ipvs: netfilter connection tracking changes
authorJulian Anastasov <ja@ssi.bg>
Tue, 21 Sep 2010 15:35:41 +0000 (17:35 +0200)
committerPatrick McHardy <kaber@trash.net>
Tue, 21 Sep 2010 15:35:41 +0000 (17:35 +0200)
Add more code to IPVS to work with Netfilter connection
tracking and fix some problems.

- Allow IPVS to be compiled without connection tracking as in
2.6.35 and before. This can avoid keeping conntracks for all
IPVS connections because this costs memory. ip_vs_ftp still
depends on connection tracking and NAT as implemented for 2.6.36.

- Add sysctl var "conntrack" to enable connection tracking for
all IPVS connections. For loaded IPVS directors it needs
tuning of nf_conntrack_max limit.

- Add IP_VS_CONN_F_NFCT connection flag to request the connection
to use connection tracking. This allows user space to provide this
flag, for example, in dest->conn_flags. This can be useful to
request connection tracking per real server instead of forcing it
for all connections with the "conntrack" sysctl. This flag is
set currently only by ip_vs_ftp and of course by "conntrack" sysctl.

- Add ip_vs_nfct.c file to hold all connection tracking code,
by this way main code should not depend of netfilter conntrack
support.

- Return back the ip_vs_post_routing handler as in 2.6.35 and use
skb->ipvs_property=1 to allow IPVS to work without connection
tracking

Connection tracking:

- most of the code is already in 2.6.36-rc

- alter conntrack reply tuple for LVS-NAT connections when first packet
from client is forwarded and conntrack state is NEW or RELATED.
Additionally, alter reply for RELATED connections from real server,
again for packet in original direction.

- add IP_VS_XMIT_TUNNEL to confirm conntrack (without altering
reply) for LVS-TUN early because we want to call nf_reset. It is
needed because we add IPIP header and the original conntrack
should be preserved, not destroyed. The transmitted IPIP packets
can reuse same conntrack, so we do not set skb->ipvs_property.

- try to destroy conntrack when the IPVS connection is destroyed.
It is not fatal if conntrack disappears before that, it depends
on the used timers.

Fix problems from long time:

- add skb->ip_summed = CHECKSUM_NONE for the LVS-TUN transmitters

Signed-off-by: Julian Anastasov <ja@ssi.bg>
Signed-off-by: Patrick McHardy <kaber@trash.net>
include/linux/ip_vs.h
include/net/ip_vs.h
net/netfilter/ipvs/Kconfig
net/netfilter/ipvs/Makefile
net/netfilter/ipvs/ip_vs_conn.c
net/netfilter/ipvs/ip_vs_core.c
net/netfilter/ipvs/ip_vs_ctl.c
net/netfilter/ipvs/ip_vs_ftp.c
net/netfilter/ipvs/ip_vs_nfct.c [new file with mode: 0644]
net/netfilter/ipvs/ip_vs_xmit.c

index 003d75f..df77286 100644 (file)
 #define IP_VS_CONN_F_ONE_PACKET        0x2000          /* forward only one packet */
 
 /* Flags that are not sent to backup server start from bit 16 */
+#define IP_VS_CONN_F_NFCT      (1 << 16)       /* use netfilter conntrack */
 
 /* Connection flags from destination that can be changed by user space */
 #define IP_VS_CONN_F_DEST_MASK (IP_VS_CONN_F_FWD_MASK | \
                                IP_VS_CONN_F_ONE_PACKET | \
+                               IP_VS_CONN_F_NFCT | \
                                0)
 
 #define IP_VS_SCHEDNAME_MAXLEN 16
index 62698a9..e8ec523 100644 (file)
@@ -25,7 +25,9 @@
 #include <linux/ip.h>
 #include <linux/ipv6.h>                        /* for struct ipv6hdr */
 #include <net/ipv6.h>                  /* for ipv6_addr_copy */
-
+#ifdef CONFIG_IP_VS_NFCT
+#include <net/netfilter/nf_conntrack.h>
+#endif
 
 /* Connections' size value needed by ip_vs_ctl.c */
 extern int ip_vs_conn_tab_size;
@@ -798,6 +800,7 @@ extern int sysctl_ip_vs_expire_nodest_conn;
 extern int sysctl_ip_vs_expire_quiescent_template;
 extern int sysctl_ip_vs_sync_threshold[2];
 extern int sysctl_ip_vs_nat_icmp_send;
+extern int sysctl_ip_vs_conntrack;
 extern struct ip_vs_stats ip_vs_stats;
 extern const struct ctl_path net_vs_ctl_path[];
 
@@ -955,8 +958,47 @@ static inline __wsum ip_vs_check_diff2(__be16 old, __be16 new, __wsum oldsum)
        return csum_partial(diff, sizeof(diff), oldsum);
 }
 
+#ifdef CONFIG_IP_VS_NFCT
+/*
+ *      Netfilter connection tracking
+ *      (from ip_vs_nfct.c)
+ */
+static inline int ip_vs_conntrack_enabled(void)
+{
+       return sysctl_ip_vs_conntrack;
+}
+
 extern void ip_vs_update_conntrack(struct sk_buff *skb, struct ip_vs_conn *cp,
                                   int outin);
+extern int ip_vs_confirm_conntrack(struct sk_buff *skb, struct ip_vs_conn *cp);
+extern void ip_vs_nfct_expect_related(struct sk_buff *skb, struct nf_conn *ct,
+                                     struct ip_vs_conn *cp, u_int8_t proto,
+                                     const __be16 port, int from_rs);
+extern void ip_vs_conn_drop_conntrack(struct ip_vs_conn *cp);
+
+#else
+
+static inline int ip_vs_conntrack_enabled(void)
+{
+       return 0;
+}
+
+static inline void ip_vs_update_conntrack(struct sk_buff *skb,
+                                         struct ip_vs_conn *cp, int outin)
+{
+}
+
+static inline int ip_vs_confirm_conntrack(struct sk_buff *skb,
+                                         struct ip_vs_conn *cp)
+{
+       return NF_ACCEPT;
+}
+
+static inline void ip_vs_conn_drop_conntrack(struct ip_vs_conn *cp)
+{
+}
+/* CONFIG_IP_VS_NFCT */
+#endif
 
 #endif /* __KERNEL__ */
 
index 46a77d5..af3c9f4 100644 (file)
@@ -3,7 +3,7 @@
 #
 menuconfig IP_VS
        tristate "IP virtual server support"
-       depends on NET && INET && NETFILTER && NF_CONNTRACK
+       depends on NET && INET && NETFILTER
        ---help---
          IP Virtual Server support will let you build a high-performance
          virtual server based on cluster of two or more real servers. This
@@ -235,7 +235,8 @@ comment 'IPVS application helper'
 
 config IP_VS_FTP
        tristate "FTP protocol helper"
-        depends on IP_VS_PROTO_TCP && NF_NAT
+        depends on IP_VS_PROTO_TCP && NF_CONNTRACK && NF_NAT
+       select IP_VS_NFCT
        ---help---
          FTP is a protocol that transfers IP address and/or port number in
          the payload. In the virtual server via Network Address Translation,
@@ -247,4 +248,12 @@ config     IP_VS_FTP
          If you want to compile it in kernel, say Y. To compile it as a
          module, choose M here. If unsure, say N.
 
+config IP_VS_NFCT
+       bool "Netfilter connection tracking"
+       depends on NF_CONNTRACK
+       ---help---
+         The Netfilter connection tracking support allows the IPVS
+         connection state to be exported to the Netfilter framework
+         for filtering purposes.
+
 endif # IP_VS
index e3baefd..349fe88 100644 (file)
@@ -9,10 +9,13 @@ ip_vs_proto-objs-$(CONFIG_IP_VS_PROTO_UDP) += ip_vs_proto_udp.o
 ip_vs_proto-objs-$(CONFIG_IP_VS_PROTO_AH_ESP) += ip_vs_proto_ah_esp.o
 ip_vs_proto-objs-$(CONFIG_IP_VS_PROTO_SCTP) += ip_vs_proto_sctp.o
 
+ip_vs-extra_objs-y :=
+ip_vs-extra_objs-$(CONFIG_IP_VS_NFCT) += ip_vs_nfct.o
+
 ip_vs-objs :=  ip_vs_conn.o ip_vs_core.o ip_vs_ctl.o ip_vs_sched.o        \
                ip_vs_xmit.o ip_vs_app.o ip_vs_sync.o                      \
                ip_vs_est.o ip_vs_proto.o                                  \
-               $(ip_vs_proto-objs-y)
+               $(ip_vs_proto-objs-y) $(ip_vs-extra_objs-y)
 
 
 # IPVS core
index 9fe1da7..a970d96 100644 (file)
@@ -721,6 +721,9 @@ static void ip_vs_conn_expire(unsigned long data)
                if (cp->control)
                        ip_vs_control_del(cp);
 
+               if (cp->flags & IP_VS_CONN_F_NFCT)
+                       ip_vs_conn_drop_conntrack(cp);
+
                if (unlikely(cp->app != NULL))
                        ip_vs_unbind_app(cp);
                ip_vs_unbind_dest(cp);
@@ -816,6 +819,16 @@ ip_vs_conn_new(int af, int proto, const union nf_inet_addr *caddr, __be16 cport,
        if (unlikely(pp && atomic_read(&pp->appcnt)))
                ip_vs_bind_app(cp, pp);
 
+       /*
+        * Allow conntrack to be preserved. By default, conntrack
+        * is created and destroyed for every packet.
+        * Sometimes keeping conntrack can be useful for
+        * IP_VS_CONN_F_ONE_PACKET too.
+        */
+
+       if (ip_vs_conntrack_enabled())
+               cp->flags |= IP_VS_CONN_F_NFCT;
+
        /* Hash it in the ip_vs_conn_tab finally */
        ip_vs_conn_hash(cp);
 
index 319991d..7fbc80d 100644 (file)
@@ -537,6 +537,23 @@ int ip_vs_leave(struct ip_vs_service *svc, struct sk_buff *skb,
        return NF_DROP;
 }
 
+/*
+ * It is hooked before NF_IP_PRI_NAT_SRC at the NF_INET_POST_ROUTING
+ * chain and is used to avoid double NAT and confirmation when we do
+ * not want to keep the conntrack structure
+ */
+static unsigned int ip_vs_post_routing(unsigned int hooknum,
+                                      struct sk_buff *skb,
+                                      const struct net_device *in,
+                                      const struct net_device *out,
+                                      int (*okfn)(struct sk_buff *))
+{
+       if (!skb->ipvs_property)
+               return NF_ACCEPT;
+       /* The packet was sent from IPVS, exit this chain */
+       return NF_STOP;
+}
+
 __sum16 ip_vs_checksum_complete(struct sk_buff *skb, int offset)
 {
        return csum_fold(skb_checksum(skb, offset, skb->len - offset, 0));
@@ -695,7 +712,10 @@ static int handle_response_icmp(int af, struct sk_buff *skb,
        /* do the statistics and put it back */
        ip_vs_out_stats(cp, skb);
 
-       skb->ipvs_property = 1;
+       if (!(cp->flags & IP_VS_CONN_F_NFCT))
+               skb->ipvs_property = 1;
+       else
+               ip_vs_update_conntrack(skb, cp, 0);
        verdict = NF_ACCEPT;
 
 out:
@@ -928,17 +948,19 @@ handle_response(int af, struct sk_buff *skb, struct ip_vs_protocol *pp,
 
        ip_vs_out_stats(cp, skb);
        ip_vs_set_state(cp, IP_VS_DIR_OUTPUT, skb, pp);
-       ip_vs_update_conntrack(skb, cp, 0);
+       if (!(cp->flags & IP_VS_CONN_F_NFCT))
+               skb->ipvs_property = 1;
+       else
+               ip_vs_update_conntrack(skb, cp, 0);
        ip_vs_conn_put(cp);
 
-       skb->ipvs_property = 1;
-
        LeaveFunction(11);
        return NF_ACCEPT;
 
 drop:
        ip_vs_conn_put(cp);
        kfree_skb(skb);
+       LeaveFunction(11);
        return NF_STOLEN;
 }
 
@@ -1483,6 +1505,14 @@ static struct nf_hook_ops ip_vs_ops[] __read_mostly = {
                .hooknum        = NF_INET_FORWARD,
                .priority       = 99,
        },
+       /* Before the netfilter connection tracking, exit from POST_ROUTING */
+       {
+               .hook           = ip_vs_post_routing,
+               .owner          = THIS_MODULE,
+               .pf             = PF_INET,
+               .hooknum        = NF_INET_POST_ROUTING,
+               .priority       = NF_IP_PRI_NAT_SRC-1,
+       },
 #ifdef CONFIG_IP_VS_IPV6
        /* After packet filtering, forward packet through VS/DR, VS/TUN,
         * or VS/NAT(change destination), so that filtering rules can be
@@ -1511,6 +1541,14 @@ static struct nf_hook_ops ip_vs_ops[] __read_mostly = {
                .hooknum        = NF_INET_FORWARD,
                .priority       = 99,
        },
+       /* Before the netfilter connection tracking, exit from POST_ROUTING */
+       {
+               .hook           = ip_vs_post_routing,
+               .owner          = THIS_MODULE,
+               .pf             = PF_INET6,
+               .hooknum        = NF_INET_POST_ROUTING,
+               .priority       = NF_IP6_PRI_NAT_SRC-1,
+       },
 #endif
 };
 
index 7bd41d2..d2d842f 100644 (file)
@@ -88,6 +88,9 @@ int sysctl_ip_vs_expire_nodest_conn = 0;
 int sysctl_ip_vs_expire_quiescent_template = 0;
 int sysctl_ip_vs_sync_threshold[2] = { 3, 50 };
 int sysctl_ip_vs_nat_icmp_send = 0;
+#ifdef CONFIG_IP_VS_NFCT
+int sysctl_ip_vs_conntrack;
+#endif
 
 
 #ifdef CONFIG_IP_VS_DEBUG
@@ -1580,6 +1583,15 @@ static struct ctl_table vs_vars[] = {
                .mode           = 0644,
                .proc_handler   = proc_do_defense_mode,
        },
+#ifdef CONFIG_IP_VS_NFCT
+       {
+               .procname       = "conntrack",
+               .data           = &sysctl_ip_vs_conntrack,
+               .maxlen         = sizeof(int),
+               .mode           = 0644,
+               .proc_handler   = &proc_dointvec,
+       },
+#endif
        {
                .procname       = "secure_tcp",
                .data           = &sysctl_ip_vs_secure_tcp,
index 7e9af5b..9cd375f 100644 (file)
  *
  * Author:     Wouter Gadeyne
  *
- *
- * Code for ip_vs_expect_related and ip_vs_expect_callback is taken from
- * http://www.ssi.bg/~ja/nfct/:
- *
- * ip_vs_nfct.c:       Netfilter connection tracking support for IPVS
- *
- * Portions Copyright (C) 2001-2002
- * Antefacto Ltd, 181 Parnell St, Dublin 1, Ireland.
- *
- * Portions Copyright (C) 2003-2008
- * Julian Anastasov
  */
 
 #define KMSG_COMPONENT "IPVS"
 #define SERVER_STRING "227 Entering Passive Mode ("
 #define CLIENT_STRING "PORT "
 
-#define FMT_TUPLE      "%pI4:%u->%pI4:%u/%u"
-#define ARG_TUPLE(T)   &(T)->src.u3.ip, ntohs((T)->src.u.all), \
-                       &(T)->dst.u3.ip, ntohs((T)->dst.u.all), \
-                       (T)->dst.protonum
-
-#define FMT_CONN       "%pI4:%u->%pI4:%u->%pI4:%u/%u:%u"
-#define ARG_CONN(C)    &((C)->caddr.ip), ntohs((C)->cport), \
-                       &((C)->vaddr.ip), ntohs((C)->vport), \
-                       &((C)->daddr.ip), ntohs((C)->dport), \
-                       (C)->protocol, (C)->state
 
 /*
  * List of ports (up to IP_VS_APP_MAX_PORTS) to be handled by helper
@@ -85,6 +64,8 @@ static int ip_vs_ftp_pasv;
 static int
 ip_vs_ftp_init_conn(struct ip_vs_app *app, struct ip_vs_conn *cp)
 {
+       /* We use connection tracking for the command connection */
+       cp->flags |= IP_VS_CONN_F_NFCT;
        return 0;
 }
 
@@ -149,120 +130,6 @@ static int ip_vs_ftp_get_addrport(char *data, char *data_limit,
 }
 
 /*
- * Called from init_conntrack() as expectfn handler.
- */
-static void
-ip_vs_expect_callback(struct nf_conn *ct,
-                     struct nf_conntrack_expect *exp)
-{
-       struct nf_conntrack_tuple *orig, new_reply;
-       struct ip_vs_conn *cp;
-
-       if (exp->tuple.src.l3num != PF_INET)
-               return;
-
-       /*
-        * We assume that no NF locks are held before this callback.
-        * ip_vs_conn_out_get and ip_vs_conn_in_get should match their
-        * expectations even if they use wildcard values, now we provide the
-        * actual values from the newly created original conntrack direction.
-        * The conntrack is confirmed when packet reaches IPVS hooks.
-        */
-
-       /* RS->CLIENT */
-       orig = &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple;
-       cp = ip_vs_conn_out_get(exp->tuple.src.l3num, orig->dst.protonum,
-                               &orig->src.u3, orig->src.u.tcp.port,
-                               &orig->dst.u3, orig->dst.u.tcp.port);
-       if (cp) {
-               /* Change reply CLIENT->RS to CLIENT->VS */
-               new_reply = ct->tuplehash[IP_CT_DIR_REPLY].tuple;
-               IP_VS_DBG(7, "%s(): ct=%p, status=0x%lX, tuples=" FMT_TUPLE ", "
-                         FMT_TUPLE ", found inout cp=" FMT_CONN "\n",
-                         __func__, ct, ct->status,
-                         ARG_TUPLE(orig), ARG_TUPLE(&new_reply),
-                         ARG_CONN(cp));
-               new_reply.dst.u3 = cp->vaddr;
-               new_reply.dst.u.tcp.port = cp->vport;
-               IP_VS_DBG(7, "%s(): ct=%p, new tuples=" FMT_TUPLE ", " FMT_TUPLE
-                         ", inout cp=" FMT_CONN "\n",
-                         __func__, ct,
-                         ARG_TUPLE(orig), ARG_TUPLE(&new_reply),
-                         ARG_CONN(cp));
-               goto alter;
-       }
-
-       /* CLIENT->VS */
-       cp = ip_vs_conn_in_get(exp->tuple.src.l3num, orig->dst.protonum,
-                              &orig->src.u3, orig->src.u.tcp.port,
-                              &orig->dst.u3, orig->dst.u.tcp.port);
-       if (cp) {
-               /* Change reply VS->CLIENT to RS->CLIENT */
-               new_reply = ct->tuplehash[IP_CT_DIR_REPLY].tuple;
-               IP_VS_DBG(7, "%s(): ct=%p, status=0x%lX, tuples=" FMT_TUPLE ", "
-                         FMT_TUPLE ", found outin cp=" FMT_CONN "\n",
-                         __func__, ct, ct->status,
-                         ARG_TUPLE(orig), ARG_TUPLE(&new_reply),
-                         ARG_CONN(cp));
-               new_reply.src.u3 = cp->daddr;
-               new_reply.src.u.tcp.port = cp->dport;
-               IP_VS_DBG(7, "%s(): ct=%p, new tuples=" FMT_TUPLE ", "
-                         FMT_TUPLE ", outin cp=" FMT_CONN "\n",
-                         __func__, ct,
-                         ARG_TUPLE(orig), ARG_TUPLE(&new_reply),
-                         ARG_CONN(cp));
-               goto alter;
-       }
-
-       IP_VS_DBG(7, "%s(): ct=%p, status=0x%lX, tuple=" FMT_TUPLE
-                 " - unknown expect\n",
-                 __func__, ct, ct->status, ARG_TUPLE(orig));
-       return;
-
-alter:
-       /* Never alter conntrack for non-NAT conns */
-       if (IP_VS_FWD_METHOD(cp) == IP_VS_CONN_F_MASQ)
-               nf_conntrack_alter_reply(ct, &new_reply);
-       ip_vs_conn_put(cp);
-       return;
-}
-
-/*
- * Create NF conntrack expectation with wildcard (optional) source port.
- * Then the default callback function will alter the reply and will confirm
- * the conntrack entry when the first packet comes.
- */
-static void
-ip_vs_expect_related(struct sk_buff *skb, struct nf_conn *ct,
-                    struct ip_vs_conn *cp, u_int8_t proto,
-                    const __be16 *port, int from_rs)
-{
-       struct nf_conntrack_expect *exp;
-
-       BUG_ON(!ct || ct == &nf_conntrack_untracked);
-
-       exp = nf_ct_expect_alloc(ct);
-       if (!exp)
-               return;
-
-       if (from_rs)
-               nf_ct_expect_init(exp, NF_CT_EXPECT_CLASS_DEFAULT,
-                                 nf_ct_l3num(ct), &cp->daddr, &cp->caddr,
-                                 proto, port, &cp->cport);
-       else
-               nf_ct_expect_init(exp, NF_CT_EXPECT_CLASS_DEFAULT,
-                                 nf_ct_l3num(ct), &cp->caddr, &cp->vaddr,
-                                 proto, port, &cp->vport);
-
-       exp->expectfn = ip_vs_expect_callback;
-
-       IP_VS_DBG(7, "%s(): ct=%p, expect tuple=" FMT_TUPLE "\n",
-                 __func__, ct, ARG_TUPLE(&exp->tuple));
-       nf_ct_expect_related(exp);
-       nf_ct_expect_put(exp);
-}
-
-/*
  * Look at outgoing ftp packets to catch the response to a PASV command
  * from the server (inside-to-outside).
  * When we see one, we build a connection entry with the client address,
@@ -335,7 +202,8 @@ static int ip_vs_ftp_out(struct ip_vs_app *app, struct ip_vs_conn *cp,
                                              &cp->caddr, 0,
                                              &cp->vaddr, port,
                                              &from, port,
-                                             IP_VS_CONN_F_NO_CPORT,
+                                             IP_VS_CONN_F_NO_CPORT |
+                                             IP_VS_CONN_F_NFCT,
                                              cp->dest);
                        if (!n_cp)
                                return 0;
@@ -371,8 +239,8 @@ static int ip_vs_ftp_out(struct ip_vs_app *app, struct ip_vs_conn *cp,
                                                       start-data, end-start,
                                                       buf, buf_len);
                        if (ret)
-                               ip_vs_expect_related(skb, ct, n_cp,
-                                                    IPPROTO_TCP, NULL, 0);
+                               ip_vs_nfct_expect_related(skb, ct, n_cp,
+                                                         IPPROTO_TCP, 0, 0);
                }
 
                /*
@@ -487,7 +355,7 @@ static int ip_vs_ftp_in(struct ip_vs_app *app, struct ip_vs_conn *cp,
                                      &to, port,
                                      &cp->vaddr, htons(ntohs(cp->vport)-1),
                                      &cp->daddr, htons(ntohs(cp->dport)-1),
-                                     0,
+                                     IP_VS_CONN_F_NFCT,
                                      cp->dest);
                if (!n_cp)
                        return 0;
diff --git a/net/netfilter/ipvs/ip_vs_nfct.c b/net/netfilter/ipvs/ip_vs_nfct.c
new file mode 100644 (file)
index 0000000..c038458
--- /dev/null
@@ -0,0 +1,292 @@
+/*
+ * ip_vs_nfct.c:       Netfilter connection tracking support for IPVS
+ *
+ * Portions Copyright (C) 2001-2002
+ * Antefacto Ltd, 181 Parnell St, Dublin 1, Ireland.
+ *
+ * Portions Copyright (C) 2003-2010
+ * Julian Anastasov
+ *
+ *
+ * This code is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
+ *
+ *
+ * Authors:
+ * Ben North <ben@redfrontdoor.org>
+ * Julian Anastasov <ja@ssi.bg>                Reorganize and sync with latest kernels
+ * Hannes Eder <heder@google.com>      Extend NFCT support for FTP, ipvs match
+ *
+ *
+ * Current status:
+ *
+ * - provide conntrack confirmation for new and related connections, by
+ * this way we can see their proper conntrack state in all hooks
+ * - support for all forwarding methods, not only NAT
+ * - FTP support (NAT), ability to support other NAT apps with expectations
+ * - to correctly create expectations for related NAT connections the proper
+ * NF conntrack support must be already installed, eg. ip_vs_ftp requires
+ * nf_conntrack_ftp ... iptables_nat for the same ports (but no iptables
+ * NAT rules are needed)
+ * - alter reply for NAT when forwarding packet in original direction:
+ * conntrack from client in NEW or RELATED (Passive FTP DATA) state or
+ * when RELATED conntrack is created from real server (Active FTP DATA)
+ * - if iptables_nat is not loaded the Passive FTP will not work (the
+ * PASV response can not be NAT-ed) but Active FTP should work
+ *
+ */
+
+#define KMSG_COMPONENT "IPVS"
+#define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
+
+#include <linux/module.h>
+#include <linux/types.h>
+#include <linux/kernel.h>
+#include <linux/errno.h>
+#include <linux/compiler.h>
+#include <linux/vmalloc.h>
+#include <linux/skbuff.h>
+#include <net/ip.h>
+#include <linux/netfilter.h>
+#include <linux/netfilter_ipv4.h>
+#include <net/ip_vs.h>
+#include <net/netfilter/nf_conntrack_core.h>
+#include <net/netfilter/nf_conntrack_expect.h>
+#include <net/netfilter/nf_conntrack_helper.h>
+#include <net/netfilter/nf_conntrack_zones.h>
+
+
+#define FMT_TUPLE      "%pI4:%u->%pI4:%u/%u"
+#define ARG_TUPLE(T)   &(T)->src.u3.ip, ntohs((T)->src.u.all), \
+                       &(T)->dst.u3.ip, ntohs((T)->dst.u.all), \
+                       (T)->dst.protonum
+
+#define FMT_CONN       "%pI4:%u->%pI4:%u->%pI4:%u/%u:%u"
+#define ARG_CONN(C)    &((C)->caddr.ip), ntohs((C)->cport), \
+                       &((C)->vaddr.ip), ntohs((C)->vport), \
+                       &((C)->daddr.ip), ntohs((C)->dport), \
+                       (C)->protocol, (C)->state
+
+void
+ip_vs_update_conntrack(struct sk_buff *skb, struct ip_vs_conn *cp, int outin)
+{
+       enum ip_conntrack_info ctinfo;
+       struct nf_conn *ct = ct = nf_ct_get(skb, &ctinfo);
+       struct nf_conntrack_tuple new_tuple;
+
+       if (ct == NULL || nf_ct_is_confirmed(ct) || nf_ct_is_untracked(ct) ||
+           nf_ct_is_dying(ct))
+               return;
+
+       /* Never alter conntrack for non-NAT conns */
+       if (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ)
+               return;
+
+       /* Alter reply only in original direction */
+       if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL)
+               return;
+
+       /*
+        * The connection is not yet in the hashtable, so we update it.
+        * CIP->VIP will remain the same, so leave the tuple in
+        * IP_CT_DIR_ORIGINAL untouched.  When the reply comes back from the
+        * real-server we will see RIP->DIP.
+        */
+       new_tuple = ct->tuplehash[IP_CT_DIR_REPLY].tuple;
+       /*
+        * This will also take care of UDP and other protocols.
+        */
+       if (outin) {
+               new_tuple.src.u3 = cp->daddr;
+               if (new_tuple.dst.protonum != IPPROTO_ICMP &&
+                   new_tuple.dst.protonum != IPPROTO_ICMPV6)
+                       new_tuple.src.u.tcp.port = cp->dport;
+       } else {
+               new_tuple.dst.u3 = cp->vaddr;
+               if (new_tuple.dst.protonum != IPPROTO_ICMP &&
+                   new_tuple.dst.protonum != IPPROTO_ICMPV6)
+                       new_tuple.dst.u.tcp.port = cp->vport;
+       }
+       IP_VS_DBG(7, "%s: Updating conntrack ct=%p, status=0x%lX, "
+                 "ctinfo=%d, old reply=" FMT_TUPLE
+                 ", new reply=" FMT_TUPLE ", cp=" FMT_CONN "\n",
+                 __func__, ct, ct->status, ctinfo,
+                 ARG_TUPLE(&ct->tuplehash[IP_CT_DIR_REPLY].tuple),
+                 ARG_TUPLE(&new_tuple), ARG_CONN(cp));
+       nf_conntrack_alter_reply(ct, &new_tuple);
+}
+
+int ip_vs_confirm_conntrack(struct sk_buff *skb, struct ip_vs_conn *cp)
+{
+       return nf_conntrack_confirm(skb);
+}
+
+/*
+ * Called from init_conntrack() as expectfn handler.
+ */
+static void ip_vs_nfct_expect_callback(struct nf_conn *ct,
+       struct nf_conntrack_expect *exp)
+{
+       struct nf_conntrack_tuple *orig, new_reply;
+       struct ip_vs_conn *cp;
+
+       if (exp->tuple.src.l3num != PF_INET)
+               return;
+
+       /*
+        * We assume that no NF locks are held before this callback.
+        * ip_vs_conn_out_get and ip_vs_conn_in_get should match their
+        * expectations even if they use wildcard values, now we provide the
+        * actual values from the newly created original conntrack direction.
+        * The conntrack is confirmed when packet reaches IPVS hooks.
+        */
+
+       /* RS->CLIENT */
+       orig = &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple;
+       cp = ip_vs_conn_out_get(exp->tuple.src.l3num, orig->dst.protonum,
+                               &orig->src.u3, orig->src.u.tcp.port,
+                               &orig->dst.u3, orig->dst.u.tcp.port);
+       if (cp) {
+               /* Change reply CLIENT->RS to CLIENT->VS */
+               new_reply = ct->tuplehash[IP_CT_DIR_REPLY].tuple;
+               IP_VS_DBG(7, "%s: ct=%p, status=0x%lX, tuples=" FMT_TUPLE ", "
+                         FMT_TUPLE ", found inout cp=" FMT_CONN "\n",
+                         __func__, ct, ct->status,
+                         ARG_TUPLE(orig), ARG_TUPLE(&new_reply),
+                         ARG_CONN(cp));
+               new_reply.dst.u3 = cp->vaddr;
+               new_reply.dst.u.tcp.port = cp->vport;
+               IP_VS_DBG(7, "%s: ct=%p, new tuples=" FMT_TUPLE ", " FMT_TUPLE
+                         ", inout cp=" FMT_CONN "\n",
+                         __func__, ct,
+                         ARG_TUPLE(orig), ARG_TUPLE(&new_reply),
+                         ARG_CONN(cp));
+               goto alter;
+       }
+
+       /* CLIENT->VS */
+       cp = ip_vs_conn_in_get(exp->tuple.src.l3num, orig->dst.protonum,
+                              &orig->src.u3, orig->src.u.tcp.port,
+                              &orig->dst.u3, orig->dst.u.tcp.port);
+       if (cp) {
+               /* Change reply VS->CLIENT to RS->CLIENT */
+               new_reply = ct->tuplehash[IP_CT_DIR_REPLY].tuple;
+               IP_VS_DBG(7, "%s: ct=%p, status=0x%lX, tuples=" FMT_TUPLE ", "
+                         FMT_TUPLE ", found outin cp=" FMT_CONN "\n",
+                         __func__, ct, ct->status,
+                         ARG_TUPLE(orig), ARG_TUPLE(&new_reply),
+                         ARG_CONN(cp));
+               new_reply.src.u3 = cp->daddr;
+               new_reply.src.u.tcp.port = cp->dport;
+               IP_VS_DBG(7, "%s: ct=%p, new tuples=" FMT_TUPLE ", "
+                         FMT_TUPLE ", outin cp=" FMT_CONN "\n",
+                         __func__, ct,
+                         ARG_TUPLE(orig), ARG_TUPLE(&new_reply),
+                         ARG_CONN(cp));
+               goto alter;
+       }
+
+       IP_VS_DBG(7, "%s: ct=%p, status=0x%lX, tuple=" FMT_TUPLE
+                 " - unknown expect\n",
+                 __func__, ct, ct->status, ARG_TUPLE(orig));
+       return;
+
+alter:
+       /* Never alter conntrack for non-NAT conns */
+       if (IP_VS_FWD_METHOD(cp) == IP_VS_CONN_F_MASQ)
+               nf_conntrack_alter_reply(ct, &new_reply);
+       ip_vs_conn_put(cp);
+       return;
+}
+
+/*
+ * Create NF conntrack expectation with wildcard (optional) source port.
+ * Then the default callback function will alter the reply and will confirm
+ * the conntrack entry when the first packet comes.
+ * Use port 0 to expect connection from any port.
+ */
+void ip_vs_nfct_expect_related(struct sk_buff *skb, struct nf_conn *ct,
+                              struct ip_vs_conn *cp, u_int8_t proto,
+                              const __be16 port, int from_rs)
+{
+       struct nf_conntrack_expect *exp;
+
+       if (ct == NULL || nf_ct_is_untracked(ct))
+               return;
+
+       exp = nf_ct_expect_alloc(ct);
+       if (!exp)
+               return;
+
+       nf_ct_expect_init(exp, NF_CT_EXPECT_CLASS_DEFAULT, nf_ct_l3num(ct),
+                       from_rs ? &cp->daddr : &cp->caddr,
+                       from_rs ? &cp->caddr : &cp->vaddr,
+                       proto, port ? &port : NULL,
+                       from_rs ? &cp->cport : &cp->vport);
+
+       exp->expectfn = ip_vs_nfct_expect_callback;
+
+       IP_VS_DBG(7, "%s: ct=%p, expect tuple=" FMT_TUPLE "\n",
+               __func__, ct, ARG_TUPLE(&exp->tuple));
+       nf_ct_expect_related(exp);
+       nf_ct_expect_put(exp);
+}
+EXPORT_SYMBOL(ip_vs_nfct_expect_related);
+
+/*
+ * Our connection was terminated, try to drop the conntrack immediately
+ */
+void ip_vs_conn_drop_conntrack(struct ip_vs_conn *cp)
+{
+       struct nf_conntrack_tuple_hash *h;
+       struct nf_conn *ct;
+       struct nf_conntrack_tuple tuple;
+
+       if (!cp->cport)
+               return;
+
+       tuple = (struct nf_conntrack_tuple) {
+               .dst = { .protonum = cp->protocol, .dir = IP_CT_DIR_ORIGINAL } };
+       tuple.src.u3 = cp->caddr;
+       tuple.src.u.all = cp->cport;
+       tuple.src.l3num = cp->af;
+       tuple.dst.u3 = cp->vaddr;
+       tuple.dst.u.all = cp->vport;
+
+       IP_VS_DBG(7, "%s: dropping conntrack with tuple=" FMT_TUPLE
+               " for conn " FMT_CONN "\n",
+               __func__, ARG_TUPLE(&tuple), ARG_CONN(cp));
+
+       h = nf_conntrack_find_get(&init_net, NF_CT_DEFAULT_ZONE, &tuple);
+       if (h) {
+               ct = nf_ct_tuplehash_to_ctrack(h);
+               /* Show what happens instead of calling nf_ct_kill() */
+               if (del_timer(&ct->timeout)) {
+                       IP_VS_DBG(7, "%s: ct=%p, deleted conntrack timer for tuple="
+                               FMT_TUPLE "\n",
+                               __func__, ct, ARG_TUPLE(&tuple));
+                       if (ct->timeout.function)
+                               ct->timeout.function(ct->timeout.data);
+               } else {
+                       IP_VS_DBG(7, "%s: ct=%p, no conntrack timer for tuple="
+                               FMT_TUPLE "\n",
+                               __func__, ct, ARG_TUPLE(&tuple));
+               }
+               nf_ct_put(ct);
+       } else {
+               IP_VS_DBG(7, "%s: no conntrack for tuple=" FMT_TUPLE "\n",
+                       __func__, ARG_TUPLE(&tuple));
+       }
+}
+
index 49df6be..8817afa 100644 (file)
@@ -28,7 +28,6 @@
 #include <net/ip6_route.h>
 #include <linux/icmpv6.h>
 #include <linux/netfilter.h>
-#include <net/netfilter/nf_conntrack.h>
 #include <linux/netfilter_ipv4.h>
 
 #include <net/ip_vs.h>
@@ -194,12 +193,37 @@ ip_vs_dst_reset(struct ip_vs_dest *dest)
        dst_release(old_dst);
 }
 
-#define IP_VS_XMIT(pf, skb, rt)                                \
+#define IP_VS_XMIT_TUNNEL(skb, cp)                             \
+({                                                             \
+       int __ret = NF_ACCEPT;                                  \
+                                                               \
+       if (unlikely((cp)->flags & IP_VS_CONN_F_NFCT))          \
+               __ret = ip_vs_confirm_conntrack(skb, cp);       \
+       if (__ret == NF_ACCEPT) {                               \
+               nf_reset(skb);                                  \
+               (skb)->ip_summed = CHECKSUM_NONE;               \
+       }                                                       \
+       __ret;                                                  \
+})
+
+#define IP_VS_XMIT_NAT(pf, skb, cp)                            \
 do {                                                   \
-       (skb)->ipvs_property = 1;                       \
+       if (likely(!((cp)->flags & IP_VS_CONN_F_NFCT))) \
+               (skb)->ipvs_property = 1;               \
+       else                                            \
+               ip_vs_update_conntrack(skb, cp, 1);     \
        skb_forward_csum(skb);                          \
        NF_HOOK(pf, NF_INET_LOCAL_OUT, (skb), NULL,     \
-               (rt)->dst.dev, dst_output);             \
+               skb_dst(skb)->dev, dst_output);         \
+} while (0)
+
+#define IP_VS_XMIT(pf, skb, cp)                                \
+do {                                                   \
+       if (likely(!((cp)->flags & IP_VS_CONN_F_NFCT))) \
+               (skb)->ipvs_property = 1;               \
+       skb_forward_csum(skb);                          \
+       NF_HOOK(pf, NF_INET_LOCAL_OUT, (skb), NULL,     \
+               skb_dst(skb)->dev, dst_output);         \
 } while (0)
 
 
@@ -271,7 +295,7 @@ ip_vs_bypass_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
        /* Another hack: avoid icmp_send in ip_fragment */
        skb->local_df = 1;
 
-       IP_VS_XMIT(NFPROTO_IPV4, skb, rt);
+       IP_VS_XMIT(NFPROTO_IPV4, skb, cp);
 
        LeaveFunction(10);
        return NF_STOLEN;
@@ -335,7 +359,7 @@ ip_vs_bypass_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp,
        /* Another hack: avoid icmp_send in ip_fragment */
        skb->local_df = 1;
 
-       IP_VS_XMIT(NFPROTO_IPV6, skb, rt);
+       IP_VS_XMIT(NFPROTO_IPV6, skb, cp);
 
        LeaveFunction(10);
        return NF_STOLEN;
@@ -349,36 +373,6 @@ ip_vs_bypass_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp,
 }
 #endif
 
-void
-ip_vs_update_conntrack(struct sk_buff *skb, struct ip_vs_conn *cp, int outin)
-{
-       struct nf_conn *ct = (struct nf_conn *)skb->nfct;
-       struct nf_conntrack_tuple new_tuple;
-
-       if (ct == NULL || nf_ct_is_untracked(ct) || nf_ct_is_confirmed(ct))
-               return;
-
-       /*
-        * The connection is not yet in the hashtable, so we update it.
-        * CIP->VIP will remain the same, so leave the tuple in
-        * IP_CT_DIR_ORIGINAL untouched.  When the reply comes back from the
-        * real-server we will see RIP->DIP.
-        */
-       new_tuple = ct->tuplehash[IP_CT_DIR_REPLY].tuple;
-       if (outin)
-               new_tuple.src.u3 = cp->daddr;
-       else
-               new_tuple.dst.u3 = cp->vaddr;
-       /*
-        * This will also take care of UDP and other protocols.
-        */
-       if (outin)
-               new_tuple.src.u.tcp.port = cp->dport;
-       else
-               new_tuple.dst.u.tcp.port = cp->vport;
-       nf_conntrack_alter_reply(ct, &new_tuple);
-}
-
 /*
  *      NAT transmitter (only for outside-to-inside nat forwarding)
  *      Not used for related ICMP
@@ -434,8 +428,6 @@ ip_vs_nat_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
 
        IP_VS_DBG_PKT(10, pp, skb, 0, "After DNAT");
 
-       ip_vs_update_conntrack(skb, cp, 1);
-
        /* FIXME: when application helper enlarges the packet and the length
           is larger than the MTU of outgoing device, there will be still
           MTU problem. */
@@ -443,7 +435,7 @@ ip_vs_nat_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
        /* Another hack: avoid icmp_send in ip_fragment */
        skb->local_df = 1;
 
-       IP_VS_XMIT(NFPROTO_IPV4, skb, rt);
+       IP_VS_XMIT_NAT(NFPROTO_IPV4, skb, cp);
 
        LeaveFunction(10);
        return NF_STOLEN;
@@ -451,8 +443,8 @@ ip_vs_nat_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
   tx_error_icmp:
        dst_link_failure(skb);
   tx_error:
-       LeaveFunction(10);
        kfree_skb(skb);
+       LeaveFunction(10);
        return NF_STOLEN;
   tx_error_put:
        ip_rt_put(rt);
@@ -512,8 +504,6 @@ ip_vs_nat_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp,
 
        IP_VS_DBG_PKT(10, pp, skb, 0, "After DNAT");
 
-       ip_vs_update_conntrack(skb, cp, 1);
-
        /* FIXME: when application helper enlarges the packet and the length
           is larger than the MTU of outgoing device, there will be still
           MTU problem. */
@@ -521,7 +511,7 @@ ip_vs_nat_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp,
        /* Another hack: avoid icmp_send in ip_fragment */
        skb->local_df = 1;
 
-       IP_VS_XMIT(NFPROTO_IPV6, skb, rt);
+       IP_VS_XMIT_NAT(NFPROTO_IPV6, skb, cp);
 
        LeaveFunction(10);
        return NF_STOLEN;
@@ -571,6 +561,7 @@ ip_vs_tunnel_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
        struct iphdr  *iph;                     /* Our new IP header */
        unsigned int max_headroom;              /* The extra header space needed */
        int    mtu;
+       int ret;
 
        EnterFunction(10);
 
@@ -655,7 +646,11 @@ ip_vs_tunnel_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
        /* Another hack: avoid icmp_send in ip_fragment */
        skb->local_df = 1;
 
-       ip_local_out(skb);
+       ret = IP_VS_XMIT_TUNNEL(skb, cp);
+       if (ret == NF_ACCEPT)
+               ip_local_out(skb);
+       else if (ret == NF_DROP)
+               kfree_skb(skb);
 
        LeaveFunction(10);
 
@@ -681,6 +676,7 @@ ip_vs_tunnel_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp,
        struct ipv6hdr  *iph;           /* Our new IP header */
        unsigned int max_headroom;      /* The extra header space needed */
        int    mtu;
+       int ret;
 
        EnterFunction(10);
 
@@ -761,7 +757,11 @@ ip_vs_tunnel_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp,
        /* Another hack: avoid icmp_send in ip_fragment */
        skb->local_df = 1;
 
-       ip6_local_out(skb);
+       ret = IP_VS_XMIT_TUNNEL(skb, cp);
+       if (ret == NF_ACCEPT)
+               ip6_local_out(skb);
+       else if (ret == NF_DROP)
+               kfree_skb(skb);
 
        LeaveFunction(10);
 
@@ -820,7 +820,7 @@ ip_vs_dr_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
        /* Another hack: avoid icmp_send in ip_fragment */
        skb->local_df = 1;
 
-       IP_VS_XMIT(NFPROTO_IPV4, skb, rt);
+       IP_VS_XMIT(NFPROTO_IPV4, skb, cp);
 
        LeaveFunction(10);
        return NF_STOLEN;
@@ -873,7 +873,7 @@ ip_vs_dr_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp,
        /* Another hack: avoid icmp_send in ip_fragment */
        skb->local_df = 1;
 
-       IP_VS_XMIT(NFPROTO_IPV6, skb, rt);
+       IP_VS_XMIT(NFPROTO_IPV6, skb, cp);
 
        LeaveFunction(10);
        return NF_STOLEN;
@@ -947,7 +947,7 @@ ip_vs_icmp_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
        /* Another hack: avoid icmp_send in ip_fragment */
        skb->local_df = 1;
 
-       IP_VS_XMIT(NFPROTO_IPV4, skb, rt);
+       IP_VS_XMIT(NFPROTO_IPV4, skb, cp);
 
        rc = NF_STOLEN;
        goto out;
@@ -1022,7 +1022,7 @@ ip_vs_icmp_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp,
        /* Another hack: avoid icmp_send in ip_fragment */
        skb->local_df = 1;
 
-       IP_VS_XMIT(NFPROTO_IPV6, skb, rt);
+       IP_VS_XMIT(NFPROTO_IPV6, skb, cp);
 
        rc = NF_STOLEN;
        goto out;