- converted to utf8

[packages/kernel.git] / pom-ng-unclean-20060504.patch

 Kconfig       |   14 +
 Makefile      |    1 
 ipt_unclean.c |  612 ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
 3 files changed, 627 insertions(+)

diff -Nur --exclude '*.orig' linux.org/net/ipv4/netfilter/Kconfig linux/net/ipv4/netfilter/Kconfig
--- linux.org/net/ipv4/netfilter/Kconfig	2006-05-02 23:38:44.000000000 +0200
+++ linux/net/ipv4/netfilter/Kconfig	2006-05-04 14:49:54.000000000 +0200
@@ -606,5 +606,19 @@
 	  Allows altering the ARP packet payload: source and destination
 	  hardware and network addresses.
 
+config IP_NF_MATCH_UNCLEAN
+	tristate  'Unclean match support (DANGEROUS)'
+	depends on EXPERIMENTAL && IP_NF_IPTABLES
+	help
+	  Unclean packet matching matches any strange or invalid packets, by
+	  looking at a series of fields in the IP, TCP, UDP and ICMP headers.  
+
+	  Please note that this kind of matching is considered dangerous and
+	  might harm the future compatibility of your packet filter.  
+	  
+	  It has happened before, search on the net for ECN blackholes :(
+
+
+
 endmenu
 
diff -Nur --exclude '*.orig' linux.org/net/ipv4/netfilter/Makefile linux/net/ipv4/netfilter/Makefile
--- linux.org/net/ipv4/netfilter/Makefile	2006-05-02 23:38:44.000000000 +0200
+++ linux/net/ipv4/netfilter/Makefile	2006-05-04 14:49:54.000000000 +0200
@@ -0,0 +0,1 @@
+obj-$(CONFIG_IP_NF_MATCH_UNCLEAN) += ipt_unclean.o
diff -Nur --exclude '*.orig' linux.org/net/ipv4/netfilter/ipt_unclean.c linux/net/ipv4/netfilter/ipt_unclean.c
--- linux.org/net/ipv4/netfilter/ipt_unclean.c	1970-01-01 01:00:00.000000000 +0100
+++ linux/net/ipv4/netfilter/ipt_unclean.c	2006-05-04 14:49:54.000000000 +0200
@@ -0,0 +1,612 @@
+/* Kernel module to match suspect packets. */
+#include <linux/module.h>
+#include <linux/skbuff.h>
+#include <linux/ip.h>
+#include <linux/udp.h>
+#include <linux/tcp.h>
+#include <linux/icmp.h>
+#include <net/checksum.h>
+#include <net/ip.h>
+
+#include <linux/netfilter_ipv4/ip_tables.h>
+
+#define limpk(format, args...)						 \
+do {									 \
+	if (net_ratelimit())						 \
+		printk("ipt_unclean: %s" format,			 \
+		       embedded ? "(embedded packet) " : "" , ## args);  \
+} while(0)
+
+enum icmp_error_status
+{
+	ICMP_MAY_BE_ERROR,
+	ICMP_IS_ERROR,
+	ICMP_NOT_ERROR
+};
+
+struct icmp_info
+{
+	size_t min_len, max_len;
+	enum icmp_error_status err;
+	u_int8_t min_code, max_code;
+};
+
+static int
+check_ip(const struct sk_buff *skb, unsigned int offset);
+
+/* ICMP-specific checks. */
+static int
+check_icmp(const struct sk_buff *skb,
+	   unsigned int offset,
+	   unsigned int fragoff,
+	   int more_frags,
+	   int embedded)
+{
+	struct icmphdr icmph;
+	static struct icmp_info info[]
+		= { [ICMP_ECHOREPLY]
+		    = { 8, 65536, ICMP_NOT_ERROR, 0, 0 },
+		    [ICMP_DEST_UNREACH]
+		    = { 8 + 28, 65536, ICMP_IS_ERROR, 0, 15 },
+		    [ICMP_SOURCE_QUENCH]
+		    = { 8 + 28, 65536, ICMP_IS_ERROR, 0, 0 },
+		    [ICMP_REDIRECT]
+		    = { 8 + 28, 65536, ICMP_IS_ERROR, 0, 3 },
+		    [ICMP_ECHO]
+		    = { 8, 65536, ICMP_NOT_ERROR, 0, 0  },
+		    /* Router advertisement. */
+		    [9]
+		    = { 8, 8 + 255 * 8, ICMP_NOT_ERROR, 0, 0 },
+		    /* Router solicitation. */
+		    [10]
+		    = { 8, 8, ICMP_NOT_ERROR, 0, 0 },
+		    [ICMP_TIME_EXCEEDED]
+		    = { 8 + 28, 65536, ICMP_IS_ERROR, 0, 1  },
+		    [ICMP_PARAMETERPROB]
+		    = { 8 + 28, 65536, ICMP_IS_ERROR, 0, 1 },
+		    [ICMP_TIMESTAMP]
+		    = { 20, 20, ICMP_NOT_ERROR, 0, 0 },
+		    [ICMP_TIMESTAMPREPLY]
+		    = { 20, 20, ICMP_NOT_ERROR, 0, 0 },
+		    [ICMP_INFO_REQUEST]
+		    = { 8, 65536, ICMP_NOT_ERROR, 0, 0 },
+		    [ICMP_INFO_REPLY]
+		    = { 8, 65536, ICMP_NOT_ERROR, 0, 0 },
+		    [ICMP_ADDRESS]
+		    = { 12, 12, ICMP_NOT_ERROR, 0, 0 },
+		    [ICMP_ADDRESSREPLY]
+		    = { 12, 12, ICMP_NOT_ERROR, 0, 0 } };
+
+	/* Can't do anything if it's a fragment. */
+	if (fragoff)
+		return 1;
+
+	/* CHECK: Must have whole header.. */
+	if (skb_copy_bits(skb, offset, &icmph, sizeof(icmph)) < 0) {
+		limpk("ICMP len=%u too short\n", skb->len - offset);
+		return 0;
+	}
+
+	/* If not embedded in an ICMP error already. */
+	if (!embedded) {
+		/* CHECK: Truncated ICMP (even if first fragment). */
+		if (icmph.type < sizeof(info)/sizeof(struct icmp_info)
+		    && info[icmph.type].min_len != 0
+		    && skb->len - offset < info[icmph.type].min_len) {
+			limpk("ICMP type %u len %u too short\n",
+			      icmph.type, skb->len - offset);
+			return 0;
+		}
+
+		/* CHECK: Check within known error ICMPs. */
+		if (icmph.type < sizeof(info)/sizeof(struct icmp_info)
+		    && info[icmph.type].err == ICMP_IS_ERROR) {
+			/* Max IP header size = 60 */
+			char inner[60 + 8];
+			struct iphdr *inner_ip = (struct iphdr *)inner;
+
+			/* CHECK: Embedded packet must be at least
+			   length of iph + 8 bytes. */
+			if (skb_copy_bits(skb, offset + sizeof(icmph),
+					  inner, sizeof(struct iphdr)+8) < 0) {
+				limpk("ICMP error internal way too short\n");
+				return 0;
+			}
+
+			/* iphhdr may actually be longer: still need 8
+                           actual protocol bytes. */
+			if (offset + sizeof(icmph) + inner_ip->ihl*4 + 8
+			    > skb->len) {
+				limpk("ICMP error internal too short\n");
+				return 0;
+			}
+			if (!check_ip(skb, offset + sizeof(icmph)))
+				return 0;
+		}
+	} else {
+		/* CHECK: Can't embed ICMP unless known non-error. */
+		if (icmph.type >= sizeof(info)/sizeof(struct icmp_info)
+		    || info[icmph.type].err != ICMP_NOT_ERROR) {
+			limpk("ICMP type %u not embeddable\n",
+			      icmph.type);
+			return 0;
+		}
+	}
+
+	/* CHECK: Invalid ICMP codes. */
+	if (icmph.type < sizeof(info)/sizeof(struct icmp_info)
+	    && (icmph.code < info[icmph.type].min_code
+		|| icmph.code > info[icmph.type].max_code)) {
+		limpk("ICMP type=%u code=%u\n",
+		      icmph.type, icmph.code);
+		return 0;
+	}
+
+	/* CHECK: Above maximum length. */
+	if (icmph.type < sizeof(info)/sizeof(struct icmp_info)
+	    && info[icmph.type].max_len != 0
+	    && skb->len - offset > info[icmph.type].max_len) {
+		limpk("ICMP type=%u too long: %u bytes\n",
+		      icmph.type, skb->len - offset);
+		return 0;
+	}
+
+	switch (icmph.type) {
+	case ICMP_PARAMETERPROB: {
+		/* CHECK: Problem param must be within error packet's
+		 * IP header. */
+		u_int32_t arg = ntohl(icmph.un.gateway);
+
+		if (icmph.code == 0) {
+			/* We've already made sure it's long enough. */
+			struct iphdr iph;
+			skb_copy_bits(skb, offset + sizeof(icmph), &iph,
+				      sizeof(iph));
+			/* Code 0 means that upper 8 bits is pointer
+                           to problem. */
+			if ((arg >> 24) >= iph.ihl*4) {
+				limpk("ICMP PARAMETERPROB ptr = %u\n",
+				      ntohl(icmph.un.gateway) >> 24);
+				return 0;
+			}
+			arg &= 0x00FFFFFF;
+		}
+
+		/* CHECK: Rest must be zero. */
+		if (arg) {
+			limpk("ICMP PARAMETERPROB nonzero arg = %u\n",
+			      arg);
+			return 0;
+		}
+		break;
+	}
+
+	case ICMP_TIME_EXCEEDED:
+	case ICMP_SOURCE_QUENCH:
+		/* CHECK: Unused must be zero. */
+		if (icmph.un.gateway != 0) {
+			limpk("ICMP type=%u unused = %u\n",
+			      icmph.type, ntohl(icmph.un.gateway));
+			return 0;
+		}
+		break;
+	}
+
+	return 1;
+}
+
+/* UDP-specific checks. */
+static int
+check_udp(const struct sk_buff *skb,
+	  unsigned int offset,
+	  unsigned int fragoff,
+	  int more_frags,
+	  int embedded)
+{
+	struct udphdr udph;
+
+	/* Can't do anything if it's a fragment. */
+	if (fragoff)
+		return 1;
+
+	/* CHECK: Must cover UDP header. */
+	if (skb_copy_bits(skb, offset, &udph, sizeof(udph)) < 0) {
+		limpk("UDP len=%u too short\n", skb->len - offset);
+		return 0;
+	}
+
+	/* CHECK: Destination port can't be zero. */
+	if (!udph.dest) {
+		limpk("UDP zero destination port\n");
+		return 0;
+	}
+
+	if (!more_frags) {
+		if (!embedded) {
+			/* CHECK: UDP length must match. */
+			if (ntohs(udph.len) != skb->len - offset) {
+				limpk("UDP len too short %u vs %u\n",
+				      ntohs(udph.len), skb->len - offset);
+				return 0;
+			}
+		} else {
+			/* CHECK: UDP length be >= this truncated pkt. */
+			if (ntohs(udph.len) < skb->len - offset) {
+				limpk("UDP len too long %u vs %u\n",
+				      ntohs(udph.len), skb->len - offset);
+				return 0;
+			}
+		}
+	} else {
+		/* CHECK: UDP length must be > this frag's length. */
+		if (ntohs(udph.len) <= skb->len - offset) {
+			limpk("UDP fragment len too short %u vs %u\n",
+			      ntohs(udph.len), skb->len - offset);
+			return 0;
+		}
+	}
+
+	return 1;
+}
+
+/* TCP-specific checks. */
+static int
+check_tcp(const struct sk_buff *skb,
+	  unsigned int offset,
+	  unsigned int fragoff,
+	  int more_frags,
+	  int embedded)
+{
+	struct tcphdr tcph; 
+	unsigned char opt[15 * 4 - sizeof(struct tcphdr)];
+	u32 tcpflags;
+	int end_of_options = 0;
+	unsigned int i, optlen;
+
+	/* CHECK: Can't have offset=1: used to override TCP syn-checks. */
+	/* In fact, this is caught below (offset < 516). */
+
+	/* Can't do anything if it's a fragment. */
+	if (fragoff)
+		return 1;
+
+	/* CHECK: Smaller than minimal TCP hdr. */
+	if (skb_copy_bits(skb, offset, &tcph, sizeof(tcph)) < 0) {
+		u16 ports[2];
+
+		if (!embedded) {
+			limpk("Packet length %u < TCP header.\n",
+			      skb->len - offset);
+			return 0;
+		}
+
+		/* Must have ports available (datalen >= 8), from
+                   check_icmp which set embedded = 1 */
+		/* CHECK: TCP ports inside ICMP error */
+		skb_copy_bits(skb, offset, ports, sizeof(ports));
+		if (!ports[0] || !ports[1]) {
+			limpk("Zero TCP ports %u/%u.\n",
+			      htons(ports[0]), htons(ports[1]));
+			return 0;
+		}
+		return 1;
+	}
+
+	/* CHECK: TCP header claims tiny size. */
+	if (tcph.doff * 4 < sizeof(tcph)) {
+		limpk("TCP header claims tiny size %u\n", tcph.doff * 4);
+		return 0;
+	}
+
+	/* CHECK: Packet smaller than actual TCP hdr. */
+	optlen = tcph.doff*4 - sizeof(tcph);
+	if (skb_copy_bits(skb, offset + sizeof(tcph), opt, optlen) < 0) {
+		if (!embedded) {
+			limpk("Packet length %u < actual TCP header.\n",
+			      skb->len - offset);
+			return 0;
+		} else
+			return 1;
+	}
+
+	/* CHECK: TCP ports non-zero */
+	if (!tcph.source || !tcph.dest) {
+		limpk("Zero TCP ports %u/%u.\n",
+		      htons(tcph.source), htons(tcph.dest));
+		return 0;
+	}
+
+	tcpflags = tcp_flag_word(&tcph);
+
+	/* CHECK: TCP reserved bits zero. */
+	if (tcpflags & TCP_RESERVED_BITS) {
+		limpk("TCP reserved bits not zero\n");
+		return 0;
+	}
+
+	tcpflags &= ~(TCP_DATA_OFFSET | TCP_FLAG_CWR | TCP_FLAG_ECE
+		      | __constant_htonl(0x0000FFFF));
+
+	/* CHECK: TCP flags. */
+	if (tcpflags != TCP_FLAG_SYN
+	    && tcpflags != (TCP_FLAG_SYN|TCP_FLAG_ACK)
+	    && tcpflags != TCP_FLAG_RST
+	    && tcpflags != (TCP_FLAG_RST|TCP_FLAG_ACK)
+	    && tcpflags != (TCP_FLAG_RST|TCP_FLAG_ACK|TCP_FLAG_PSH)
+	    && tcpflags != (TCP_FLAG_FIN|TCP_FLAG_ACK)
+	    && tcpflags != TCP_FLAG_ACK
+	    && tcpflags != (TCP_FLAG_ACK|TCP_FLAG_PSH)
+	    && tcpflags != (TCP_FLAG_ACK|TCP_FLAG_URG)
+	    && tcpflags != (TCP_FLAG_ACK|TCP_FLAG_URG|TCP_FLAG_PSH)
+	    && tcpflags != (TCP_FLAG_FIN|TCP_FLAG_ACK|TCP_FLAG_PSH)
+	    && tcpflags != (TCP_FLAG_FIN|TCP_FLAG_ACK|TCP_FLAG_URG)
+	    && tcpflags != (TCP_FLAG_FIN|TCP_FLAG_ACK|TCP_FLAG_URG
+			    |TCP_FLAG_PSH)) {
+		limpk("TCP flags bad: 0x%04X\n", ntohl(tcpflags) >> 16);
+		return 0;
+	}
+
+	for (i = 0; i < optlen; ) {
+		switch (opt[i]) {
+		case 0:
+			end_of_options = 1;
+			i++;
+			break;
+		case 1:
+			i++;
+			break;
+		default:
+			/* CHECK: options after EOO. */
+			if (end_of_options) {
+				limpk("TCP option %u after end\n",
+				      opt[i]);
+				return 0;
+			}
+			/* CHECK: options at tail. */
+			else if (i+1 >= optlen) {
+				limpk("TCP option %u at tail\n",
+				      opt[i]);
+				return 0;
+			}
+			/* CHECK: zero-length options. */
+			else if (opt[i+1] == 0) {
+				limpk("TCP option %u 0 len\n",
+				      opt[i]);
+				return 0;
+			}
+			/* CHECK: oversize options. */
+			else if (i + opt[i+1] > optlen) {
+				limpk("TCP option %u at %u too long\n",
+				      (unsigned int) opt[i], i);
+				return 0;
+			}
+			/* Move to next option */
+			i += opt[i+1];
+		}
+	}
+
+	return 1;
+}
+
+/* Returns 1 if ok */
+/* Standard IP checks. */
+static int
+check_ip(const struct sk_buff *skb, unsigned int offset)
+{
+	int end_of_options = 0;
+	unsigned int datalen, optlen;
+	unsigned int i;
+	unsigned int fragoff;
+	struct iphdr iph;
+	unsigned char opt[15 * 4 - sizeof(struct iphdr)];
+	int embedded = offset;
+
+	/* Should only happen for local outgoing raw-socket packets. */
+	/* CHECK: length >= ip header. */
+	if (skb_copy_bits(skb, offset, &iph, sizeof(iph)) < 0) {
+		limpk("Packet length %u < IP header.\n", skb->len - offset);
+		return 0;
+	}
+	if (iph.ihl * 4 < sizeof(iph)) {
+		limpk("IP len %u < minimum IP header.\n", iph.ihl*4);
+		return 0;
+	}
+
+	optlen = iph.ihl * 4 - sizeof(iph);
+	if (skb_copy_bits(skb, offset+sizeof(struct iphdr), opt, optlen)<0) {
+		limpk("Packet length %u < IP header %u.\n",
+		      skb->len - offset, iph.ihl * 4);
+		return 0;
+	}
+
+	fragoff = (ntohs(iph.frag_off) & IP_OFFSET);
+	datalen = skb->len - (offset + sizeof(struct iphdr) + optlen);
+
+	/* CHECK: Embedded fragment. */
+	if (offset && fragoff) {
+		limpk("Embedded fragment.\n");
+		return 0;
+	}
+
+	for (i = 0; i < optlen; ) {
+		switch (opt[i]) {
+		case 0:
+			end_of_options = 1;
+			i++;
+			break;
+		case 1:
+			i++;
+			break;
+		default:
+			/* CHECK: options after EOO. */
+			if (end_of_options) {
+				limpk("IP option %u after end\n",
+				      opt[i]);
+				return 0;
+			}
+			/* CHECK: options at tail. */
+			else if (i+1 >= optlen) {
+				limpk("IP option %u at tail\n",
+				      opt[i]);
+				return 0;
+			}
+			/* CHECK: zero-length or one-length options. */
+			else if (opt[i+1] < 2) {
+				limpk("IP option %u %u len\n",
+				      opt[i], opt[i+1]);
+				return 0;
+			}
+			/* CHECK: oversize options. */
+			else if (i + opt[i+1] > optlen) {
+				limpk("IP option %u at %u too long\n",
+				      opt[i], i);
+				return 0;
+			}
+			/* Move to next option */
+			i += opt[i+1];
+		}
+	}
+
+	/* Fragment checks. */
+
+	/* CHECK: More fragments, but doesn't fill 8-byte boundary. */
+	if ((ntohs(iph.frag_off) & IP_MF)
+	    && (ntohs(iph.tot_len) % 8) != 0) {
+		limpk("Truncated fragment %u long.\n", ntohs(iph.tot_len));
+		return 0;
+	}
+
+	/* CHECK: Oversize fragment a-la Ping of Death. */
+	if (fragoff * 8 + datalen > 65535) {
+		limpk("Oversize fragment to %u.\n", fragoff * 8);
+		return 0;
+	}
+
+	/* CHECK: DF set and fragoff or MF set. */
+	if ((ntohs(iph.frag_off) & IP_DF)
+	    && (fragoff || (ntohs(iph.frag_off) & IP_MF))) {
+		limpk("DF set and offset=%u, MF=%u.\n",
+		      fragoff, ntohs(iph.frag_off) & IP_MF);
+		return 0;
+	}
+
+	/* CHECK: Zero-sized fragments. */
+	if ((fragoff || (ntohs(iph.frag_off) & IP_MF))
+	    && datalen == 0) {
+		limpk("Zero size fragment offset=%u\n", fragoff);
+		return 0;
+	}
+
+	/* Note: we can have even middle fragments smaller than this:
+	   consider a large packet passing through a 600MTU then
+	   576MTU link: this gives a fragment of 24 data bytes.  But
+	   everyone packs fragments largest first, hence a fragment
+	   can't START before 576 - MAX_IP_HEADER_LEN. */
+
+	/* Used to be min-size 576: I recall Alan Cox saying ax25 goes
+	   down to 128 (576 taken from RFC 791: All hosts must be
+	   prepared to accept datagrams of up to 576 octets).  Use 128
+	   here. */
+#define MIN_LIKELY_MTU 128
+	/* CHECK: Min size of first frag = 128. */
+	if ((ntohs(iph.frag_off) & IP_MF)
+	    && fragoff == 0
+	    && ntohs(iph.tot_len) < MIN_LIKELY_MTU) {
+		limpk("First fragment size %u < %u\n", ntohs(iph.tot_len),
+		      MIN_LIKELY_MTU);
+		return 0;
+	}
+
+	/* CHECK: Min offset of frag = 128 - IP hdr len. */
+	if (fragoff && fragoff * 8 < MIN_LIKELY_MTU - iph.ihl * 4) {
+		limpk("Fragment starts at %u < %u\n", fragoff * 8,
+		      MIN_LIKELY_MTU - iph.ihl * 4);
+		return 0;
+	}
+
+	/* CHECK: Protocol specification non-zero. */
+	if (iph.protocol == 0) {
+		limpk("Zero protocol\n");
+		return 0;
+	}
+
+	/* FIXME: This is already checked for in "Oversize fragment"
+           above --RR */
+	/* CHECK: Do not use what is unused.
+	 * First bit of fragmentation flags should be unused.
+	 * May be used by OS fingerprinting tools.
+	 * 04 Jun 2002, Maciej Soltysiak, solt@dns.toxicfilms.tv
+	 */
+	if (ntohs(iph.frag_off)>>15) {
+		limpk("IP unused bit set\n");
+		return 0;
+	}
+
+	/* Per-protocol checks. */
+	switch (iph.protocol) {
+	case IPPROTO_ICMP:
+		return check_icmp(skb, offset + iph.ihl*4, fragoff,
+				  (ntohs(iph.frag_off) & IP_MF),
+				  embedded);
+
+	case IPPROTO_UDP:
+		return check_udp(skb, offset + iph.ihl*4, fragoff,
+				 (ntohs(iph.frag_off) & IP_MF),
+				 embedded);
+
+	case IPPROTO_TCP:
+		return check_tcp(skb, offset + iph.ihl*4, fragoff,
+				 (ntohs(iph.frag_off) & IP_MF),
+				 embedded);
+	default:
+		/* Ignorance is bliss. */
+		return 1;
+	}
+}
+
+static int
+match(const struct sk_buff *skb,
+      const struct net_device *in,
+      const struct net_device *out,
+      const void *matchinfo,
+      int offset,
+      unsigned int protoff,
+      int *hotdrop)
+{
+	return !check_ip(skb, 0);
+}
+
+/* Called when user tries to insert an entry of this type. */
+static int
+checkentry(const char *tablename,
+	   const void *ip,
+	   void *matchinfo,
+	   unsigned int matchsize,
+	   unsigned int hook_mask)
+{
+	if (matchsize != IPT_ALIGN(0))
+		return 0;
+
+	return 1;
+}
+
+static struct ipt_match unclean_match = {
+	.name		= "unclean",
+	.match		= &match,
+	.checkentry	= &checkentry,
+	.me		= THIS_MODULE,
+};
+
+static int __init init(void)
+{
+	return ipt_register_match(&unclean_match);
+}
+
+static void __exit fini(void)
+{
+	ipt_unregister_match(&unclean_match);
+}
+
+module_init(init);
+module_exit(fini);
+MODULE_LICENSE("GPL");