1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * ip_vs_nfct.c: Netfilter connection tracking support for IPVS
4 *
5 * Portions Copyright (C) 2001-2002
6 * Antefacto Ltd, 181 Parnell St, Dublin 1, Ireland.
7 *
8 * Portions Copyright (C) 2003-2010
9 * Julian Anastasov
10 *
11 * Authors:
12 * Ben North <ben@redfrontdoor.org>
13 * Julian Anastasov <ja@ssi.bg> Reorganize and sync with latest kernels
14 * Hannes Eder <heder@google.com> Extend NFCT support for FTP, ipvs match
15 *
16 * Current status:
17 *
18 * - provide conntrack confirmation for new and related connections, by
19 * this way we can see their proper conntrack state in all hooks
20 * - support for all forwarding methods, not only NAT
21 * - FTP support (NAT), ability to support other NAT apps with expectations
22 * - to correctly create expectations for related NAT connections the proper
23 * NF conntrack support must be already installed, eg. ip_vs_ftp requires
24 * nf_conntrack_ftp ... iptables_nat for the same ports (but no iptables
25 * NAT rules are needed)
26 * - alter reply for NAT when forwarding packet in original direction:
27 * conntrack from client in NEW or RELATED (Passive FTP DATA) state or
28 * when RELATED conntrack is created from real server (Active FTP DATA)
29 * - if iptables_nat is not loaded the Passive FTP will not work (the
30 * PASV response can not be NAT-ed) but Active FTP should work
31 */
32
33#define KMSG_COMPONENT "IPVS"
34#define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
35
36#include <linux/module.h>
37#include <linux/types.h>
38#include <linux/kernel.h>
39#include <linux/errno.h>
40#include <linux/compiler.h>
41#include <linux/vmalloc.h>
42#include <linux/skbuff.h>
43#include <net/ip.h>
44#include <linux/netfilter.h>
45#include <linux/netfilter_ipv4.h>
46#include <net/ip_vs.h>
47#include <net/netfilter/nf_conntrack_core.h>
48#include <net/netfilter/nf_conntrack_expect.h>
49#include <net/netfilter/nf_conntrack_seqadj.h>
50#include <net/netfilter/nf_conntrack_helper.h>
51#include <net/netfilter/nf_conntrack_zones.h>
52
53
54#define FMT_TUPLE "%s:%u->%s:%u/%u"
55#define ARG_TUPLE(T) IP_VS_DBG_ADDR((T)->src.l3num, &(T)->src.u3), \
56 ntohs((T)->src.u.all), \
57 IP_VS_DBG_ADDR((T)->src.l3num, &(T)->dst.u3), \
58 ntohs((T)->dst.u.all), \
59 (T)->dst.protonum
60
61#define FMT_CONN "%s:%u->%s:%u->%s:%u/%u:%u"
62#define ARG_CONN(C) IP_VS_DBG_ADDR((C)->af, &((C)->caddr)), \
63 ntohs((C)->cport), \
64 IP_VS_DBG_ADDR((C)->af, &((C)->vaddr)), \
65 ntohs((C)->vport), \
66 IP_VS_DBG_ADDR((C)->daf, &((C)->daddr)), \
67 ntohs((C)->dport), \
68 (C)->protocol, (C)->state
69
70void
71ip_vs_update_conntrack(struct sk_buff *skb, struct ip_vs_conn *cp, int outin)
72{
73 enum ip_conntrack_info ctinfo;
74 struct nf_conn *ct = nf_ct_get(skb, ctinfo: &ctinfo);
75 struct nf_conntrack_tuple new_tuple;
76
77 if (ct == NULL || nf_ct_is_confirmed(ct) ||
78 nf_ct_is_dying(ct))
79 return;
80
81 /* Never alter conntrack for non-NAT conns */
82 if (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ)
83 return;
84
85 /* Never alter conntrack for OPS conns (no reply is expected) */
86 if (cp->flags & IP_VS_CONN_F_ONE_PACKET)
87 return;
88
89 /* Alter reply only in original direction */
90 if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL)
91 return;
92
93 /* Applications may adjust TCP seqs */
94 if (cp->app && nf_ct_protonum(ct) == IPPROTO_TCP &&
95 !nfct_seqadj(ct) && !nfct_seqadj_ext_add(ct))
96 return;
97
98 /*
99 * The connection is not yet in the hashtable, so we update it.
100 * CIP->VIP will remain the same, so leave the tuple in
101 * IP_CT_DIR_ORIGINAL untouched. When the reply comes back from the
102 * real-server we will see RIP->DIP.
103 */
104 new_tuple = ct->tuplehash[IP_CT_DIR_REPLY].tuple;
105 /*
106 * This will also take care of UDP and other protocols.
107 */
108 if (outin) {
109 new_tuple.src.u3 = cp->daddr;
110 if (new_tuple.dst.protonum != IPPROTO_ICMP &&
111 new_tuple.dst.protonum != IPPROTO_ICMPV6)
112 new_tuple.src.u.tcp.port = cp->dport;
113 } else {
114 new_tuple.dst.u3 = cp->vaddr;
115 if (new_tuple.dst.protonum != IPPROTO_ICMP &&
116 new_tuple.dst.protonum != IPPROTO_ICMPV6)
117 new_tuple.dst.u.tcp.port = cp->vport;
118 }
119 IP_VS_DBG_BUF(7, "%s: Updating conntrack ct=%p, status=0x%lX, "
120 "ctinfo=%d, old reply=" FMT_TUPLE "\n",
121 __func__, ct, ct->status, ctinfo,
122 ARG_TUPLE(&ct->tuplehash[IP_CT_DIR_REPLY].tuple));
123 IP_VS_DBG_BUF(7, "%s: Updating conntrack ct=%p, status=0x%lX, "
124 "ctinfo=%d, new reply=" FMT_TUPLE "\n",
125 __func__, ct, ct->status, ctinfo,
126 ARG_TUPLE(&new_tuple));
127 nf_conntrack_alter_reply(ct, newreply: &new_tuple);
128 IP_VS_DBG_BUF(7, "%s: Updated conntrack ct=%p for cp=" FMT_CONN "\n",
129 __func__, ct, ARG_CONN(cp));
130}
131
132int ip_vs_confirm_conntrack(struct sk_buff *skb)
133{
134 return nf_conntrack_confirm(skb);
135}
136
137/*
138 * Called from init_conntrack() as expectfn handler.
139 */
140static void ip_vs_nfct_expect_callback(struct nf_conn *ct,
141 struct nf_conntrack_expect *exp)
142{
143 struct nf_conntrack_tuple *orig, new_reply;
144 struct ip_vs_conn *cp;
145 struct ip_vs_conn_param p;
146 struct net *net = nf_ct_net(ct);
147
148 /*
149 * We assume that no NF locks are held before this callback.
150 * ip_vs_conn_out_get and ip_vs_conn_in_get should match their
151 * expectations even if they use wildcard values, now we provide the
152 * actual values from the newly created original conntrack direction.
153 * The conntrack is confirmed when packet reaches IPVS hooks.
154 */
155
156 /* RS->CLIENT */
157 orig = &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple;
158 ip_vs_conn_fill_param(ipvs: net_ipvs(net), af: exp->tuple.src.l3num, protocol: orig->dst.protonum,
159 caddr: &orig->src.u3, cport: orig->src.u.tcp.port,
160 vaddr: &orig->dst.u3, vport: orig->dst.u.tcp.port, p: &p);
161 cp = ip_vs_conn_out_get(p: &p);
162 if (cp) {
163 /* Change reply CLIENT->RS to CLIENT->VS */
164 IP_VS_DBG_BUF(7, "%s: for ct=%p, status=0x%lX found inout cp="
165 FMT_CONN "\n",
166 __func__, ct, ct->status, ARG_CONN(cp));
167 new_reply = ct->tuplehash[IP_CT_DIR_REPLY].tuple;
168 IP_VS_DBG_BUF(7, "%s: ct=%p before alter: reply tuple="
169 FMT_TUPLE "\n",
170 __func__, ct, ARG_TUPLE(&new_reply));
171 new_reply.dst.u3 = cp->vaddr;
172 new_reply.dst.u.tcp.port = cp->vport;
173 goto alter;
174 }
175
176 /* CLIENT->VS */
177 cp = ip_vs_conn_in_get(p: &p);
178 if (cp) {
179 /* Change reply VS->CLIENT to RS->CLIENT */
180 IP_VS_DBG_BUF(7, "%s: for ct=%p, status=0x%lX found outin cp="
181 FMT_CONN "\n",
182 __func__, ct, ct->status, ARG_CONN(cp));
183 new_reply = ct->tuplehash[IP_CT_DIR_REPLY].tuple;
184 IP_VS_DBG_BUF(7, "%s: ct=%p before alter: reply tuple="
185 FMT_TUPLE "\n",
186 __func__, ct, ARG_TUPLE(&new_reply));
187 new_reply.src.u3 = cp->daddr;
188 new_reply.src.u.tcp.port = cp->dport;
189 goto alter;
190 }
191
192 IP_VS_DBG_BUF(7, "%s: ct=%p, status=0x%lX, tuple=" FMT_TUPLE
193 " - unknown expect\n",
194 __func__, ct, ct->status, ARG_TUPLE(orig));
195 return;
196
197alter:
198 /* Never alter conntrack for non-NAT conns */
199 if (IP_VS_FWD_METHOD(cp) == IP_VS_CONN_F_MASQ)
200 nf_conntrack_alter_reply(ct, newreply: &new_reply);
201 ip_vs_conn_put(cp);
202 return;
203}
204
205/*
206 * Create NF conntrack expectation with wildcard (optional) source port.
207 * Then the default callback function will alter the reply and will confirm
208 * the conntrack entry when the first packet comes.
209 * Use port 0 to expect connection from any port.
210 */
211void ip_vs_nfct_expect_related(struct sk_buff *skb, struct nf_conn *ct,
212 struct ip_vs_conn *cp, u_int8_t proto,
213 const __be16 port, int from_rs)
214{
215 struct nf_conntrack_expect *exp;
216
217 if (ct == NULL)
218 return;
219
220 exp = nf_ct_expect_alloc(me: ct);
221 if (!exp)
222 return;
223
224 nf_ct_expect_init(exp, NF_CT_EXPECT_CLASS_DEFAULT, nf_ct_l3num(ct),
225 from_rs ? &cp->daddr : &cp->caddr,
226 from_rs ? &cp->caddr : &cp->vaddr,
227 proto, port ? &port : NULL,
228 from_rs ? &cp->cport : &cp->vport);
229
230 exp->expectfn = ip_vs_nfct_expect_callback;
231
232 IP_VS_DBG_BUF(7, "%s: ct=%p, expect tuple=" FMT_TUPLE "\n",
233 __func__, ct, ARG_TUPLE(&exp->tuple));
234 nf_ct_expect_related(expect: exp, flags: 0);
235 nf_ct_expect_put(exp);
236}
237EXPORT_SYMBOL(ip_vs_nfct_expect_related);
238
239/*
240 * Our connection was terminated, try to drop the conntrack immediately
241 */
242void ip_vs_conn_drop_conntrack(struct ip_vs_conn *cp)
243{
244 struct nf_conntrack_tuple_hash *h;
245 struct nf_conn *ct;
246 struct nf_conntrack_tuple tuple;
247
248 if (!cp->cport)
249 return;
250
251 tuple = (struct nf_conntrack_tuple) {
252 .dst = { .protonum = cp->protocol, .dir = IP_CT_DIR_ORIGINAL } };
253 tuple.src.u3 = cp->caddr;
254 tuple.src.u.all = cp->cport;
255 tuple.src.l3num = cp->af;
256 tuple.dst.u3 = cp->vaddr;
257 tuple.dst.u.all = cp->vport;
258
259 IP_VS_DBG_BUF(7, "%s: dropping conntrack for conn " FMT_CONN "\n",
260 __func__, ARG_CONN(cp));
261
262 h = nf_conntrack_find_get(net: cp->ipvs->net, zone: &nf_ct_zone_dflt, tuple: &tuple);
263 if (h) {
264 ct = nf_ct_tuplehash_to_ctrack(hash: h);
265 if (nf_ct_kill(ct)) {
266 IP_VS_DBG_BUF(7, "%s: ct=%p deleted for tuple="
267 FMT_TUPLE "\n",
268 __func__, ct, ARG_TUPLE(&tuple));
269 } else {
270 IP_VS_DBG_BUF(7, "%s: ct=%p, no conntrack for tuple="
271 FMT_TUPLE "\n",
272 __func__, ct, ARG_TUPLE(&tuple));
273 }
274 nf_ct_put(ct);
275 } else {
276 IP_VS_DBG_BUF(7, "%s: no conntrack for tuple=" FMT_TUPLE "\n",
277 __func__, ARG_TUPLE(&tuple));
278 }
279}
280
281

source code of linux/net/netfilter/ipvs/ip_vs_nfct.c