diff mbox

netfilter: unfold two critical loops in ip_packet_match()

Message ID 499C3871.4030600@cosmosbay.com
State Not Applicable, archived
Delegated to: David Miller
Headers show

Commit Message

Eric Dumazet Feb. 18, 2009, 4:33 p.m. UTC 
Patrick McHardy a écrit :
> Eric Dumazet wrote:
>> Patrick McHardy a écrit :
>>> The interface name matching has shown up in profiles forever
>>> though and we've actually already tried to optimize it IIRC.
>>>
>>> Eric, I'm trying to keep all the *tables files synchronized,
>>> could you send me a patch updating the other ones as well
>>> please?
>>
>> While doing this, I found arp_tables is still using loop using
>> byte operations.
>>
>> Also, I could not find how iniface_mask[], outiface_mask[], iniface[]
>> and outiface[] were forced to long word alignment ...
>> (in struct ipt_ip, struct ip6t_ip6, struct arpt_arp)
> 
> In case of IPv4 and IPv6 they are already suitable aligned, it
> simply performing the comparison in unsigned long quantities.
> struct arpt_arp unfortunately doesn't properly align the interface
> names, so we need to continue to do byte-wise comparisons.
> 
> 

I see, but #ifdef CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS can help here ;)

ifname_compare() should be static in three files (ipv4_ip_tables, ipv6_ip_tables and arp_tables),
since only arp_tables variant has the alignement problem.

[PATCH] netfilter: unfold two critical loops in arp_packet_match()

x86 and powerpc can perform long word accesses in an efficient maner.
We can use this to unroll two loops in arp_packet_match(), to
perform arithmetic on long words instead of bytes. This is a win
on x86_64 for example.

Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
---
 net/ipv4/netfilter/arp_tables.c |   44 +++++++++++++++++++++++-------
 1 files changed, 34 insertions(+), 10 deletions(-)


--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Comments

Patrick McHardy Feb. 18, 2009, 4:52 p.m. UTC | #1 
Eric Dumazet wrote:
> Patrick McHardy a écrit :
>> In case of IPv4 and IPv6 they are already suitable aligned, it
>> simply performing the comparison in unsigned long quantities.
>> struct arpt_arp unfortunately doesn't properly align the interface
>> names, so we need to continue to do byte-wise comparisons.
>>
> I see, but #ifdef CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS can help here ;)

get_unaligned() would work as well I guess. But we don't seem to
have a get_unaligned_long().

> ifname_compare() should be static in three files (ipv4_ip_tables, ipv6_ip_tables and arp_tables),
> since only arp_tables variant has the alignement problem.
> 
> [PATCH] netfilter: unfold two critical loops in arp_packet_match()
> 
> x86 and powerpc can perform long word accesses in an efficient maner.
> We can use this to unroll two loops in arp_packet_match(), to
> perform arithmetic on long words instead of bytes. This is a win
> on x86_64 for example.

This looks good to me. Applied, thanks.
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
diff mbox

Patch

diff --git a/net/ipv4/netfilter/arp_tables.c b/net/ipv4/netfilter/arp_tables.c
index 7ea88b6..b5db463 100644
--- a/net/ipv4/netfilter/arp_tables.c
+++ b/net/ipv4/netfilter/arp_tables.c
@@ -73,6 +73,36 @@  static inline int arp_devaddr_compare(const struct arpt_devaddr_info *ap,
 	return (ret != 0);
 }
 
+/*
+ * Unfortunatly, _b and _mask are not aligned to an int (or long int)
+ * Some arches dont care, unrolling the loop is a win on them.
+ */
+static unsigned long ifname_compare(const char *_a, const char *_b, const char *_mask)
+{
+#ifdef CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS
+	const unsigned long *a = (const unsigned long *)_a;
+	const unsigned long *b = (const unsigned long *)_b;
+	const unsigned long *mask = (const unsigned long *)_mask;
+	unsigned long ret;
+
+	ret = (a[0] ^ b[0]) & mask[0];
+	if (IFNAMSIZ > sizeof(unsigned long))
+		ret |= (a[1] ^ b[1]) & mask[1];
+	if (IFNAMSIZ > 2 * sizeof(unsigned long))
+		ret |= (a[2] ^ b[2]) & mask[2];
+	if (IFNAMSIZ > 3 * sizeof(unsigned long))
+		ret |= (a[3] ^ b[3]) & mask[3];
+	BUILD_BUG_ON(IFNAMSIZ > 4 * sizeof(unsigned long));
+#else
+	unsigned long ret = 0;
+	int i;
+
+	for (i = 0; i < IFNAMSIZ; i++)
+		ret |= (_a[i] ^ _b[i]) & _mask[i];
+#endif
+	return ret;
+}
+
 /* Returns whether packet matches rule or not. */
 static inline int arp_packet_match(const struct arphdr *arphdr,
 				   struct net_device *dev,
@@ -83,7 +113,7 @@  static inline int arp_packet_match(const struct arphdr *arphdr,
 	const char *arpptr = (char *)(arphdr + 1);
 	const char *src_devaddr, *tgt_devaddr;
 	__be32 src_ipaddr, tgt_ipaddr;
-	int i, ret;
+	long ret;
 
 #define FWINV(bool, invflg) ((bool) ^ !!(arpinfo->invflags & (invflg)))
 
@@ -156,10 +186,7 @@  static inline int arp_packet_match(const struct arphdr *arphdr,
 	}
 
 	/* Look for ifname matches.  */
-	for (i = 0, ret = 0; i < IFNAMSIZ; i++) {
-		ret |= (indev[i] ^ arpinfo->iniface[i])
-			& arpinfo->iniface_mask[i];
-	}
+	ret = ifname_compare(indev, arpinfo->iniface, arpinfo->iniface_mask);
 
 	if (FWINV(ret != 0, ARPT_INV_VIA_IN)) {
 		dprintf("VIA in mismatch (%s vs %s).%s\n",
@@ -168,10 +195,7 @@  static inline int arp_packet_match(const struct arphdr *arphdr,
 		return 0;
 	}
 
-	for (i = 0, ret = 0; i < IFNAMSIZ; i++) {
-		ret |= (outdev[i] ^ arpinfo->outiface[i])
-			& arpinfo->outiface_mask[i];
-	}
+	ret = ifname_compare(outdev, arpinfo->outiface, arpinfo->outiface_mask);
 
 	if (FWINV(ret != 0, ARPT_INV_VIA_OUT)) {
 		dprintf("VIA out mismatch (%s vs %s).%s\n",
@@ -221,7 +245,7 @@  unsigned int arpt_do_table(struct sk_buff *skb,
 			   const struct net_device *out,
 			   struct xt_table *table)
 {
-	static const char nulldevname[IFNAMSIZ];
+	static const char nulldevname[IFNAMSIZ] __attribute__((aligned(sizeof(long))));
 	unsigned int verdict = NF_DROP;
 	const struct arphdr *arp;
 	bool hotdrop = false;