NFQUEUE balancing extension (userspace changes)

NFQUEUE balancing extension (userspace changes)

[Florian Westphal]

Hello list,

this adds a new revision of the NFQUEUE target.

In particular, a new "--queue-balance" option, which allows to
specify a range of queues to use.
Packets are then balanced across the given queues by the kernel.

This is useful for multicore systems:
start multiple instances of the userspace program on queues
x, x+1, .. x+n and use "--queue-balance x:x+n".

As for naming, if you'd prefer a different name (or just extend
the existing --queue-num option to accept "firstid:lastid" instead),
please let me know and I'll be happy to change it.

Both paches can also be pulled from
git://git.breakpoint.cc/fw/iptables.git nfq-balance

Thanks for reviewing,
Florian

[PATCH 1/2] NFQUEUE: use NFPROTO_UNSPEC

[Florian Westphal]

also, xtables_strtoui() does the range check for us, no need for binary "&".

Signed-off-by: Florian Westphal <fw@strlen.de>
---
 extensions/libxt_NFQUEUE.c |   20 +++-----------------
 1 files changed, 3 insertions(+), 17 deletions(-)

diff --git a/extensions/libxt_NFQUEUE.c b/extensions/libxt_NFQUEUE.c
index 3ca2239..6939c6f 100644
--- a/extensions/libxt_NFQUEUE.c
+++ b/extensions/libxt_NFQUEUE.c
@@ -32,12 +32,12 @@ static void
 parse_num(const char *s, struct xt_NFQ_info *tinfo)
 {
 	unsigned int num;
-       
+
 	if (!xtables_strtoui(s, NULL, &num, 0, UINT16_MAX))
 		xtables_error(PARAMETER_PROBLEM,
 			   "Invalid queue number `%s'\n", s);
 
-    	tinfo->queuenum = num & 0xffff;
+	tinfo->queuenum = num;
 }
 
 static int
@@ -78,7 +78,7 @@ static void NFQUEUE_save(const void *ip, const struct xt_entry_target *target)
 }
 
 static struct xtables_target nfqueue_target = {
-	.family		= NFPROTO_IPV4,
+	.family		= NFPROTO_UNSPEC,
 	.name		= "NFQUEUE",
 	.version	= XTABLES_VERSION,
 	.size		= XT_ALIGN(sizeof(struct xt_NFQ_info)),
@@ -90,21 +90,7 @@ static struct xtables_target nfqueue_target = {
 	.extra_opts	= NFQUEUE_opts
 };
 
-static struct xtables_target nfqueue_target6 = {
-	.family		= NFPROTO_IPV6,
-	.name		= "NFQUEUE",
-	.version	= XTABLES_VERSION,
-	.size		= XT_ALIGN(sizeof(struct xt_NFQ_info)),
-	.userspacesize	= XT_ALIGN(sizeof(struct xt_NFQ_info)),
-	.help		= NFQUEUE_help,
-	.parse		= NFQUEUE_parse,
-	.print		= NFQUEUE_print,
-	.save		= NFQUEUE_save,
-	.extra_opts	= NFQUEUE_opts,
-};
-
 void _init(void)
 {
 	xtables_register_target(&nfqueue_target);
-	xtables_register_target(&nfqueue_target6);
 }
-- 
1.6.0.6

Re: [PATCH 1/2] NFQUEUE: use NFPROTO_UNSPEC

[Jan Engelhardt]

On Friday 2009-06-05 03:17, Florian Westphal wrote:
>--- a/extensions/libxt_NFQUEUE.c
>+++ b/extensions/libxt_NFQUEUE.c
>@@ -32,12 +32,12 @@ static void
> parse_num(const char *s, struct xt_NFQ_info *tinfo)
> {
> 	unsigned int num;
>-       
>+
> 	if (!xtables_strtoui(s, NULL, &num, 0, UINT16_MAX))
> 		xtables_error(PARAMETER_PROBLEM,
> 			   "Invalid queue number `%s'\n", s);

Extra line unintended?

Re: [PATCH 1/2] NFQUEUE: use NFPROTO_UNSPEC

[Florian Westphal]

Jan Engelhardt <jengelh@medozas.de> wrote:
> On Friday 2009-06-05 03:17, Florian Westphal wrote:
> >--- a/extensions/libxt_NFQUEUE.c
> >+++ b/extensions/libxt_NFQUEUE.c
> >@@ -32,12 +32,12 @@ static void
> > parse_num(const char *s, struct xt_NFQ_info *tinfo)
> > {
> > 	unsigned int num;
> >-       
> >+
> > 	if (!xtables_strtoui(s, NULL, &num, 0, UINT16_MAX))
> > 		xtables_error(PARAMETER_PROBLEM,
> > 			   "Invalid queue number `%s'\n", s);
> 
> Extra line unintended?

It trims extra whitespace.

[PATCH 2/2] NFQUEUE: add new v1 version with queue-balance option

[Florian Westphal]

new version that adds support for specifying a queue range instead
of a single queue id.
The kernel will distribute flows across the given queue range.

This is useful for multicore systems, simply start multiple instances
of the userspace program on queues x, x+1, .. x+n and use
"--queue-balance x:x+n".
Packets belonging to the same connection are put into the same queue.

Signed-off-by: Florian Westphal <fwestphal@astaro.com>
---
 extensions/libxt_NFQUEUE.c           |  127 +++++++++++++++++++++++++++++++++-
 extensions/libxt_NFQUEUE.man         |   10 +++
 include/linux/netfilter/xt_NFQUEUE.h |    5 ++
 3 files changed, 140 insertions(+), 2 deletions(-)

diff --git a/extensions/libxt_NFQUEUE.c b/extensions/libxt_NFQUEUE.c
index 6939c6f..bf75e63 100644
--- a/extensions/libxt_NFQUEUE.c
+++ b/extensions/libxt_NFQUEUE.c
@@ -23,19 +23,36 @@ static void NFQUEUE_help(void)
 );
 }
 
+static void NFQUEUE_help_v1(void)
+{
+	NFQUEUE_help();
+	printf(
+"  --queue-balance first:last	Balance flows between queues <value> to <value>.\n");
+}
+
 static const struct option NFQUEUE_opts[] = {
 	{ "queue-num", 1, NULL, 'F' },
 	{ .name = NULL }
 };
 
+static const struct option NFQUEUE_opts_v1[] = {
+	{ "queue-num", 1, NULL, 'F' },
+	{ "queue-balance", 1, NULL, 'B' },
+	{ .name = NULL }
+};
+
+static void exit_badqueue(const char *s)
+{
+	xtables_error(PARAMETER_PROBLEM, "Invalid queue number `%s'\n", s);
+}
+
 static void
 parse_num(const char *s, struct xt_NFQ_info *tinfo)
 {
 	unsigned int num;
 
 	if (!xtables_strtoui(s, NULL, &num, 0, UINT16_MAX))
-		xtables_error(PARAMETER_PROBLEM,
-			   "Invalid queue number `%s'\n", s);
+		exit_badqueue(s);
 
 	tinfo->queuenum = num;
 }
@@ -61,6 +78,47 @@ NFQUEUE_parse(int c, char **argv, int invert, unsigned int *flags,
 	return 1;
 }
 
+static int
+NFQUEUE_parse_v1(int c, char **argv, int invert, unsigned int *flags,
+                 const void *entry, struct xt_entry_target **target)
+{
+	struct xt_NFQ_info_v1 *info = (void *)(*target)->data;
+	char *colon;
+	unsigned int firstqueue, lastqueue;
+
+	switch (c) {
+	case 'F': /* fallthrough */
+	case 'B':
+		if (*flags)
+			xtables_error(PARAMETER_PROBLEM, "NFQUEUE target: "
+				   "Only use --queue-num ONCE!");
+
+		if (!xtables_strtoui(optarg, &colon, &firstqueue, 0, UINT16_MAX))
+			exit_badqueue(optarg);
+
+		info->queuenum = firstqueue;
+
+		if (c == 'F')
+			break;
+
+		if (*colon != ':')
+			xtables_error(PARAMETER_PROBLEM, "Bad range \"%s\"", optarg);
+
+		if (!xtables_strtoui(colon + 1, NULL, &lastqueue, 1, UINT16_MAX))
+			exit_badqueue(optarg);
+
+		if (firstqueue >= lastqueue)
+			xtables_error(PARAMETER_PROBLEM, "%u should be less than %u",
+							firstqueue, lastqueue);
+		info->queues_total = lastqueue - firstqueue + 1;
+		break;
+	default:
+		return 0;
+	}
+
+	return 1;
+}
+
 static void NFQUEUE_print(const void *ip,
                           const struct xt_entry_target *target, int numeric)
 {
@@ -69,6 +127,20 @@ static void NFQUEUE_print(const void *ip,
 	printf("NFQUEUE num %u", tinfo->queuenum);
 }
 
+static void NFQUEUE_print_v1(const void *ip,
+                             const struct xt_entry_target *target, int numeric)
+{
+	const struct xt_NFQ_info_v1 *tinfo = (const void *)target->data;
+	unsigned int last = tinfo->queues_total;
+
+	if (last > 1) {
+		last += tinfo->queuenum - 1;
+		printf("NFQUEUE balance %u:%u", tinfo->queuenum, last);
+	} else {
+		printf("NFQUEUE num %u", tinfo->queuenum);
+	}
+}
+
 static void NFQUEUE_save(const void *ip, const struct xt_entry_target *target)
 {
 	const struct xt_NFQ_info *tinfo =
@@ -77,6 +149,25 @@ static void NFQUEUE_save(const void *ip, const struct xt_entry_target *target)
 	printf("--queue-num %u ", tinfo->queuenum);
 }
 
+static void NFQUEUE_save_v1(const void *ip, const struct xt_entry_target *target)
+{
+	const struct xt_NFQ_info_v1 *tinfo = (const void *)target->data;
+	unsigned int last = tinfo->queues_total;
+
+	if (last > 1) {
+		last += tinfo->queuenum - 1;
+		printf("--queue-balance %u:%u ", tinfo->queuenum, last);
+	} else {
+		printf("--queue-num %u ", tinfo->queuenum);
+	}
+}
+
+static void NFQUEUE_init_v1(struct xt_entry_target *t)
+{
+	struct xt_NFQ_info_v1 *tinfo = (void *)t->data;
+	tinfo->queues_total = 1;
+}
+
 static struct xtables_target nfqueue_target = {
 	.family		= NFPROTO_UNSPEC,
 	.name		= "NFQUEUE",
@@ -90,7 +181,39 @@ static struct xtables_target nfqueue_target = {
 	.extra_opts	= NFQUEUE_opts
 };
 
+static struct xtables_target nfqueue_target4_v1 = {
+	.family		= NFPROTO_IPV4,
+	.revision	= 1,
+	.name		= "NFQUEUE",
+	.version	= XTABLES_VERSION,
+	.size		= XT_ALIGN(sizeof(struct xt_NFQ_info_v1)),
+	.userspacesize	= XT_ALIGN(sizeof(struct xt_NFQ_info_v1)),
+	.help		= NFQUEUE_help_v1,
+	.init		= NFQUEUE_init_v1,
+	.parse		= NFQUEUE_parse_v1,
+	.print		= NFQUEUE_print_v1,
+	.save		= NFQUEUE_save_v1,
+	.extra_opts	= NFQUEUE_opts_v1,
+};
+
+static struct xtables_target nfqueue_target6_v1 = {
+	.family		= NFPROTO_IPV6,
+	.revision	= 1,
+	.name		= "NFQUEUE",
+	.version	= XTABLES_VERSION,
+	.size		= XT_ALIGN(sizeof(struct xt_NFQ_info_v1)),
+	.userspacesize	= XT_ALIGN(sizeof(struct xt_NFQ_info_v1)),
+	.help		= NFQUEUE_help_v1,
+	.init		= NFQUEUE_init_v1,
+	.parse		= NFQUEUE_parse_v1,
+	.print		= NFQUEUE_print_v1,
+	.save		= NFQUEUE_save_v1,
+	.extra_opts	= NFQUEUE_opts_v1,
+};
+
 void _init(void)
 {
 	xtables_register_target(&nfqueue_target);
+	xtables_register_target(&nfqueue_target4_v1);
+	xtables_register_target(&nfqueue_target6_v1);
 }
diff --git a/extensions/libxt_NFQUEUE.man b/extensions/libxt_NFQUEUE.man
index b2c90bb..db01021 100644
--- a/extensions/libxt_NFQUEUE.man
+++ b/extensions/libxt_NFQUEUE.man
@@ -5,8 +5,18 @@ number.
 \fB\-\-queue\-num\fP \fIvalue\fP
 This specifies the QUEUE number to use. Valid queue numbers are 0 to 65535. The default value is 0.
 .PP
+.TP
+\fB\-\-queue\-balance\fP \fIvalue\fP:\fIvalue\fP
+This specifies a range of queues to use. Packets are then balanced across the given queues.
+This is useful for multicore systems: start multiple instances of the userspace program on
+queues x, x+1, .. x+n and use "--queue-balance x:x+n".
+Packets belonging to the same connection are put into the same nfqueue.
+.PP
 It can only be used with Kernel versions 2.6.14 or later, since it requires
 the
 .B
 nfnetlink_queue
 kernel support.
+.B
+queue-balance
+support was added in Linux 2.6.31.
diff --git a/include/linux/netfilter/xt_NFQUEUE.h b/include/linux/netfilter/xt_NFQUEUE.h
index 9a9af79..ab6d62b 100644
--- a/include/linux/netfilter/xt_NFQUEUE.h
+++ b/include/linux/netfilter/xt_NFQUEUE.h
@@ -13,4 +13,9 @@ struct xt_NFQ_info {
 	u_int16_t queuenum;
 };
 
+struct xt_NFQ_info_v1 {
+	u_int16_t queuenum;
+	u_int16_t queues_total;
+};
+
 #endif /* _XT_NFQ_TARGET_H */
-- 
1.6.0.6

Re: [PATCH 2/2] NFQUEUE: add new v1 version with queue-balance option

[Jan Engelhardt]

On Friday 2009-06-05 03:17, Florian Westphal wrote:
>target4_v1{
>+	.revision	= 1,
>+	.name		= "NFQUEUE",
>+	.version	= XTABLES_VERSION,
>+	.size		= XT_ALIGN(sizeof(struct xt_NFQ_info_v1)),
>+	.userspacesize	= XT_ALIGN(sizeof(struct xt_NFQ_info_v1)),
>+	.help		= NFQUEUE_help_v1,
>+	.init		= NFQUEUE_init_v1,
>+	.parse		= NFQUEUE_parse_v1,
>+	.print		= NFQUEUE_print_v1,
>+	.save		= NFQUEUE_save_v1,
>+	.extra_opts	= NFQUEUE_opts_v1,
>+};
>+
>+static struct xtables_target nfqueue_target6_v1 = {
>+	.family		= NFPROTO_IPV6,
>+	.revision	= 1,

You can combine these two to NFPROTO_UNSPEC, like you did to v0,
is not it?

>index b2c90bb..db01021 100644
>--- a/extensions/libxt_NFQUEUE.man
>+++ b/extensions/libxt_NFQUEUE.man
>@@ -5,8 +5,18 @@ number.
> \fB\-\-queue\-num\fP \fIvalue\fP
> This specifies the QUEUE number to use. Valid queue numbers are 0 to 65535. The default value is 0.
> .PP
>+.TP
>+\fB\-\-queue\-balance\fP \fIvalue\fP:\fIvalue\fP

Minor nitpick:  "\fIvalue\fP\fB:\fP\fIvale\fP", because the : is to
be typed verbatim.

>+This specifies a range of queues to use. Packets are then balanced across the given queues.
>+This is useful for multicore systems: start multiple instances of the userspace program on
>+queues x, x+1, .. x+n and use "--queue-balance x:x+n".

\-\-queue\-balance here, too.

>+Packets belonging to the same connection are put into the same nfqueue.
>+.PP
> It can only be used with Kernel versions 2.6.14 or later, since it requires
> the
> .B
> nfnetlink_queue
> kernel support.
>+.B
>+queue-balance
>+support was added in Linux 2.6.31.

Mh, try to use \fB...\fP, as that's somehow easier to recognize given
people are familiar with HTML.

(\fBqueue-balance\fP support wa added in Linux 2.6.31.)


Looks good generally.

Re: [PATCH 2/2] NFQUEUE: add new v1 version with queue-balance option

[Florian Westphal]

Jan Engelhardt <jengelh@medozas.de> wrote:
> On Friday 2009-06-05 03:17, Florian Westphal wrote:
> >target4_v1{
> >+	.revision	= 1,
> >+	.name		= "NFQUEUE",
> >+	.version	= XTABLES_VERSION,
> >+	.size		= XT_ALIGN(sizeof(struct xt_NFQ_info_v1)),
> >+	.userspacesize	= XT_ALIGN(sizeof(struct xt_NFQ_info_v1)),
> >+	.help		= NFQUEUE_help_v1,
> >+	.init		= NFQUEUE_init_v1,
> >+	.parse		= NFQUEUE_parse_v1,
> >+	.print		= NFQUEUE_print_v1,
> >+	.save		= NFQUEUE_save_v1,
> >+	.extra_opts	= NFQUEUE_opts_v1,
> >+};
> >+
> >+static struct xtables_target nfqueue_target6_v1 = {
> >+	.family		= NFPROTO_IPV6,
> >+	.revision	= 1,
> 
> You can combine these two to NFPROTO_UNSPEC, like you did to v0,
> is not it?

Hm, v1 is tied to ipv4/6 on the kernel side, thats why I wanted to make
this explicit here, too.

> >--- a/extensions/libxt_NFQUEUE.man
> >+++ b/extensions/libxt_NFQUEUE.man
> >@@ -5,8 +5,18 @@ number.
> > \fB\-\-queue\-num\fP \fIvalue\fP
> > This specifies the QUEUE number to use. Valid queue numbers are 0 to 65535. The default value is 0.
> > .PP
> >+.TP
> >+\fB\-\-queue\-balance\fP \fIvalue\fP:\fIvalue\fP
> 
> Minor nitpick:  "\fIvalue\fP\fB:\fP\fIvale\fP", because the : is to
> be typed verbatim.

right, will fix.

> >+This specifies a range of queues to use. Packets are then balanced across the given queues.
> >+This is useful for multicore systems: start multiple instances of the userspace program on
> >+queues x, x+1, .. x+n and use "--queue-balance x:x+n".
> 
> \-\-queue\-balance here, too.

Doh -- I should have thought of that when typing this 8-/

> >+Packets belonging to the same connection are put into the same nfqueue.
> >+.PP
> > It can only be used with Kernel versions 2.6.14 or later, since it requires
> > the
> > .B
> > nfnetlink_queue
> > kernel support.
> >+.B
> >+queue-balance
> >+support was added in Linux 2.6.31.
> 
> Mh, try to use \fB...\fP, as that's somehow easier to recognize given
> people are familiar with HTML.
> 
> (\fBqueue-balance\fP support wa added in Linux 2.6.31.)

Agreed.
I will re-spin with these fixups after the 2.6.30 iptables release is out.

Thank you for reviewing this patch.

Re: [PATCH 2/2] NFQUEUE: add new v1 version with queue-balance option

[Jan Engelhardt]

On Friday 2009-06-05 15:27, Florian Westphal wrote:
>> >target4_v1{
>> >+	.revision	= 1,
>> >+	.name		= "NFQUEUE",
>> >+	.version	= XTABLES_VERSION,
>> >+	.size		= XT_ALIGN(sizeof(struct xt_NFQ_info_v1)),
>> >+	.userspacesize	= XT_ALIGN(sizeof(struct xt_NFQ_info_v1)),
>> >+	.help		= NFQUEUE_help_v1,
>> >+	.init		= NFQUEUE_init_v1,
>> >+	.parse		= NFQUEUE_parse_v1,
>> >+	.print		= NFQUEUE_print_v1,
>> >+	.save		= NFQUEUE_save_v1,
>> >+	.extra_opts	= NFQUEUE_opts_v1,
>> >+};
>> >+
>> >+static struct xtables_target nfqueue_target6_v1 = {
>> >+	.family		= NFPROTO_IPV6,
>> >+	.revision	= 1,
>> 
>> You can combine these two to NFPROTO_UNSPEC, like you did to v0,
>> is not it?
>
>Hm, v1 is tied to ipv4/6 on the kernel side, thats why I wanted to make
>this explicit here, too.

It is not needed. iptables's .family and the kernel's .family fields
are not linked in any way. That is why libxt_length.c can have
NFPROTO_UNSPEC even if the kernel's xt_length.c happens not to
(because it inspects L3proto-dependent headers).

Re: [PATCH 2/2] NFQUEUE: add new v1 version with queue-balance option

[Florian Westphal]

Jan Engelhardt <jengelh@medozas.de> wrote:
> >Hm, v1 is tied to ipv4/6 on the kernel side, thats why I wanted to make
> >this explicit here, too.
> 
> It is not needed. iptables's .family and the kernel's .family fields
> are not linked in any way. That is why libxt_length.c can have
> NFPROTO_UNSPEC even if the kernel's xt_length.c happens not to
> (because it inspects L3proto-dependent headers).

I didn't know this was common practice. Thanks for explaining, I will use
_UNSPEC as you suggested.

Re: NFQUEUE balancing extension (userspace changes)

[Patrick McHardy]

Florian Westphal wrote:
> Hello list,
> 
> this adds a new revision of the NFQUEUE target.
> 
> In particular, a new "--queue-balance" option, which allows to
> specify a range of queues to use.
> Packets are then balanced across the given queues by the kernel.
> 
> This is useful for multicore systems:
> start multiple instances of the userspace program on queues
> x, x+1, .. x+n and use "--queue-balance x:x+n".
> 
> As for naming, if you'd prefer a different name (or just extend
> the existing --queue-num option to accept "firstid:lastid" instead),
> please let me know and I'll be happy to change it.
> 
> Both paches can also be pulled from
> git://git.breakpoint.cc/fw/iptables.git nfq-balance

Thanks, I'll wait with pulling these until we've made the iptables
release for 2.6.30.