xref: /openbmc/qemu/net/filter-rewriter.c (revision a18d4369)
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 "colo.h"
15 #include "net/filter.h"
16 #include "net/net.h"
17 #include "qemu/error-report.h"
18 #include "qom/object.h"
19 #include "qemu/main-loop.h"
20 #include "qemu/iov.h"
21 #include "net/checksum.h"
22 #include "net/colo.h"
23 #include "migration/colo.h"
24 #include "util.h"
25 
26 #define TYPE_FILTER_REWRITER "filter-rewriter"
27 OBJECT_DECLARE_SIMPLE_TYPE(RewriterState, FILTER_REWRITER)
28 
29 #define FAILOVER_MODE_ON  true
30 #define FAILOVER_MODE_OFF false
31 
32 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     bool failover_mode;
39 };
40 
filter_rewriter_failover_mode(RewriterState * s)41 static void filter_rewriter_failover_mode(RewriterState *s)
42 {
43     s->failover_mode = FAILOVER_MODE_ON;
44 }
45 
filter_rewriter_flush(NetFilterState * nf)46 static void filter_rewriter_flush(NetFilterState *nf)
47 {
48     RewriterState *s = FILTER_REWRITER(nf);
49 
50     if (!qemu_net_queue_flush(s->incoming_queue)) {
51         /* Unable to empty the queue, purge remaining packets */
52         qemu_net_queue_purge(s->incoming_queue, nf->netdev);
53     }
54 }
55 
56 /*
57  * Return 1 on success, if return 0 means the pkt
58  * is not TCP packet
59  */
is_tcp_packet(Packet * pkt)60 static int is_tcp_packet(Packet *pkt)
61 {
62     if (!parse_packet_early(pkt) &&
63         pkt->ip->ip_p == IPPROTO_TCP) {
64         return 1;
65     } else {
66         return 0;
67     }
68 }
69 
70 /* handle tcp packet from primary guest */
handle_primary_tcp_pkt(RewriterState * rf,Connection * conn,Packet * pkt,ConnectionKey * key)71 static int handle_primary_tcp_pkt(RewriterState *rf,
72                                   Connection *conn,
73                                   Packet *pkt, ConnectionKey *key)
74 {
75     struct tcp_hdr *tcp_pkt;
76 
77     tcp_pkt = (struct tcp_hdr *)pkt->transport_header;
78     if (trace_event_get_state_backends(TRACE_COLO_FILTER_REWRITER_PKT_INFO)) {
79         trace_colo_filter_rewriter_pkt_info(__func__,
80                     inet_ntoa(pkt->ip->ip_src), inet_ntoa(pkt->ip->ip_dst),
81                     ntohl(tcp_pkt->th_seq), ntohl(tcp_pkt->th_ack),
82                     tcp_pkt->th_flags);
83     }
84     if (trace_event_get_state_backends(
85           TRACE_COLO_FILTER_REWRITER_CONN_OFFSET)) {
86         trace_colo_filter_rewriter_conn_offset(conn->offset);
87     }
88 
89     if (((tcp_pkt->th_flags & (TH_ACK | TH_SYN)) == (TH_ACK | TH_SYN)) &&
90         conn->tcp_state == TCPS_SYN_SENT) {
91         conn->tcp_state = TCPS_ESTABLISHED;
92     }
93 
94     if (((tcp_pkt->th_flags & (TH_ACK | TH_SYN)) == TH_SYN)) {
95         /*
96          * we use this flag update offset func
97          * run once in independent tcp connection
98          */
99         conn->tcp_state = TCPS_SYN_RECEIVED;
100     }
101 
102     if (((tcp_pkt->th_flags & (TH_ACK | TH_SYN)) == TH_ACK)) {
103         if (conn->tcp_state == TCPS_SYN_RECEIVED) {
104             /*
105              * offset = secondary_seq - primary seq
106              * ack packet sent by guest from primary node,
107              * so we use th_ack - 1 get primary_seq
108              */
109             conn->offset -= (ntohl(tcp_pkt->th_ack) - 1);
110             conn->tcp_state = TCPS_ESTABLISHED;
111         }
112         if (conn->offset) {
113             /* handle packets to the secondary from the primary */
114             tcp_pkt->th_ack = htonl(ntohl(tcp_pkt->th_ack) + conn->offset);
115 
116             net_checksum_calculate((uint8_t *)pkt->data + pkt->vnet_hdr_len,
117                                    pkt->size - pkt->vnet_hdr_len, CSUM_TCP);
118         }
119 
120         /*
121          * Passive close step 3
122          */
123         if ((conn->tcp_state == TCPS_LAST_ACK) &&
124             (ntohl(tcp_pkt->th_ack) == (conn->fin_ack_seq + 1))) {
125             conn->tcp_state = TCPS_CLOSED;
126             g_hash_table_remove(rf->connection_track_table, key);
127         }
128     }
129 
130     if ((tcp_pkt->th_flags & TH_FIN) == TH_FIN) {
131         /*
132          * Passive close.
133          * Step 1:
134          * The *server* side of this connect is VM, *client* tries to close
135          * the connection. We will into CLOSE_WAIT status.
136          *
137          * Step 2:
138          * In this step we will into LAST_ACK status.
139          *
140          * We got 'fin=1, ack=1' packet from server side, we need to
141          * record the seq of 'fin=1, ack=1' packet.
142          *
143          * Step 3:
144          * We got 'ack=1' packets from client side, it acks 'fin=1, ack=1'
145          * packet from server side. From this point, we can ensure that there
146          * will be no packets in the connection, except that, some errors
147          * happen between the path of 'filter object' and vNIC, if this rare
148          * case really happen, we can still create a new connection,
149          * So it is safe to remove the connection from connection_track_table.
150          *
151          */
152         if (conn->tcp_state == TCPS_ESTABLISHED) {
153             conn->tcp_state = TCPS_CLOSE_WAIT;
154         }
155 
156         /*
157          * Active close step 2.
158          */
159         if (conn->tcp_state == TCPS_FIN_WAIT_1) {
160             /*
161              * For simplify implementation, we needn't wait 2MSL time
162              * in filter rewriter. Because guest kernel will track the
163              * TCP status and wait 2MSL time, if client resend the FIN
164              * packet, guest will apply the last ACK too.
165              * So, we skip the TCPS_TIME_WAIT state here and go straight
166              * to TCPS_CLOSED state.
167              */
168             conn->tcp_state = TCPS_CLOSED;
169             g_hash_table_remove(rf->connection_track_table, key);
170         }
171     }
172 
173     return 0;
174 }
175 
176 /* handle tcp packet from secondary guest */
handle_secondary_tcp_pkt(RewriterState * rf,Connection * conn,Packet * pkt,ConnectionKey * key)177 static int handle_secondary_tcp_pkt(RewriterState *rf,
178                                     Connection *conn,
179                                     Packet *pkt, ConnectionKey *key)
180 {
181     struct tcp_hdr *tcp_pkt;
182 
183     tcp_pkt = (struct tcp_hdr *)pkt->transport_header;
184 
185     if (trace_event_get_state_backends(TRACE_COLO_FILTER_REWRITER_PKT_INFO)) {
186         trace_colo_filter_rewriter_pkt_info(__func__,
187                     inet_ntoa(pkt->ip->ip_src), inet_ntoa(pkt->ip->ip_dst),
188                     ntohl(tcp_pkt->th_seq), ntohl(tcp_pkt->th_ack),
189                     tcp_pkt->th_flags);
190     }
191     if (trace_event_get_state_backends(
192           TRACE_COLO_FILTER_REWRITER_CONN_OFFSET)) {
193         trace_colo_filter_rewriter_conn_offset(conn->offset);
194     }
195 
196     if (conn->tcp_state == TCPS_SYN_RECEIVED &&
197         ((tcp_pkt->th_flags & (TH_ACK | TH_SYN)) == (TH_ACK | TH_SYN))) {
198         /*
199          * save offset = secondary_seq and then
200          * in handle_primary_tcp_pkt make offset
201          * = secondary_seq - primary_seq
202          */
203         conn->offset = ntohl(tcp_pkt->th_seq);
204     }
205 
206     /* VM active connect */
207     if (conn->tcp_state == TCPS_CLOSED &&
208         ((tcp_pkt->th_flags & (TH_ACK | TH_SYN)) == TH_SYN)) {
209         conn->tcp_state = TCPS_SYN_SENT;
210     }
211 
212     if ((tcp_pkt->th_flags & (TH_ACK | TH_SYN)) == TH_ACK) {
213         /* Only need to adjust seq while offset is Non-zero */
214         if (conn->offset) {
215             /* handle packets to the primary from the secondary*/
216             tcp_pkt->th_seq = htonl(ntohl(tcp_pkt->th_seq) - conn->offset);
217 
218             net_checksum_calculate((uint8_t *)pkt->data + pkt->vnet_hdr_len,
219                                    pkt->size - pkt->vnet_hdr_len, CSUM_TCP);
220         }
221     }
222 
223     /*
224      * Passive close step 2:
225      */
226     if (conn->tcp_state == TCPS_CLOSE_WAIT &&
227         (tcp_pkt->th_flags & (TH_ACK | TH_FIN)) == (TH_ACK | TH_FIN)) {
228         conn->fin_ack_seq = ntohl(tcp_pkt->th_seq);
229         conn->tcp_state = TCPS_LAST_ACK;
230     }
231 
232     /*
233      * Active close
234      *
235      * Step 1:
236      * The *server* side of this connect is VM, *server* tries to close
237      * the connection.
238      *
239      * Step 2:
240      * We will into CLOSE_WAIT status.
241      * We simplify the TCPS_FIN_WAIT_2, TCPS_TIME_WAIT and
242      * CLOSING status.
243      */
244     if (conn->tcp_state == TCPS_ESTABLISHED &&
245         (tcp_pkt->th_flags & (TH_ACK | TH_FIN)) == TH_FIN) {
246         conn->tcp_state = TCPS_FIN_WAIT_1;
247     }
248 
249     return 0;
250 }
251 
colo_rewriter_receive_iov(NetFilterState * nf,NetClientState * sender,unsigned flags,const struct iovec * iov,int iovcnt,NetPacketSent * sent_cb)252 static ssize_t colo_rewriter_receive_iov(NetFilterState *nf,
253                                          NetClientState *sender,
254                                          unsigned flags,
255                                          const struct iovec *iov,
256                                          int iovcnt,
257                                          NetPacketSent *sent_cb)
258 {
259     RewriterState *s = FILTER_REWRITER(nf);
260     Connection *conn;
261     ConnectionKey key;
262     Packet *pkt;
263     ssize_t size = iov_size(iov, iovcnt);
264     ssize_t vnet_hdr_len = 0;
265     char *buf = g_malloc0(size);
266 
267     iov_to_buf(iov, iovcnt, 0, buf, size);
268 
269     if (s->vnet_hdr) {
270         vnet_hdr_len = nf->netdev->vnet_hdr_len;
271     }
272 
273     pkt = packet_new_nocopy(buf, size, vnet_hdr_len);
274 
275     /*
276      * if we get tcp packet
277      * we will rewrite it to make secondary guest's
278      * connection established successfully
279      */
280     if (pkt && is_tcp_packet(pkt)) {
281 
282         fill_connection_key(pkt, &key, sender == nf->netdev);
283 
284         /* After failover we needn't change new TCP packet */
285         if (s->failover_mode &&
286             !connection_has_tracked(s->connection_track_table, &key)) {
287             goto out;
288         }
289 
290         conn = connection_get(s->connection_track_table,
291                               &key,
292                               NULL);
293 
294         if (sender == nf->netdev) {
295             /* NET_FILTER_DIRECTION_TX */
296             if (!handle_primary_tcp_pkt(s, conn, pkt, &key)) {
297                 qemu_net_queue_send(s->incoming_queue, sender, 0,
298                 (const uint8_t *)pkt->data, pkt->size, NULL);
299                 packet_destroy(pkt, NULL);
300                 pkt = NULL;
301                 /*
302                  * We block the packet here,after rewrite pkt
303                  * and will send it
304                  */
305                 return 1;
306             }
307         } else {
308             /* NET_FILTER_DIRECTION_RX */
309             if (!handle_secondary_tcp_pkt(s, conn, pkt, &key)) {
310                 qemu_net_queue_send(s->incoming_queue, sender, 0,
311                 (const uint8_t *)pkt->data, pkt->size, NULL);
312                 packet_destroy(pkt, NULL);
313                 pkt = NULL;
314                 /*
315                  * We block the packet here,after rewrite pkt
316                  * and will send it
317                  */
318                 return 1;
319             }
320         }
321     }
322 
323 out:
324     packet_destroy(pkt, NULL);
325     pkt = NULL;
326     return 0;
327 }
328 
reset_seq_offset(gpointer key,gpointer value,gpointer user_data)329 static void reset_seq_offset(gpointer key, gpointer value, gpointer user_data)
330 {
331     Connection *conn = (Connection *)value;
332 
333     conn->offset = 0;
334 }
335 
offset_is_nonzero(gpointer key,gpointer value,gpointer user_data)336 static gboolean offset_is_nonzero(gpointer key,
337                                   gpointer value,
338                                   gpointer user_data)
339 {
340     Connection *conn = (Connection *)value;
341 
342     return conn->offset ? true : false;
343 }
344 
colo_rewriter_handle_event(NetFilterState * nf,int event,Error ** errp)345 static void colo_rewriter_handle_event(NetFilterState *nf, int event,
346                                        Error **errp)
347 {
348     RewriterState *rs = FILTER_REWRITER(nf);
349 
350     switch (event) {
351     case COLO_EVENT_CHECKPOINT:
352         g_hash_table_foreach(rs->connection_track_table,
353                             reset_seq_offset, NULL);
354         break;
355     case COLO_EVENT_FAILOVER:
356         if (!g_hash_table_find(rs->connection_track_table,
357                               offset_is_nonzero, NULL)) {
358             filter_rewriter_failover_mode(rs);
359         }
360         break;
361     default:
362         break;
363     }
364 }
365 
colo_rewriter_cleanup(NetFilterState * nf)366 static void colo_rewriter_cleanup(NetFilterState *nf)
367 {
368     RewriterState *s = FILTER_REWRITER(nf);
369 
370     /* flush packets */
371     if (s->incoming_queue) {
372         filter_rewriter_flush(nf);
373         g_free(s->incoming_queue);
374     }
375 
376     g_hash_table_destroy(s->connection_track_table);
377 }
378 
colo_rewriter_setup(NetFilterState * nf,Error ** errp)379 static void colo_rewriter_setup(NetFilterState *nf, Error **errp)
380 {
381     RewriterState *s = FILTER_REWRITER(nf);
382 
383     s->connection_track_table = g_hash_table_new_full(connection_key_hash,
384                                                       connection_key_equal,
385                                                       g_free,
386                                                       NULL);
387     s->incoming_queue = qemu_new_net_queue(qemu_netfilter_pass_to_next, nf);
388 }
389 
filter_rewriter_get_vnet_hdr(Object * obj,Error ** errp)390 static bool filter_rewriter_get_vnet_hdr(Object *obj, Error **errp)
391 {
392     RewriterState *s = FILTER_REWRITER(obj);
393 
394     return s->vnet_hdr;
395 }
396 
filter_rewriter_set_vnet_hdr(Object * obj,bool value,Error ** errp)397 static void filter_rewriter_set_vnet_hdr(Object *obj,
398                                          bool value,
399                                          Error **errp)
400 {
401     RewriterState *s = FILTER_REWRITER(obj);
402 
403     s->vnet_hdr = value;
404 }
405 
filter_rewriter_init(Object * obj)406 static void filter_rewriter_init(Object *obj)
407 {
408     RewriterState *s = FILTER_REWRITER(obj);
409 
410     s->vnet_hdr = false;
411     s->failover_mode = FAILOVER_MODE_OFF;
412 }
413 
colo_rewriter_class_init(ObjectClass * oc,void * data)414 static void colo_rewriter_class_init(ObjectClass *oc, void *data)
415 {
416     NetFilterClass *nfc = NETFILTER_CLASS(oc);
417 
418     object_class_property_add_bool(oc, "vnet_hdr_support",
419                                    filter_rewriter_get_vnet_hdr,
420                                    filter_rewriter_set_vnet_hdr);
421 
422     nfc->setup = colo_rewriter_setup;
423     nfc->cleanup = colo_rewriter_cleanup;
424     nfc->receive_iov = colo_rewriter_receive_iov;
425     nfc->handle_event = colo_rewriter_handle_event;
426 }
427 
428 static const TypeInfo colo_rewriter_info = {
429     .name = TYPE_FILTER_REWRITER,
430     .parent = TYPE_NETFILTER,
431     .class_init = colo_rewriter_class_init,
432     .instance_init = filter_rewriter_init,
433     .instance_size = sizeof(RewriterState),
434 };
435 
register_types(void)436 static void register_types(void)
437 {
438     type_register_static(&colo_rewriter_info);
439 }
440 
441 type_init(register_types);
442