ip_vs_nfct.c

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