Skip to contentSkip to conversation

conntrack: restrict runtime hashsize modifications

PATCH Applied Alexey Dobriyan 3 participants 28 messages Feb 3, 2010 – Feb 8, 2010

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 conntrack: restrict runtime hashsize modifications

Jon Masters correctly points out that conntrack hash sizes (nf_conntrack_htable_size) are global (not per-netns) and modifiable at runtime via /sys/module/nf_conntrack/hashsize .

Steps to reproduce:

	clone(CLONE_NEWNET)
	[grow /sys/module/nf_conntrack/hashsize]
	exit()

At netns exit we are going to scan random memory for conntracks to be killed.

Apparently there is a code which deals with hashtable resize for init_net (and it was there befode netns conntrack code), so prohibit hashsize modification if there is more than one netns exists.

To change hashtable sizes, you need to reload module.

Expectation hashtable size was simply glued to a variable with no code to rehash expectations, so it was a bug to allow writing to it. Make "expect_hashsize" readonly.

This is temporarily until we figure out what to do.

---

 net/netfilter/nf_conntrack_core.c   |   15 +++++++++++++++
 net/netfilter/nf_conntrack_expect.c |    2 +-
 2 files changed, 16 insertions(+), 1 deletion(-)
net/netfilter/nf_conntrack_core.c +15 -0
@@ -21,6 +21,7 @@
 #include <linux/stddef.h>
 #include <linux/slab.h>
 #include <linux/random.h>
+#include <linux/rtnetlink.h>
 #include <linux/jhash.h>
 #include <linux/err.h>
 #include <linux/percpu.h>
@@ -1198,6 +1199,20 @@ int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp)
 	if (!nf_conntrack_htable_size)
 		return param_set_uint(val, kp);
 
+	{
+		struct net *net;
+		unsigned int nr;
+
+		nr = 0;
+		rtnl_lock();
+		for_each_net(net)
+			nr++;
+		rtnl_unlock();
+		/* init_net always exists */
+		if (nr != 1)
+			return -EINVAL;
+	}
+
 	hashsize = simple_strtoul(val, NULL, 0);
 	if (!hashsize)
 		return -EINVAL;
net/netfilter/nf_conntrack_expect.c +1 -1
@@ -569,7 +569,7 @@ static void exp_proc_remove(struct net *net)
 #endif /* CONFIG_PROC_FS */
 }
 
-module_param_named(expect_hashsize, nf_ct_expect_hsize, uint, 0600);
+module_param_named(expect_hashsize, nf_ct_expect_hsize, uint, 0400);
 
 int nf_conntrack_expect_init(struct net *net)
 {
#1 lore raw
Quoted 3 lines · Alexey Dobriyan

Jon Masters correctly points out that conntrack hash sizes (nf_conntrack_htable_size) are global (not per-netns) and modifiable at runtime via /sys/module/nf_conntrack/hashsize .

Thanks.

Jon.

#2 lore raw
In reply to Alexey Dobriyan
Quoted 15 lines · Alexey Dobriyan

Jon Masters correctly points out that conntrack hash sizes (nf_conntrack_htable_size) are global (not per-netns) and modifiable at runtime via /sys/module/nf_conntrack/hashsize .

Steps to reproduce:

	clone(CLONE_NEWNET)
	[grow /sys/module/nf_conntrack/hashsize]
	exit()

At netns exit we are going to scan random memory for conntracks to be killed.

Apparently there is a code which deals with hashtable resize for init_net (and it was there befode netns conntrack code), so prohibit hashsize modification if there is more than one netns exists.

Quoted 7 lines

To change hashtable sizes, you need to reload module.

Expectation hashtable size was simply glued to a variable with no code to rehash expectations, so it was a bug to allow writing to it. Make "expect_hashsize" readonly.

This is temporarily until we figure out what to do.

How about alternatively moving nf_conntrack_hsize into the per-namespace struct? It doesn't look more complicated or intrusive and would allow to still change the init_net hashsize. Also seems less hackish :)

#3 lore raw
Quoted 22 lines · Patrick McHardy
Quoted 21 lines · Alexey Dobriyan

Jon Masters correctly points out that conntrack hash sizes (nf_conntrack_htable_size) are global (not per-netns) and modifiable at runtime via /sys/module/nf_conntrack/hashsize .

Steps to reproduce:

	clone(CLONE_NEWNET)
	[grow /sys/module/nf_conntrack/hashsize]
	exit()

At netns exit we are going to scan random memory for conntracks to be killed.

Apparently there is a code which deals with hashtable resize for init_net (and it was there befode netns conntrack code), so prohibit hashsize modification if there is more than one netns exists.

To change hashtable sizes, you need to reload module.

Expectation hashtable size was simply glued to a variable with no code to rehash expectations, so it was a bug to allow writing to it. Make "expect_hashsize" readonly.

Quoted 6 lines
Quoted 1 line

This is temporarily until we figure out what to do.

How about alternatively moving nf_conntrack_hsize into the per-namespace struct? It doesn't look more complicated or intrusive and would allow to still change the init_net hashsize. Also seems less hackish :)

Just to avoid duplicate work, I'm currently trying that.

#4 lore raw
Quoted 23 lines · Patrick McHardy
Quoted 22 lines · Patrick McHardy
Quoted 21 lines · Alexey Dobriyan

Jon Masters correctly points out that conntrack hash sizes (nf_conntrack_htable_size) are global (not per-netns) and modifiable at runtime via /sys/module/nf_conntrack/hashsize .

Steps to reproduce:

	clone(CLONE_NEWNET)
	[grow /sys/module/nf_conntrack/hashsize]
	exit()

At netns exit we are going to scan random memory for conntracks to be killed.

Apparently there is a code which deals with hashtable resize for init_net (and it was there befode netns conntrack code), so prohibit hashsize modification if there is more than one netns exists.

To change hashtable sizes, you need to reload module.

Expectation hashtable size was simply glued to a variable with no code to rehash expectations, so it was a bug to allow writing to it. Make "expect_hashsize" readonly.

Quoted 8 lines
Quoted 6 lines
Quoted 1 line

This is temporarily until we figure out what to do.

How about alternatively moving nf_conntrack_hsize into the per-namespace struct? It doesn't look more complicated or intrusive and would allow to still change the init_net hashsize. Also seems less hackish :)

Just to avoid duplicate work, I'm currently trying that.

Bah. I already worked a set of patches to do that as I mentioned, but you've probably done it by now - can clean up and post if not :)

Jon.

#5 lore raw
Quoted 25 lines · Jon Masters
Quoted 24 lines · Patrick McHardy
Quoted 23 lines · Patrick McHardy
Quoted 22 lines · Alexey Dobriyan

Jon Masters correctly points out that conntrack hash sizes (nf_conntrack_htable_size) are global (not per-netns) and modifiable at runtime via /sys/module/nf_conntrack/hashsize .

Steps to reproduce:

	clone(CLONE_NEWNET)
	[grow /sys/module/nf_conntrack/hashsize]
	exit()

At netns exit we are going to scan random memory for conntracks to be killed.

Apparently there is a code which deals with hashtable resize for init_net (and it was there befode netns conntrack code), so prohibit hashsize modification if there is more than one netns exists.

To change hashtable sizes, you need to reload module.

Expectation hashtable size was simply glued to a variable with no code to rehash expectations, so it was a bug to allow writing to it. Make "expect_hashsize" readonly.

This is temporarily until we figure out what to do.

Quoted 8 lines
Quoted 5 lines
Quoted 4 lines

How about alternatively moving nf_conntrack_hsize into the per-namespace struct? It doesn't look more complicated or intrusive and would allow to still change the init_net hashsize. Also seems less hackish :)

Just to avoid duplicate work, I'm currently trying that.

Bah. I already worked a set of patches to do that as I mentioned, but you've probably done it by now - can clean up and post if not :)

Sorry, I missed that in your mail. I'm pretty much done, will finish testing shortly.

#6 lore raw
Quoted 30 lines · Patrick McHardy
Quoted 29 lines · Jon Masters
Quoted 28 lines · Patrick McHardy
Quoted 27 lines · Patrick McHardy
Quoted 15 lines · Alexey Dobriyan

Jon Masters correctly points out that conntrack hash sizes (nf_conntrack_htable_size) are global (not per-netns) and modifiable at runtime via /sys/module/nf_conntrack/hashsize .

Steps to reproduce:

	clone(CLONE_NEWNET)
	[grow /sys/module/nf_conntrack/hashsize]
	exit()

At netns exit we are going to scan random memory for conntracks to be killed.

Apparently there is a code which deals with hashtable resize for init_net (and it was there befode netns conntrack code), so prohibit hashsize modification if there is more than one netns exists.

Quoted 7 lines

To change hashtable sizes, you need to reload module.

Expectation hashtable size was simply glued to a variable with no code to rehash expectations, so it was a bug to allow writing to it. Make "expect_hashsize" readonly.

This is temporarily until we figure out what to do.

How about alternatively moving nf_conntrack_hsize into the per-namespace struct? It doesn't look more complicated or intrusive and would allow to still change the init_net hashsize. Also seems less hackish :)

Quoted 7 lines
Quoted 4 lines
Quoted 1 line

Just to avoid duplicate work, I'm currently trying that.

Bah. I already worked a set of patches to do that as I mentioned, but you've probably done it by now - can clean up and post if not :)

Sorry, I missed that in your mail. I'm pretty much done, will finish testing shortly.

Oh, it's cool. I hacked it together on my test box but I'm happy to go with whatever you post later, I will just try to be forthcoming next time with my bits first to save you hassle. Please do keep CCing me on these things and I'll try to test over the weekend as time permits.

Jon.

#7 lore raw
Quoted 11 lines · Jon Masters
Quoted 5 lines · Patrick McHardy
Quoted 3 lines
Quoted 1 line

Just to avoid duplicate work, I'm currently trying that.

Bah. I already worked a set of patches to do that as I mentioned, but you've probably done it by now - can clean up and post if not :)

Sorry, I missed that in your mail. I'm pretty much done, will finish testing shortly.

Oh, it's cool. I hacked it together on my test box but I'm happy to go with whatever you post later, I will just try to be forthcoming next time with my bits first to save you hassle. Please do keep CCing me on these things and I'll try to test over the weekend as time permits.

Will do, thanks for all your help Jon.

#8 lore raw
In reply to Patrick McHardy
Quoted 22 lines · Patrick McHardy
Quoted 21 lines · Alexey Dobriyan

Jon Masters correctly points out that conntrack hash sizes (nf_conntrack_htable_size) are global (not per-netns) and modifiable at runtime via /sys/module/nf_conntrack/hashsize .

Steps to reproduce:

	clone(CLONE_NEWNET)
	[grow /sys/module/nf_conntrack/hashsize]
	exit()

At netns exit we are going to scan random memory for conntracks to be killed.

Apparently there is a code which deals with hashtable resize for init_net (and it was there befode netns conntrack code), so prohibit hashsize modification if there is more than one netns exists.

To change hashtable sizes, you need to reload module.

Expectation hashtable size was simply glued to a variable with no code to rehash expectations, so it was a bug to allow writing to it. Make "expect_hashsize" readonly.

Quoted 6 lines
Quoted 1 line

This is temporarily until we figure out what to do.

How about alternatively moving nf_conntrack_hsize into the per-namespace struct? It doesn't look more complicated or intrusive and would allow to still change the init_net hashsize. Also seems less hackish :)

How about this (so far untested) patch? The htable_size is moved into the per-namespace struct and initialized from the current (global) value of nf_conntrack_htable_size. Changes through sysfs are still permitted, but only affect the init namespace and newly created ones.

Additionally I removed reinitializing the hash random value when changing the hash size since that also requires to rehash in all namespaces.

include/net/netns/conntrack.h +1 -0
index aed23b6..63d4498 100644
@@ -11,6 +11,7 @@ struct nf_conntrack_ecache;
 struct netns_ct {
 	atomic_t		count;
 	unsigned int		expect_count;
+	unsigned int		htable_size;
 	struct kmem_cache	*nf_conntrack_cachep;
 	struct hlist_nulls_head	*hash;
 	struct hlist_head	*expect_hash;
include/net/netns/ipv4.h +1 -0
index 2eb3814..9a4b8b7 100644
@@ -40,6 +40,7 @@ struct netns_ipv4 {
 	struct xt_table		*iptable_security;
 	struct xt_table		*nat_table;
 	struct hlist_head	*nat_bysource;
+	unsigned int		nat_htable_size;
 	int			nat_vmalloced;
 #endif
 
net/ipv4/netfilter/nf_conntrack_l3proto_ipv4.c +1 -1
index d171b12..d1ea38a 100644
@@ -210,7 +210,7 @@ static ctl_table ip_ct_sysctl_table[] = {
 	},
 	{
 		.procname	= "ip_conntrack_buckets",
-		.data		= &nf_conntrack_htable_size,
+		.data		= &init_net.ct.htable_size,
 		.maxlen		= sizeof(unsigned int),
 		.mode		= 0444,
 		.proc_handler	= proc_dointvec,
net/ipv4/netfilter/nf_conntrack_l3proto_ipv4_compat.c +2 -2
index 8668a3d..2fb7b76 100644
@@ -32,7 +32,7 @@ static struct hlist_nulls_node *ct_get_first(struct seq_file *seq)
 	struct hlist_nulls_node *n;
 
 	for (st->bucket = 0;
-	     st->bucket < nf_conntrack_htable_size;
+	     st->bucket < net->ct.htable_size;
 	     st->bucket++) {
 		n = rcu_dereference(net->ct.hash[st->bucket].first);
 		if (!is_a_nulls(n))
@@ -50,7 +50,7 @@ static struct hlist_nulls_node *ct_get_next(struct seq_file *seq,
 	head = rcu_dereference(head->next);
 	while (is_a_nulls(head)) {
 		if (likely(get_nulls_value(head) == st->bucket)) {
-			if (++st->bucket >= nf_conntrack_htable_size)
+			if (++st->bucket >= net->ct.htable_size)
 				return NULL;
 		}
 		head = rcu_dereference(net->ct.hash[st->bucket].first);
net/ipv4/netfilter/nf_nat_core.c +10 -13
index fe1a644..5cff943 100644
@@ -35,9 +35,6 @@ static DEFINE_SPINLOCK(nf_nat_lock);
 
 static struct nf_conntrack_l3proto *l3proto __read_mostly;
 
-/* Calculated at init based on memory size */
-static unsigned int nf_nat_htable_size __read_mostly;
-
 #define MAX_IP_NAT_PROTO 256
 static const struct nf_nat_protocol *nf_nat_protos[MAX_IP_NAT_PROTO]
 						__read_mostly;
@@ -72,7 +69,7 @@ EXPORT_SYMBOL_GPL(nf_nat_proto_put);
 
 /* We keep an extra hash for each conntrack, for fast searching. */
 static inline unsigned int
-hash_by_src(const struct nf_conntrack_tuple *tuple)
+hash_by_src(const struct net *net, const struct nf_conntrack_tuple *tuple)
 {
 	unsigned int hash;
 
@@ -80,7 +77,7 @@ hash_by_src(const struct nf_conntrack_tuple *tuple)
 	hash = jhash_3words((__force u32)tuple->src.u3.ip,
 			    (__force u32)tuple->src.u.all,
 			    tuple->dst.protonum, 0);
-	return ((u64)hash * nf_nat_htable_size) >> 32;
+	return ((u64)hash * net->ipv4.nat_htable_size) >> 32;
 }
 
 /* Is this tuple already taken? (not by us) */
@@ -147,7 +144,7 @@ find_appropriate_src(struct net *net,
 		     struct nf_conntrack_tuple *result,
 		     const struct nf_nat_range *range)
 {
-	unsigned int h = hash_by_src(tuple);
+	unsigned int h = hash_by_src(net, tuple);
 	const struct nf_conn_nat *nat;
 	const struct nf_conn *ct;
 	const struct hlist_node *n;
@@ -330,7 +327,7 @@ nf_nat_setup_info(struct nf_conn *ct,
 	if (have_to_hash) {
 		unsigned int srchash;
 
-		srchash = hash_by_src(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
+		srchash = hash_by_src(net, &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 		spin_lock_bh(&nf_nat_lock);
 		/* nf_conntrack_alter_reply might re-allocate exntension aera */
 		nat = nfct_nat(ct);
@@ -679,8 +676,11 @@ nfnetlink_parse_nat_setup(struct nf_conn *ct,
 
 static int __net_init nf_nat_net_init(struct net *net)
 {
-	net->ipv4.nat_bysource = nf_ct_alloc_hashtable(&nf_nat_htable_size,
-						      &net->ipv4.nat_vmalloced, 0);
+	/* Leave them the same for the moment. */
+	net->ipv4.nat_htable_size = net->ct.htable_size;
+
+	net->ipv4.nat_bysource = nf_ct_alloc_hashtable(&net->ipv4.nat_htable_size,
+						       &net->ipv4.nat_vmalloced, 0);
 	if (!net->ipv4.nat_bysource)
 		return -ENOMEM;
 	return 0;
@@ -703,7 +703,7 @@ static void __net_exit nf_nat_net_exit(struct net *net)
 	nf_ct_iterate_cleanup(net, &clean_nat, NULL);
 	synchronize_rcu();
 	nf_ct_free_hashtable(net->ipv4.nat_bysource, net->ipv4.nat_vmalloced,
-			     nf_nat_htable_size);
+			     net->ipv4.nat_htable_size);
 }
 
 static struct pernet_operations nf_nat_net_ops = {
@@ -724,9 +724,6 @@ static int __init nf_nat_init(void)
 		return ret;
 	}
 
-	/* Leave them the same for the moment. */
-	nf_nat_htable_size = nf_conntrack_htable_size;
-
 	ret = register_pernet_subsys(&nf_nat_net_ops);
 	if (ret < 0)
 		goto cleanup_extend;
net/netfilter/nf_conntrack_core.c +24 -25

154 lines not on this page · raw message

net/netfilter/nf_conntrack_expect.c +1 -1
index fdf5d2a..1e751b1 100644
@@ -577,7 +577,7 @@ int nf_conntrack_expect_init(struct net *net)
 
 	if (net_eq(net, &init_net)) {
 		if (!nf_ct_expect_hsize) {
-			nf_ct_expect_hsize = nf_conntrack_htable_size / 256;
+			nf_ct_expect_hsize = net->ct.htable_size / 256;
 			if (!nf_ct_expect_hsize)
 				nf_ct_expect_hsize = 1;
 		}
net/netfilter/nf_conntrack_helper.c +1 -1
index 65c2a7b..4b1a56b 100644
@@ -192,7 +192,7 @@ static void __nf_conntrack_helper_unregister(struct nf_conntrack_helper *me,
 	/* Get rid of expecteds, set helpers to NULL. */
 	hlist_nulls_for_each_entry(h, nn, &net->ct.unconfirmed, hnnode)
 		unhelp(h, me);
-	for (i = 0; i < nf_conntrack_htable_size; i++) {
+	for (i = 0; i < net->ct.htable_size; i++) {
 		hlist_nulls_for_each_entry(h, nn, &net->ct.hash[i], hnnode)
 			unhelp(h, me);
 	}
net/netfilter/nf_conntrack_netlink.c +1 -1
index 42f21c0..0ffe689 100644
@@ -594,7 +594,7 @@ ctnetlink_dump_table(struct sk_buff *skb, struct netlink_callback *cb)
 
 	rcu_read_lock();
 	last = (struct nf_conn *)cb->args[1];
-	for (; cb->args[0] < nf_conntrack_htable_size; cb->args[0]++) {
+	for (; cb->args[0] < init_net.ct.htable_size; cb->args[0]++) {
 restart:
 		hlist_nulls_for_each_entry_rcu(h, n, &init_net.ct.hash[cb->args[0]],
 					 hnnode) {
net/netfilter/nf_conntrack_standalone.c +4 -3
index 028aba6..e310f15 100644
@@ -51,7 +51,7 @@ static struct hlist_nulls_node *ct_get_first(struct seq_file *seq)
 	struct hlist_nulls_node *n;
 
 	for (st->bucket = 0;
-	     st->bucket < nf_conntrack_htable_size;
+	     st->bucket < net->ct.htable_size;
 	     st->bucket++) {
 		n = rcu_dereference(net->ct.hash[st->bucket].first);
 		if (!is_a_nulls(n))
@@ -69,7 +69,7 @@ static struct hlist_nulls_node *ct_get_next(struct seq_file *seq,
 	head = rcu_dereference(head->next);
 	while (is_a_nulls(head)) {
 		if (likely(get_nulls_value(head) == st->bucket)) {
-			if (++st->bucket >= nf_conntrack_htable_size)
+			if (++st->bucket >= net->ct.htable_size)
 				return NULL;
 		}
 		head = rcu_dereference(net->ct.hash[st->bucket].first);
@@ -355,7 +355,7 @@ static ctl_table nf_ct_sysctl_table[] = {
 	},
 	{
 		.procname       = "nf_conntrack_buckets",
-		.data           = &nf_conntrack_htable_size,
+		.data           = &init_net.ct.htable_size,
 		.maxlen         = sizeof(unsigned int),
 		.mode           = 0444,
 		.proc_handler   = proc_dointvec,
@@ -421,6 +421,7 @@ static int nf_conntrack_standalone_init_sysctl(struct net *net)
 		goto out_kmemdup;
 
 	table[1].data = &net->ct.count;
+	table[2].data = &net->ct.htable_size;
 	table[3].data = &net->ct.sysctl_checksum;
 	table[4].data = &net->ct.sysctl_log_invalid;
 
#9 lore raw
Quoted 30 lines · Patrick McHardy
Quoted 22 lines · Patrick McHardy
Quoted 21 lines · Alexey Dobriyan

Jon Masters correctly points out that conntrack hash sizes (nf_conntrack_htable_size) are global (not per-netns) and modifiable at runtime via /sys/module/nf_conntrack/hashsize .

Steps to reproduce:

	clone(CLONE_NEWNET)
	[grow /sys/module/nf_conntrack/hashsize]
	exit()

At netns exit we are going to scan random memory for conntracks to be killed.

Apparently there is a code which deals with hashtable resize for init_net (and it was there befode netns conntrack code), so prohibit hashsize modification if there is more than one netns exists.

To change hashtable sizes, you need to reload module.

Expectation hashtable size was simply glued to a variable with no code to rehash expectations, so it was a bug to allow writing to it. Make "expect_hashsize" readonly.

Quoted 6 lines
Quoted 1 line

This is temporarily until we figure out what to do.

How about alternatively moving nf_conntrack_hsize into the per-namespace struct? It doesn't look more complicated or intrusive and would allow to still change the init_net hashsize. Also seems less hackish :)

Quoted 4 lines

How about this (so far untested) patch? The htable_size is moved into the per-namespace struct and initialized from the current (global) value of nf_conntrack_htable_size. Changes through sysfs are still permitted, but only affect the init namespace and newly created ones.

No matter what we do, it's a hack!

Quoted 3 lines

Additionally I removed reinitializing the hash random value when changing the hash size since that also requires to rehash in all namespaces.

I'm not fond of this, because we're not even closely going to allow changing hashtable size per-netns. As such having actual per-netns hashtable size just slows down everything.

#10 lore raw
Quoted 11 lines · Alexey Dobriyan
Quoted 6 lines · Patrick McHardy
Quoted 1 line · Patrick McHardy

Alexey Dobriyan wrote:

Additionally I removed reinitializing the hash random value when changing the hash size since that also requires to rehash in all namespaces.

I'm not fond of this, because we're not even closely going to allow changing hashtable size per-netns. As such having actual per-netns hashtable size just slows down everything.

Are you sure, as compared with a shared global size that this is really going to slow things down all that much? We already need to poke in that struct to pull out various things just hash. I haven't profiled to see.

Jon.

#11 lore raw
In reply to Alexey Dobriyan
Quoted 36 lines · Alexey Dobriyan
Quoted 24 lines · Patrick McHardy
Quoted 23 lines · Patrick McHardy
Quoted 22 lines · Alexey Dobriyan

Jon Masters correctly points out that conntrack hash sizes (nf_conntrack_htable_size) are global (not per-netns) and modifiable at runtime via /sys/module/nf_conntrack/hashsize .

Steps to reproduce:

	clone(CLONE_NEWNET)
	[grow /sys/module/nf_conntrack/hashsize]
	exit()

At netns exit we are going to scan random memory for conntracks to be killed.

Apparently there is a code which deals with hashtable resize for init_net (and it was there befode netns conntrack code), so prohibit hashsize modification if there is more than one netns exists.

To change hashtable sizes, you need to reload module.

Expectation hashtable size was simply glued to a variable with no code to rehash expectations, so it was a bug to allow writing to it. Make "expect_hashsize" readonly.

This is temporarily until we figure out what to do.

Quoted 8 lines
Quoted 4 lines

How about alternatively moving nf_conntrack_hsize into the per-namespace struct? It doesn't look more complicated or intrusive and would allow to still change the init_net hashsize. Also seems less hackish :)

How about this (so far untested) patch? The htable_size is moved into the per-namespace struct and initialized from the current (global) value of nf_conntrack_htable_size. Changes through sysfs are still permitted, but only affect the init namespace and newly created ones.

No matter what we do, it's a hack!

Quoted 7 lines
Quoted 3 lines

Additionally I removed reinitializing the hash random value when changing the hash size since that also requires to rehash in all namespaces.

I'm not fond of this, because we're not even closely going to allow changing hashtable size per-netns. As such having actual per-netns hashtable size just slows down everything.

Actually it doesn't seem like much more work to allow changing table size, the main problem is that sysfs module parameters don't seem to fit into the network namespace model at all.

Please be more specific about your suspected slowdowns. What's "everything"? What's different about the hashsize compared to the many members we already moved to per-netns structs?

#12 lore raw
Quoted 41 lines · Patrick McHardy
Quoted 40 lines · Alexey Dobriyan
Quoted 24 lines · Patrick McHardy
Quoted 23 lines · Patrick McHardy
Quoted 22 lines · Alexey Dobriyan

Jon Masters correctly points out that conntrack hash sizes (nf_conntrack_htable_size) are global (not per-netns) and modifiable at runtime via /sys/module/nf_conntrack/hashsize .

Steps to reproduce:

	clone(CLONE_NEWNET)
	[grow /sys/module/nf_conntrack/hashsize]
	exit()

At netns exit we are going to scan random memory for conntracks to be killed.

Apparently there is a code which deals with hashtable resize for init_net (and it was there befode netns conntrack code), so prohibit hashsize modification if there is more than one netns exists.

To change hashtable sizes, you need to reload module.

Expectation hashtable size was simply glued to a variable with no code to rehash expectations, so it was a bug to allow writing to it. Make "expect_hashsize" readonly.

This is temporarily until we figure out what to do.

Quoted 8 lines
Quoted 4 lines

How about alternatively moving nf_conntrack_hsize into the per-namespace struct? It doesn't look more complicated or intrusive and would allow to still change the init_net hashsize. Also seems less hackish :)

How about this (so far untested) patch? The htable_size is moved into the per-namespace struct and initialized from the current (global) value of nf_conntrack_htable_size. Changes through sysfs are still permitted, but only affect the init namespace and newly created ones.

No matter what we do, it's a hack!

Quoted 3 lines

Additionally I removed reinitializing the hash random value when changing the hash size since that also requires to rehash in all namespaces.

Quoted 7 lines
Quoted 3 lines

I'm not fond of this, because we're not even closely going to allow changing hashtable size per-netns. As such having actual per-netns hashtable size just slows down everything.

Actually it doesn't seem like much more work to allow changing table size, the main problem is that sysfs module parameters don't seem to fit into the network namespace model at all.

That was the reason I initially suggested we need a better way to expose netns topology through sysfs, which I still think is a good idea. How about this...it's dangerous as it is right now to leave things global. I suggest leaving the existing sysfs module parameter that only actually touches the init_net ct and get the rest fixed up, then adding support for exposing the topology better in sysfs and tweaking per-ns bits.

But maybe you want to fix it all at the same time.

Jon.

#13 lore raw
Quoted 17 lines · Jon Masters
Quoted 9 lines · Patrick McHardy
Quoted 9 lines
Quoted 4 lines

How about this (so far untested) patch? The htable_size is moved into the per-namespace struct and initialized from the current (global) value of nf_conntrack_htable_size. Changes through sysfs are still permitted, but only affect the init namespace and newly created ones.

No matter what we do, it's a hack!

Quoted 3 lines

Additionally I removed reinitializing the hash random value when changing the hash size since that also requires to rehash in all namespaces.

Quoted 6 lines
Quoted 3 lines

I'm not fond of this, because we're not even closely going to allow changing hashtable size per-netns. As such having actual per-netns hashtable size just slows down everything.

Actually it doesn't seem like much more work to allow changing table size, the main problem is that sysfs module parameters don't seem to fit into the network namespace model at all.

Quoted 8 lines

That was the reason I initially suggested we need a better way to expose netns topology through sysfs, which I still think is a good idea. How about this...it's dangerous as it is right now to leave things global. I suggest leaving the existing sysfs module parameter that only actually touches the init_net ct and get the rest fixed up, then adding support for exposing the topology better in sysfs and tweaking per-ns bits.

But maybe you want to fix it all at the same time.

No, right now I only want to fix the remaining bugs. I'll leave everything else to the netns people.

#14 lore raw
In reply to Patrick McHardy
Quoted 14 lines · Patrick McHardy
Quoted 4 lines · Alexey Dobriyan
Quoted 4 lines

How about this (so far untested) patch? The htable_size is moved into the per-namespace struct and initialized from the current (global) value of nf_conntrack_htable_size. Changes through sysfs are still permitted, but only affect the init namespace and newly created ones.

Quoted 8 lines

No matter what we do, it's a hack!

Quoted 3 lines

Additionally I removed reinitializing the hash random value when changing the hash size since that also requires to rehash in all namespaces.

I'm not fond of this, because we're not even closely going to allow changing hashtable size per-netns. As such having actual per-netns hashtable size just slows down everything.

Quoted 8 lines

Actually it doesn't seem like much more work to allow changing table size, the main problem is that sysfs module parameters don't seem to fit into the network namespace model at all.

Please be more specific about your suspected slowdowns. What's "everything"? What's different about the hashsize compared to the many members we already moved to per-netns structs?

OK testing looks fine, although I'm quite surprised that its actually possible to change module parameters from within non-init namespaces. How is this supposed to work at all? I don't see how sysfs could possibly provide a network namespace context ...

#15 lore raw
Quoted 4 lines · Patrick McHardy

OK testing looks fine, although I'm quite surprised that its actually possible to change module parameters from within non-init namespaces. How is this supposed to work at all? I don't see how sysfs could possibly provide a network namespace context ...

You can do in write hook

    if (!net_eq(current->nsproxy->net_ns, &init_net))
            return -EINVAL;
#16 lore raw
Quoted 11 lines · Alexey Dobriyan
Quoted 4 lines · Patrick McHardy

OK testing looks fine, although I'm quite surprised that its actually possible to change module parameters from within non-init namespaces. How is this supposed to work at all? I don't see how sysfs could possibly provide a network namespace context ...

You can do in write hook

    if (!net_eq(current->nsproxy->net_ns, &init_net))
            return -EINVAL;

Right, I see. So we could actually make resizing work for all namespaces quite easily. Is there any reason not to do this?

#17 lore raw
Quoted 7 lines · Patrick McHardy
Quoted 6 lines · Alexey Dobriyan
Quoted 4 lines · Patrick McHardy

OK testing looks fine, although I'm quite surprised that its actually possible to change module parameters from within non-init namespaces. How is this supposed to work at all? I don't see how sysfs could possibly provide a network namespace context ...

Quoted 8 lines
Quoted 4 lines

You can do in write hook

    if (!net_eq(current->nsproxy->net_ns, &init_net))
            return -EINVAL;

Right, I see. So we could actually make resizing work for all namespaces quite easily. Is there any reason not to do this?

Something like this (untested) patch on top of the previous one.

net/netfilter/nf_conntrack_core.c +11 -9
index ef1c856..212dac3 100644
@@ -31,6 +31,7 @@
 #include <linux/socket.h>
 #include <linux/mm.h>
 #include <linux/rculist_nulls.h>
+#include <linux/nsproxy.h>
 
 #include <net/netfilter/nf_conntrack.h>
 #include <net/netfilter/nf_conntrack_l3proto.h>
@@ -1194,6 +1195,7 @@ int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp)
 	unsigned int hashsize, old_size;
 	struct hlist_nulls_head *hash, *old_hash;
 	struct nf_conntrack_tuple_hash *h;
+	struct net *net = current->nsproxy->net_ns;
 
 	/* On boot, we can set this without any fancy locking. */
 	if (!nf_conntrack_htable_size)
@@ -1213,9 +1215,9 @@ int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp)
 	 * though since that required taking the lock.
 	 */
 	spin_lock_bh(&nf_conntrack_lock);
-	for (i = 0; i < init_net.ct.htable_size; i++) {
-		while (!hlist_nulls_empty(&init_net.ct.hash[i])) {
-			h = hlist_nulls_entry(init_net.ct.hash[i].first,
+	for (i = 0; i < net->ct.htable_size; i++) {
+		while (!hlist_nulls_empty(&net->ct.hash[i])) {
+			h = hlist_nulls_entry(net->ct.hash[i].first,
 					struct nf_conntrack_tuple_hash, hnnode);
 			hlist_nulls_del_rcu(&h->hnnode);
 			bucket = __hash_conntrack(&h->tuple, hashsize,
@@ -1223,13 +1225,13 @@ int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp)
 			hlist_nulls_add_head_rcu(&h->hnnode, &hash[bucket]);
 		}
 	}
-	old_size = init_net.ct.htable_size;
-	old_vmalloced = init_net.ct.hash_vmalloc;
-	old_hash = init_net.ct.hash;
+	old_size = net->ct.htable_size;
+	old_vmalloced = net->ct.hash_vmalloc;
+	old_hash = net->ct.hash;
 
-	init_net.ct.htable_size = nf_conntrack_htable_size = hashsize;
-	init_net.ct.hash_vmalloc = vmalloced;
-	init_net.ct.hash = hash;
+	net->ct.htable_size = nf_conntrack_htable_size = hashsize;
+	net->ct.hash_vmalloc = vmalloced;
+	net->ct.hash = hash;
 	spin_unlock_bh(&nf_conntrack_lock);
 
 	nf_ct_free_hashtable(old_hash, old_vmalloced, old_size);
#18 lore raw
In reply to Patrick McHardy
Quoted 8 lines · Patrick McHardy
Quoted 7 lines · Alexey Dobriyan
Quoted 4 lines · Patrick McHardy

OK testing looks fine, although I'm quite surprised that its actually possible to change module parameters from within non-init namespaces. How is this supposed to work at all? I don't see how sysfs could possibly provide a network namespace context ...

Quoted 7 lines
Quoted 4 lines

You can do in write hook

    if (!net_eq(current->nsproxy->net_ns, &init_net))
            return -EINVAL;

Right, I see. So we could actually make resizing work for all namespaces quite easily. Is there any reason not to do this?

Yes, but I think (2.6.34) there also needs to be a better way to expose netns topology and visualize all the hashtables on a system - if you're the admin you really want to be able to see all of them, not just for a particular namespace you might be in, in addition to the perns views.

We already discussed 2.6.34, I'm only mentioning this "for the record".

Jon.

#19 lore raw
In reply to Alexey Dobriyan
Quoted 11 lines · Alexey Dobriyan
Quoted 4 lines · Patrick McHardy

OK testing looks fine, although I'm quite surprised that its actually possible to change module parameters from within non-init namespaces. How is this supposed to work at all? I don't see how sysfs could possibly provide a network namespace context ...

You can do in write hook

   if (!net_eq(current->nsproxy->net_ns, &init_net)) return -EINVAL;

-EPERM of course.
--
To unsubscribe from this list: send the line "unsubscribe netfilter-devel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html

#20 lore raw
In reply to Patrick McHardy
Quoted 3 lines · Patrick McHardy

Actually it doesn't seem like much more work to allow changing table size, the main problem is that sysfs module parameters don't seem to fit into the network namespace model at all.

Well, they "fit" as they're global because modules are global. So we can make every netns hashtable size equals to module param, or make it bounded by module param
or make initial hashtable size and not bounded
or million other things.

Quoted 1 line

Please be more specific about your suspected slowdowns.

I meant net->ct.htable_size in hash functions _if_ you're not allowing changing it from inside netns.

Quoted 3 lines

What's "everything"? What's different about the hashsize compared to the many members we already moved to per-netns structs?

But whatever. I think per-netns hashtable size shouldn't be done that late.

#21 lore raw
Quoted 10 lines · Alexey Dobriyan
Quoted 3 lines · Patrick McHardy

Actually it doesn't seem like much more work to allow changing table size, the main problem is that sysfs module parameters don't seem to fit into the network namespace model at all.

Well, they "fit" as they're global because modules are global. So we can make every netns hashtable size equals to module param, or make it bounded by module param
or make initial hashtable size and not bounded
or million other things.

Feel free to suggest your preferred method.

Quoted 11 lines
Quoted 1 line

Please be more specific about your suspected slowdowns.

I meant net->ct.htable_size in hash functions _if_ you're not allowing changing it from inside netns.

Quoted 3 lines

What's "everything"? What's different about the hashsize compared to the many members we already moved to per-netns structs?

But whatever. I think per-netns hashtable size shouldn't be done that late.

This is a regression. You can not simply disable resizing and call it a bugfix.

#22 lore raw
Quoted 17 lines · Patrick McHardy
Quoted 9 lines · Alexey Dobriyan
Quoted 3 lines · Patrick McHardy

Actually it doesn't seem like much more work to allow changing table size, the main problem is that sysfs module parameters don't seem to fit into the network namespace model at all.

Well, they "fit" as they're global because modules are global. So we can make every netns hashtable size equals to module param, or make it bounded by module param
or make initial hashtable size and not bounded
or million other things.

Feel free to suggest your preferred method.

Quoted 4 lines
Quoted 1 line

Please be more specific about your suspected slowdowns.

I meant net->ct.htable_size in hash functions _if_ you're not allowing changing it from inside netns.

Quoted 8 lines
Quoted 5 lines
Quoted 3 lines

What's "everything"? What's different about the hashsize compared to the many members we already moved to per-netns structs?

But whatever. I think per-netns hashtable size shouldn't be done that late.

This is a regression. You can not simply disable resizing and call it a bugfix.

This is what I'm going to commit unless someone spots any problems. Resizing of non-init-namespaces is not supported, so its not necessary to decide the questions you raised above at this time.

commit 6521ddb0811ffb4fbfdf48d2c85024bd8f51aadf

Author: Patrick McHardy <[email protected]>
Date:   Mon Feb 8 15:32:18 2010 +0100

netfilter: nf_conntrack: fix hash resizing with namespaces

As noticed by Jon Masters <[email protected]>, the conntrack hash size is global and not per namespace, but modifiable at runtime through /sys/module/nf_conntrack/hashsize. Changing the hash size will only resize the hash in the current namespace however, so other namespaces will use an invalid hash size. This can cause crashes when enlarging the hashsize, or false negative lookups when shrinking it.

Move the hash size into the per-namespace data and only use the global hash size to initialize the per-namespace value when instanciating a new namespace. Additionally restrict hash resizing to init_net for now as other namespaces are not handled currently.

include/net/netns/conntrack.h +1 -0
index aed23b6..63d4498 100644
@@ -11,6 +11,7 @@ struct nf_conntrack_ecache;
 struct netns_ct {
 	atomic_t		count;
 	unsigned int		expect_count;
+	unsigned int		htable_size;
 	struct kmem_cache	*nf_conntrack_cachep;
 	struct hlist_nulls_head	*hash;
 	struct hlist_head	*expect_hash;
include/net/netns/ipv4.h +1 -0
index 2eb3814..9a4b8b7 100644
@@ -40,6 +40,7 @@ struct netns_ipv4 {
 	struct xt_table		*iptable_security;
 	struct xt_table		*nat_table;
 	struct hlist_head	*nat_bysource;
+	unsigned int		nat_htable_size;
 	int			nat_vmalloced;
 #endif
 
net/ipv4/netfilter/nf_conntrack_l3proto_ipv4.c +1 -1
index d171b12..d1ea38a 100644
@@ -210,7 +210,7 @@ static ctl_table ip_ct_sysctl_table[] = {
 	},
 	{
 		.procname	= "ip_conntrack_buckets",
-		.data		= &nf_conntrack_htable_size,
+		.data		= &init_net.ct.htable_size,
 		.maxlen		= sizeof(unsigned int),
 		.mode		= 0444,
 		.proc_handler	= proc_dointvec,
net/ipv4/netfilter/nf_conntrack_l3proto_ipv4_compat.c +2 -2
index 8668a3d..2fb7b76 100644
@@ -32,7 +32,7 @@ static struct hlist_nulls_node *ct_get_first(struct seq_file *seq)
 	struct hlist_nulls_node *n;
 
 	for (st->bucket = 0;
-	     st->bucket < nf_conntrack_htable_size;
+	     st->bucket < net->ct.htable_size;
 	     st->bucket++) {
 		n = rcu_dereference(net->ct.hash[st->bucket].first);
 		if (!is_a_nulls(n))
@@ -50,7 +50,7 @@ static struct hlist_nulls_node *ct_get_next(struct seq_file *seq,
 	head = rcu_dereference(head->next);
 	while (is_a_nulls(head)) {
 		if (likely(get_nulls_value(head) == st->bucket)) {
-			if (++st->bucket >= nf_conntrack_htable_size)
+			if (++st->bucket >= net->ct.htable_size)
 				return NULL;
 		}
 		head = rcu_dereference(net->ct.hash[st->bucket].first);
net/ipv4/netfilter/nf_nat_core.c +9 -13
index fe1a644..26066a2 100644
@@ -35,9 +35,6 @@ static DEFINE_SPINLOCK(nf_nat_lock);
 
 static struct nf_conntrack_l3proto *l3proto __read_mostly;
 
-/* Calculated at init based on memory size */
-static unsigned int nf_nat_htable_size __read_mostly;
-
 #define MAX_IP_NAT_PROTO 256
 static const struct nf_nat_protocol *nf_nat_protos[MAX_IP_NAT_PROTO]
 						__read_mostly;
@@ -72,7 +69,7 @@ EXPORT_SYMBOL_GPL(nf_nat_proto_put);
 
 /* We keep an extra hash for each conntrack, for fast searching. */
 static inline unsigned int
-hash_by_src(const struct nf_conntrack_tuple *tuple)
+hash_by_src(const struct net *net, const struct nf_conntrack_tuple *tuple)
 {
 	unsigned int hash;
 
@@ -80,7 +77,7 @@ hash_by_src(const struct nf_conntrack_tuple *tuple)
 	hash = jhash_3words((__force u32)tuple->src.u3.ip,
 			    (__force u32)tuple->src.u.all,
 			    tuple->dst.protonum, 0);
-	return ((u64)hash * nf_nat_htable_size) >> 32;
+	return ((u64)hash * net->ipv4.nat_htable_size) >> 32;
 }
 
 /* Is this tuple already taken? (not by us) */
@@ -147,7 +144,7 @@ find_appropriate_src(struct net *net,
 		     struct nf_conntrack_tuple *result,
 		     const struct nf_nat_range *range)
 {
-	unsigned int h = hash_by_src(tuple);
+	unsigned int h = hash_by_src(net, tuple);
 	const struct nf_conn_nat *nat;
 	const struct nf_conn *ct;
 	const struct hlist_node *n;
@@ -330,7 +327,7 @@ nf_nat_setup_info(struct nf_conn *ct,
 	if (have_to_hash) {
 		unsigned int srchash;
 
-		srchash = hash_by_src(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
+		srchash = hash_by_src(net, &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 		spin_lock_bh(&nf_nat_lock);
 		/* nf_conntrack_alter_reply might re-allocate exntension aera */
 		nat = nfct_nat(ct);
@@ -679,8 +676,10 @@ nfnetlink_parse_nat_setup(struct nf_conn *ct,
 
 static int __net_init nf_nat_net_init(struct net *net)
 {
-	net->ipv4.nat_bysource = nf_ct_alloc_hashtable(&nf_nat_htable_size,
-						      &net->ipv4.nat_vmalloced, 0);
+	/* Leave them the same for the moment. */
+	net->ipv4.nat_htable_size = net->ct.htable_size;
+	net->ipv4.nat_bysource = nf_ct_alloc_hashtable(&net->ipv4.nat_htable_size,
+						       &net->ipv4.nat_vmalloced, 0);
 	if (!net->ipv4.nat_bysource)
 		return -ENOMEM;
 	return 0;
@@ -703,7 +702,7 @@ static void __net_exit nf_nat_net_exit(struct net *net)
 	nf_ct_iterate_cleanup(net, &clean_nat, NULL);
 	synchronize_rcu();
 	nf_ct_free_hashtable(net->ipv4.nat_bysource, net->ipv4.nat_vmalloced,
-			     nf_nat_htable_size);
+			     net->ipv4.nat_htable_size);
 }
 
 static struct pernet_operations nf_nat_net_ops = {
@@ -724,9 +723,6 @@ static int __init nf_nat_init(void)
 		return ret;
 	}
 
-	/* Leave them the same for the moment. */
-	nf_nat_htable_size = nf_conntrack_htable_size;
-
 	ret = register_pernet_subsys(&nf_nat_net_ops);
 	if (ret < 0)
 		goto cleanup_extend;
net/netfilter/nf_conntrack_core.c +28 -25

167 lines not on this page · raw message

net/netfilter/nf_conntrack_expect.c +1 -1
index 4ad7d1d..2f25ff6 100644
@@ -577,7 +577,7 @@ int nf_conntrack_expect_init(struct net *net)
 
 	if (net_eq(net, &init_net)) {
 		if (!nf_ct_expect_hsize) {
-			nf_ct_expect_hsize = nf_conntrack_htable_size / 256;
+			nf_ct_expect_hsize = net->ct.htable_size / 256;
 			if (!nf_ct_expect_hsize)
 				nf_ct_expect_hsize = 1;
 		}
net/netfilter/nf_conntrack_helper.c +1 -1
index 65c2a7b..4b1a56b 100644
@@ -192,7 +192,7 @@ static void __nf_conntrack_helper_unregister(struct nf_conntrack_helper *me,
 	/* Get rid of expecteds, set helpers to NULL. */
 	hlist_nulls_for_each_entry(h, nn, &net->ct.unconfirmed, hnnode)
 		unhelp(h, me);
-	for (i = 0; i < nf_conntrack_htable_size; i++) {
+	for (i = 0; i < net->ct.htable_size; i++) {
 		hlist_nulls_for_each_entry(h, nn, &net->ct.hash[i], hnnode)
 			unhelp(h, me);
 	}
net/netfilter/nf_conntrack_netlink.c +1 -1
index 42f21c0..0ffe689 100644
@@ -594,7 +594,7 @@ ctnetlink_dump_table(struct sk_buff *skb, struct netlink_callback *cb)
 
 	rcu_read_lock();
 	last = (struct nf_conn *)cb->args[1];
-	for (; cb->args[0] < nf_conntrack_htable_size; cb->args[0]++) {
+	for (; cb->args[0] < init_net.ct.htable_size; cb->args[0]++) {
 restart:
 		hlist_nulls_for_each_entry_rcu(h, n, &init_net.ct.hash[cb->args[0]],
 					 hnnode) {
net/netfilter/nf_conntrack_standalone.c +4 -3
index 028aba6..e310f15 100644
@@ -51,7 +51,7 @@ static struct hlist_nulls_node *ct_get_first(struct seq_file *seq)
 	struct hlist_nulls_node *n;
 
 	for (st->bucket = 0;
-	     st->bucket < nf_conntrack_htable_size;
+	     st->bucket < net->ct.htable_size;
 	     st->bucket++) {
 		n = rcu_dereference(net->ct.hash[st->bucket].first);
 		if (!is_a_nulls(n))
@@ -69,7 +69,7 @@ static struct hlist_nulls_node *ct_get_next(struct seq_file *seq,
 	head = rcu_dereference(head->next);
 	while (is_a_nulls(head)) {
 		if (likely(get_nulls_value(head) == st->bucket)) {
-			if (++st->bucket >= nf_conntrack_htable_size)
+			if (++st->bucket >= net->ct.htable_size)
 				return NULL;
 		}
 		head = rcu_dereference(net->ct.hash[st->bucket].first);
@@ -355,7 +355,7 @@ static ctl_table nf_ct_sysctl_table[] = {
 	},
 	{
 		.procname       = "nf_conntrack_buckets",
-		.data           = &nf_conntrack_htable_size,
+		.data           = &init_net.ct.htable_size,
 		.maxlen         = sizeof(unsigned int),
 		.mode           = 0444,
 		.proc_handler   = proc_dointvec,
@@ -421,6 +421,7 @@ static int nf_conntrack_standalone_init_sysctl(struct net *net)
 		goto out_kmemdup;
 
 	table[1].data = &net->ct.count;
+	table[2].data = &net->ct.htable_size;
 	table[3].data = &net->ct.sysctl_checksum;
 	table[4].data = &net->ct.sysctl_log_invalid;
 
#23 lore raw
In reply to Patrick McHardy
Quoted 30 lines · Patrick McHardy
Quoted 22 lines · Patrick McHardy
Quoted 21 lines · Alexey Dobriyan

Jon Masters correctly points out that conntrack hash sizes (nf_conntrack_htable_size) are global (not per-netns) and modifiable at runtime via /sys/module/nf_conntrack/hashsize .

Steps to reproduce:

	clone(CLONE_NEWNET)
	[grow /sys/module/nf_conntrack/hashsize]
	exit()

At netns exit we are going to scan random memory for conntracks to be killed.

Apparently there is a code which deals with hashtable resize for init_net (and it was there befode netns conntrack code), so prohibit hashsize modification if there is more than one netns exists.

To change hashtable sizes, you need to reload module.

Expectation hashtable size was simply glued to a variable with no code to rehash expectations, so it was a bug to allow writing to it. Make "expect_hashsize" readonly.

Quoted 6 lines
Quoted 1 line

This is temporarily until we figure out what to do.

How about alternatively moving nf_conntrack_hsize into the per-namespace struct? It doesn't look more complicated or intrusive and would allow to still change the init_net hashsize. Also seems less hackish :)

Quoted 4 lines

How about this (so far untested) patch? The htable_size is moved into the per-namespace struct and initialized from the current (global) value of nf_conntrack_htable_size. Changes through sysfs are still permitted, but only affect the init namespace and newly created ones.

I moved the random seed into the per-ns context aswell. I think that's better than having a global one, and you don't need to rehash all.

Jon.

#24 lore raw
Quoted 5 lines · Jon Masters
Quoted 4 lines · Patrick McHardy
Quoted 4 lines

How about alternatively moving nf_conntrack_hsize into the per-namespace struct? It doesn't look more complicated or intrusive and would allow to still change the init_net hashsize. Also seems less hackish :)

Quoted 7 lines
Quoted 4 lines

How about this (so far untested) patch? The htable_size is moved into the per-namespace struct and initialized from the current (global) value of nf_conntrack_htable_size. Changes through sysfs are still permitted, but only affect the init namespace and newly created ones.

I moved the random seed into the per-ns context aswell. I think that's better than having a global one, and you don't need to rehash all.

That's another possibility. But we don't loose anything by not reseeding during resize. It also shouldn't be possible to determine the seed from userspace in a namespace, so there's no real need to use seperate values.

#25 lore raw
Quoted 11 lines · Patrick McHardy
Quoted 10 lines · Jon Masters
Quoted 8 lines · Patrick McHardy
Quoted 4 lines

How about alternatively moving nf_conntrack_hsize into the per-namespace struct? It doesn't look more complicated or intrusive and would allow to still change the init_net hashsize. Also seems less hackish :)

How about this (so far untested) patch? The htable_size is moved into the per-namespace struct and initialized from the current (global) value of nf_conntrack_htable_size. Changes through sysfs are still permitted, but only affect the init namespace and newly created ones.

Quoted 7 lines
Quoted 2 lines

I moved the random seed into the per-ns context aswell. I think that's better than having a global one, and you don't need to rehash all.

That's another possibility. But we don't loose anything by not reseeding during resize. It also shouldn't be possible to determine the seed from userspace in a namespace, so there's no real need to use seperate values.

Right, the risk there is hypothetical at best. But there's little lost in putting it in per-ns and then you can rehash and truly make them independent, which I think is really what netns is all about.

Jon.

#26 lore raw
Quoted 13 lines · Jon Masters
Quoted 12 lines · Patrick McHardy
Quoted 11 lines · Jon Masters
Quoted 8 lines · Patrick McHardy
Quoted 4 lines

How about alternatively moving nf_conntrack_hsize into the per-namespace struct? It doesn't look more complicated or intrusive and would allow to still change the init_net hashsize. Also seems less hackish :)

How about this (so far untested) patch? The htable_size is moved into the per-namespace struct and initialized from the current (global) value of nf_conntrack_htable_size. Changes through sysfs are still permitted, but only affect the init namespace and newly created ones.

I moved the random seed into the per-ns context aswell. I think that's better than having a global one, and you don't need to rehash all.

Quoted 8 lines
Quoted 4 lines

That's another possibility. But we don't loose anything by not reseeding during resize. It also shouldn't be possible to determine the seed from userspace in a namespace, so there's no real need to use seperate values.

Right, the risk there is hypothetical at best. But there's little lost in putting it in per-ns and then you can rehash and truly make them independent, which I think is really what netns is all about.

I don't disagree, but currently I'm trying to go for a minimal version thats suitable for 2.6.33.

#27 lore raw
In reply to Alexey Dobriyan
Quoted 15 lines · Alexey Dobriyan

Jon Masters correctly points out that conntrack hash sizes (nf_conntrack_htable_size) are global (not per-netns) and modifiable at runtime via /sys/module/nf_conntrack/hashsize .

Steps to reproduce:

	clone(CLONE_NEWNET)
	[grow /sys/module/nf_conntrack/hashsize]
	exit()

At netns exit we are going to scan random memory for conntracks to be killed.

Apparently there is a code which deals with hashtable resize for init_net (and it was there befode netns conntrack code), so prohibit hashsize modification if there is more than one netns exists.

Quoted 5 lines

To change hashtable sizes, you need to reload module.

Expectation hashtable size was simply glued to a variable with no code to rehash expectations, so it was a bug to allow writing to it. Make "expect_hashsize" readonly.

I've applied the expectation part, thanks Alexey.

Quoted 12 lines
net/netfilter/nf_conntrack_expect.c +1 -1
@@ -569,7 +569,7 @@ static void exp_proc_remove(struct net *net)
 #endif /* CONFIG_PROC_FS */
 }
 
-module_param_named(expect_hashsize, nf_ct_expect_hsize, uint, 0600);
+module_param_named(expect_hashsize, nf_ct_expect_hsize, uint, 0400);
 
 int nf_conntrack_expect_init(struct net *net)
 {