(size);
164 
165     iov_to_buf(iov, iovcnt, 0, buf, size);
166 
167     if (s->vnet_hdr) {
168         vnet_hdr_len = nf->netdev->vnet_hdr_len;
169     }
170 
171     pkt = packet_new(buf, size, vnet_hdr_len);
172     g_free(buf);
173 
174     /*
175      * if we get tcp packet
176      * we will rewrite it to make secondary guest's
177      * connection established successfully
178      */
179     if (pkt && is_tcp_packet(pkt)) {
180 
181         fill_connection_key(pkt, &key);
182 
183         if (sender == nf->netdev) {
184             /*
185              * We need make tcp TX and RX packet
186              * into one connection.
187              */
188             reverse_connection_key(&key);
189         }
190         conn = connection_get(s->connection_track_table,
191                               &key,
192                               NULL);
193 
194         if (sender == nf->netdev) {
195             /* NET_FILTER_DIRECTION_TX */
196             if (!handle_primary_tcp_pkt(nf, conn, pkt)) {
197                 qemu_net_queue_send(s->incoming_queue, sender, 0,
198                 (const uint8_t *)pkt->data, pkt->size, NULL);
199                 packet_destroy(pkt, NULL);
200                 pkt = NULL;
201                 /*
202                  * We block the packet here,after rewrite pkt
203                  * and will send it
204                  */
205                 return 1;
206             }
207         } else {
208             /* NET_FILTER_DIRECTION_RX */
209             if (!handle_secondary_tcp_pkt(nf, conn, pkt)) {
210                 qemu_net_queue_send(s->incoming_queue, sender, 0,
211                 (const uint8_t *)pkt->data, pkt->size, NULL);
212                 packet_destroy(pkt, NULL);
213                 pkt = NULL;
214                 /*
215                  * We block the packet here,after rewrite pkt
216                  * and will send it
217                  */
218                 return 1;
219             }
220         }
221     }
222 
223     packet_destroy(pkt, NULL);
224     pkt = NULL;
225     return 0;
226 }
227 
228 static void colo_rewriter_cleanup(NetFilterState *nf)
229 {
230     RewriterState *s = FILTER_COLO_REWRITER(nf);
231 
232     /* flush packets */
233     if (s->incoming_queue) {
234         filter_rewriter_flush(nf);
235         g_free(s->incoming_queue);
236     }
237 }
238 
239 static void colo_rewriter_setup(NetFilterState *nf, Error **errp)
240 {
241     RewriterState *s = FILTER_COLO_REWRITER(nf);
242 
243     s->connection_track_table = g_hash_table_new_full(connection_key_hash,
244                                                       connection_key_equal,
245                                                       g_free,
246                                                       connection_destroy);
247     s->incoming_queue = qemu_new_net_queue(qemu_netfilter_pass_to_next, nf);
248 }
249 
250 static bool filter_rewriter_get_vnet_hdr(Object *obj, Error **errp)
251 {
252     RewriterState *s = FILTER_COLO_REWRITER(obj);
253 
254     return s->vnet_hdr;
255 }
256 
257 static void filter_rewriter_set_vnet_hdr(Object *obj,
258                                          bool value,
259                                          Error **errp)
260 {
261     RewriterState *s = FILTER_COLO_REWRITER(obj);
262 
263     s->vnet_hdr = value;
264 }
265 
266 static void filter_rewriter_init(Object *obj)
267 {
268     RewriterState *s = FILTER_COLO_REWRITER(obj);
269 
270     s->vnet_hdr = false;
271     object_property_add_bool(obj, "vnet_hdr_support",
272                              filter_rewriter_get_vnet_hdr,
273                              filter_rewriter_set_vnet_hdr, NULL);
274 }
275 
276 static void colo_rewriter_class_init(ObjectClass *oc, void *data)
277 {
278     NetFilterClass *nfc = NETFILTER_CLASS(oc);
279 
280     nfc->setup = colo_rewriter_setup;
281     nfc->cleanup = colo_rewriter_cleanup;
282     nfc->receive_iov = colo_rewriter_receive_iov;
283 }
284 
285 static const TypeInfo colo_rewriter_info = {
286     .name = TYPE_FILTER_REWRITER,
287     .parent = TYPE_NETFILTER,
288     .class_init = colo_rewriter_class_init,
289     .instance_init = filter_rewriter_init,
290     .instance_size = sizeof(RewriterState),
291 };
292 
293 static void register_types(void)
294 {
295     type_register_static(&colo_rewriter_info);
296 }
297 
298 type_init(register_types);
299