Skip to contentSkip to conversation

: amanda helper: convert to textsearch infrastructure

RFC In review Patrick McHardy 2 participants 4 messages May 19, 2006 – May 24, 2006

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

RFC 1/1 : amanda helper: convert to textsearch infrastructure

I've just converted the amanda helper to the textsearch infrastructure to fix a packet corruption bug. This allows to avoid the data copying and save roughly 64k of memory. The IRC helper looks like a good candidate for this too, FTP is probably quite a bit more complicated.

Comments welcome.

[NETFILTER]: amanda helper: convert to textsearch infrastructure

When a port number within a packet is replaced by a differently sized number only the packet is resized, but not the copy of the data. Following port numbers are rewritten based on their offsets within the copy, leading to packet corruption.

Convert the amanda helper to the textsearch infrastructure to avoid the copy entirely.

---
commit 23b9a4b7fd6c16efd069b1b4de6e605fdc56634c
tree 0145077411b1866778a745d14805baaa7d5f290f
parent a54c9d30dbb06391ec4422aaf0e1dc2c8c53bd3e
author Patrick McHardy <[email protected]> Fri, 19 May 2006 18:30:54 +0200 committer Patrick McHardy <[email protected]> Fri, 19 May 2006 18:30:54 +0200

 net/ipv4/netfilter/Kconfig               |    2 
 net/ipv4/netfilter/ip_conntrack_amanda.c |  140 ++++++++++++++++++++----------
 2 files changed, 93 insertions(+), 49 deletions(-)
net/ipv4/netfilter/Kconfig +2 -0
index 3d560de..d859754 100644
@@ -142,6 +142,8 @@ config IP_NF_TFTP
 config IP_NF_AMANDA
 	tristate "Amanda backup protocol support"
 	depends on IP_NF_CONNTRACK
+	select TEXTSEARCH
+	select TEXTSEARCH_KMP
 	help
 	  If you are running the Amanda backup package <http://www.amanda.org/>
 	  on this machine or machines that will be MASQUERADED through this
net/ipv4/netfilter/ip_conntrack_amanda.c +91 -49

207 lines not on this page · raw message

#1 lore raw
Quoted 4 lines · Patrick McHardy

I've just converted the amanda helper to the textsearch infrastructure to fix a packet corruption bug. This allows to avoid the data copying and save roughly 64k of memory. The IRC helper looks like a good candidate for this too, FTP is probably quite a bit more complicated.

It looks fine, just two minor comments. BTW, thanks for doing this: This port is on my TODO list since the workshop.

Quoted 146 lines

------------------------------------------------------------------------

[NETFILTER]: amanda helper: convert to textsearch infrastructure

When a port number within a packet is replaced by a differently sized number only the packet is resized, but not the copy of the data. Following port numbers are rewritten based on their offsets within the copy, leading to packet corruption.

Convert the amanda helper to the textsearch infrastructure to avoid the copy entirely.

---
commit 23b9a4b7fd6c16efd069b1b4de6e605fdc56634c
tree 0145077411b1866778a745d14805baaa7d5f290f
parent a54c9d30dbb06391ec4422aaf0e1dc2c8c53bd3e
author Patrick McHardy <[email protected]> Fri, 19 May 2006 18:30:54 +0200 committer Patrick McHardy <[email protected]> Fri, 19 May 2006 18:30:54 +0200

 net/ipv4/netfilter/Kconfig               |    2 
 net/ipv4/netfilter/ip_conntrack_amanda.c |  140 ++++++++++++++++++++----------
 2 files changed, 93 insertions(+), 49 deletions(-)
net/ipv4/netfilter/Kconfig +2 -0
index 3d560de..d859754 100644
@@ -142,6 +142,8 @@ config IP_NF_TFTP
 config IP_NF_AMANDA
 	tristate "Amanda backup protocol support"
 	depends on IP_NF_CONNTRACK
+	select TEXTSEARCH
+	select TEXTSEARCH_KMP
 	help
 	  If you are running the Amanda backup package <http://www.amanda.org/>
 	  on this machine or machines that will be MASQUERADED through this
net/ipv4/netfilter/ip_conntrack_amanda.c +47 -23
index a604b1c..34b2309 100644
@@ -17,17 +17,16 @@
  *	this value.
  *
  */
-
-#include <linux/in.h>
 #include <linux/kernel.h>
 #include <linux/module.h>
-#include <linux/netfilter.h>
-#include <linux/ip.h>
 #include <linux/moduleparam.h>
+#include <linux/textsearch.h>
+#include <linux/skbuff.h>
+#include <linux/in.h>
+#include <linux/ip.h>
 #include <linux/udp.h>
-#include <net/checksum.h>
-#include <net/udp.h>
 
+#include <linux/netfilter.h>
 #include <linux/netfilter_ipv4/ip_conntrack_helper.h>
 #include <linux/netfilter_ipv4/ip_conntrack_amanda.h>
 
@@ -39,12 +38,6 @@ MODULE_LICENSE("GPL");
 module_param(master_timeout, uint, 0600);
 MODULE_PARM_DESC(master_timeout, "timeout for the master connection");
 
-static const char *conns[] = { "DATA ", "MESG ", "INDEX " };
-
-/* This is slow, but it's simple. --RR */
-static char *amanda_buffer;
-static DEFINE_SPINLOCK(amanda_buffer_lock);
-
 unsigned int (*ip_nat_amanda_hook)(struct sk_buff **pskb,
 				   enum ip_conntrack_info ctinfo,
 				   unsigned int matchoff,
@@ -52,12 +45,48 @@ unsigned int (*ip_nat_amanda_hook)(struc
 				   struct ip_conntrack_expect *exp);
 EXPORT_SYMBOL_GPL(ip_nat_amanda_hook);
 
+enum amanda_strings {
+	SEARCH_CONNECT,
+	SEARCH_NEWLINE,
+	SEARCH_DATA,
+	SEARCH_MESG,
+	SEARCH_INDEX,
+};
+
+static struct {
+	char			*string;
+	size_t			len;
+	struct ts_config	*ts;
+} search[] = {
+	[SEARCH_CONNECT] = {
+		.string	= "CONNECT ",
+		.len	= 8,
+	},
+	[SEARCH_NEWLINE] = {
+		.string	= "\n",
+		.len	= 1,
+	},
+	[SEARCH_DATA] = {
+		.string	= "DATA ",
+		.len	= 5,
+	},
+	[SEARCH_MESG] = {
+		.string	= "MESG ",
+		.len	= 5,
+	},
+	[SEARCH_INDEX] = {
+		.string = "INDEX ",
+		.len	= 6,
+	},
+};
+
 static int help(struct sk_buff **pskb,
                 struct ip_conntrack *ct, enum ip_conntrack_info ctinfo)
 {
+	struct ts_state ts;
 	struct ip_conntrack_expect *exp;
-	char *data, *data_limit, *tmp;
-	unsigned int dataoff, i;
+	unsigned int dataoff, start, stop, off, i;
+	char pbuf[sizeof("65535")], *tmp;
 	u_int16_t port, len;
 	int ret = NF_ACCEPT;
 
@@ -77,29 +106,34 @@ static int help(struct sk_buff **pskb,
 		return NF_ACCEPT;
 	}
 
-	spin_lock_bh(&amanda_buffer_lock);
-	skb_copy_bits(*pskb, dataoff, amanda_buffer, (*pskb)->len - dataoff);
-	data = amanda_buffer;
-	data_limit = amanda_buffer + (*pskb)->len - dataoff;
-	*data_limit = '\0';
-
-	/* Search for the CONNECT string */
-	data = strstr(data, "CONNECT ");
-	if (!data)
+	memset(&ts, 0, sizeof(ts));
+	start = skb_find_text(*pskb, dataoff, (*pskb)->len,
+			      search[SEARCH_CONNECT].ts, &ts);
+	if (start == UINT_MAX)

net/ipv4/netfilter/ip_conntrack_amanda.c static int help(struct sk_buff **pskb, · line 118

Quoted 10 lines
net/ipv4/netfilter/ip_conntrack_amanda.c +4 -4
 		goto out;
-	data += strlen("CONNECT ");
+	start += dataoff + search[SEARCH_CONNECT].len;
 
-	/* Only search first line. */	
-	if ((tmp = strchr(data, '\n')))
-		*tmp = '\0';
+	memset(&ts, 0, sizeof(ts));
+	stop = skb_find_text(*pskb, start, (*pskb)->len,
+			     search[SEARCH_NEWLINE].ts, &ts);

Better use strstr here: the textsearch infrastructure implements algorithms only introduce trade off for patterns longer than one byte. So, in this case, the infrastructure can not help in any way.

Quoted 72 lines
net/ipv4/netfilter/ip_conntrack_amanda.c +22 -18
+	if (stop == UINT_MAX)
+		goto out;
+	stop += start;
 
-	for (i = 0; i < ARRAY_SIZE(conns); i++) {
-		char *match = strstr(data, conns[i]);
-		if (!match)
+	for (i = SEARCH_DATA; i <= SEARCH_INDEX; i++) {
+		memset(&ts, 0, sizeof(ts));
+		off = skb_find_text(*pskb, start, stop, search[i].ts, &ts);
+		if (off == UINT_MAX)
 			continue;
-		tmp = data = match + strlen(conns[i]);
-		port = simple_strtoul(data, &data, 10);
-		len = data - tmp;
+		off += start + search[i].len;
+
+		len = min_t(unsigned int, sizeof(pbuf) - 1, stop - off);
+		if (skb_copy_bits(*pskb, off, pbuf, len))
+			break;
+		pbuf[len] = '\0';
+
+		port = simple_strtoul(pbuf, &tmp, 10);
+		len = tmp - pbuf;
 		if (port == 0 || len > 5)
 			break;
 
@@ -125,8 +159,7 @@ static int help(struct sk_buff **pskb,
 		exp->mask.dst.u.tcp.port = 0xFFFF;
 
 		if (ip_nat_amanda_hook)
-			ret = ip_nat_amanda_hook(pskb, ctinfo,
-						 tmp - amanda_buffer,
+			ret = ip_nat_amanda_hook(pskb, ctinfo, off - dataoff,
 						 len, exp);
 		else if (ip_conntrack_expect_related(exp) != 0)
 			ret = NF_DROP;
@@ -134,12 +167,11 @@ static int help(struct sk_buff **pskb,
 	}
 
 out:
-	spin_unlock_bh(&amanda_buffer_lock);
 	return ret;
 }
 
 static struct ip_conntrack_helper amanda_helper = {
-	.max_expected = ARRAY_SIZE(conns),
+	.max_expected = 3,
 	.timeout = 180,
 	.me = THIS_MODULE,
 	.help = help,
@@ -155,26 +187,36 @@ static struct ip_conntrack_helper amanda
 
 static void __exit ip_conntrack_amanda_fini(void)
 {
+	int i;
+
 	ip_conntrack_helper_unregister(&amanda_helper);
-	kfree(amanda_buffer);
+	for (i = 0; i < ARRAY_SIZE(search); i++)
+		textsearch_destroy(search[i].ts);
 }
 
 static int __init ip_conntrack_amanda_init(void)
 {
-	int ret;
-
-	amanda_buffer = kmalloc(65536, GFP_KERNEL);
-	if (!amanda_buffer)
-		return -ENOMEM;
-
-	ret = ip_conntrack_helper_register(&amanda_helper);

net/ipv4/netfilter/ip_conntrack_amanda.c static struct ip_conntrack_helper amanda · line 205

Quoted 10 lines
net/ipv4/netfilter/ip_conntrack_amanda.c +7 -3
-	if (ret < 0) {
-		kfree(amanda_buffer);
-		return ret;
+	int ret, i;
+
+	ret = -ENOMEM;
+	for (i = 0; i < ARRAY_SIZE(search); i++) {
+		search[i].ts = textsearch_prepare("kmp", search[i].string,
+						  search[i].len,
+						  GFP_KERNEL, TS_AUTOLOAD);

I think that the algorithm should be configurable. Although Boyer-Moore would not find a matching if the traffic is defragmented (see the header of /lib/ts_bm.c), it definitely scales better, so some sysadmins could increase performance to the detriment of defragmented traffic:

http://people.netfilter.org/~pablo/textsearch/

I have an old patch for the FTP helper here somewhere, I could give it some spins and send it to you.

Signature · 1 line
The dawn of the fourth age of Linux firewalling is coming; a time of great struggle and heroic deeds -- J.Kadlecsik got inspired by J.Morris
#2 lore raw
Quoted 7 lines · Pablo Neira Ayuso

Patrick McHardy wrote:
- *tmp = '\0';

net/ipv4/netfilter/ip_conntrack_amanda.c static int help(struct sk_buff **pskb, · line 118

Quoted 3 lines
net/ipv4/netfilter/ip_conntrack_amanda.c +3 -0
+    memset(&ts, 0, sizeof(ts));
+    stop = skb_find_text(*pskb, start, (*pskb)->len,
+                 search[SEARCH_NEWLINE].ts, &ts);

Better use strstr here: the textsearch infrastructure implements algorithms only introduce trade off for patterns longer than one byte. So, in this case, the infrastructure can not help in any way.

Please, forget this comment, strstr is not fragment-aware.

Signature · 1 line
The dawn of the fourth age of Linux firewalling is coming; a time of great struggle and heroic deeds -- J.Kadlecsik got inspired by J.Morris
#3 lore raw
In reply to Pablo Neira Ayuso
Quoted 37 lines · Pablo Neira Ayuso
Quoted 4 lines · Patrick McHardy

I've just converted the amanda helper to the textsearch infrastructure to fix a packet corruption bug. This allows to avoid the data copying and save roughly 64k of memory. The IRC helper looks like a good candidate for this too, FTP is probably quite a bit more complicated.

It looks fine, just two minor comments. BTW, thanks for doing this: This port is on my TODO list since the workshop.

Quoted 9 lines

static int __init ip_conntrack_amanda_init(void)
{

net/ipv4/netfilter/ip_conntrack_amanda.c +0 -7
-    int ret;
-
-    amanda_buffer = kmalloc(65536, GFP_KERNEL);
-    if (!amanda_buffer)
-        return -ENOMEM;
-
-    ret = ip_conntrack_helper_register(&amanda_helper);

net/ipv4/netfilter/ip_conntrack_amanda.c static struct ip_conntrack_helper amanda · line 205

Quoted 10 lines
net/ipv4/netfilter/ip_conntrack_amanda.c +7 -3
-    if (ret < 0) {
-        kfree(amanda_buffer);
-        return ret;
+    int ret, i;
+
+    ret = -ENOMEM;
+    for (i = 0; i < ARRAY_SIZE(search); i++) {
+        search[i].ts = textsearch_prepare("kmp", search[i].string,
+                          search[i].len,
+                          GFP_KERNEL, TS_AUTOLOAD);

I think that the algorithm should be configurable. Although Boyer-Moore would not find a matching if the traffic is defragmented (see the header of /lib/ts_bm.c), it definitely scales better, so some sysadmins could increase performance to the detriment of defragmented traffic:

Quoted 1 line

http://people.netfilter.org/~pablo/textsearch/

Yes, that really looks better. I guess we could add a module-parameter, but right now it wouldn't help much because the individual algorithms can't be configured by the user, they can only be chosen with "select TEXTSEARCH_...".

Quoted 2 lines

I have an old patch for the FTP helper here somewhere, I could give it some spins and send it to you.

Please, I would like to see how complicated it gets. I think what would really help is something like the fsm algorithm if it would store the offsets of the individual matches. That would allow to specify all the possible matches as a single pattern.