Skip to contentSkip to conversation

netfilter: nfnetlink: Remove VLA usage

PATCH Superseded Kees Cook 3 participants 4 messages May 30, 2018

Subsystems (3)

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore

Kees Cook

author
#0 lore raw

PATCH 1/1 netfilter: nfnetlink: Remove VLA usage

In the quest to remove all stack VLA usage from the kernel[1], this allocates the maximum size expected for all possible attrs and adds a sanity-check to make sure nothing gets out of sync.

[1] https://lkml.kernel.org/r/CA+55aFzCG-zNmZwX4A2FQpadafLfEzK6CC=qPXydAacU1RqZWA@mail.gmail.com

---
 net/netfilter/nfnetlink.c | 22 ++++++++++++++++++++--
 1 file changed, 20 insertions(+), 2 deletions(-)
net/netfilter/nfnetlink.c +20 -2
index 03ead8a9e90c..0cb395f9627e 100644
@@ -28,6 +28,7 @@
 
 #include <net/netlink.h>
 #include <linux/netfilter/nfnetlink.h>
+#include <linux/netfilter/nf_tables.h>
 
 MODULE_LICENSE("GPL");
 MODULE_AUTHOR("Harald Welte <[email protected]>");
@@ -37,6 +38,11 @@ MODULE_ALIAS_NET_PF_PROTO(PF_NETLINK, NETLINK_NETFILTER);
 	rcu_dereference_protected(table[(id)].subsys, \
 				  lockdep_nfnl_is_held((id)))
 
+#define NFTA_MAX_ATTR	max(max(max(NFTA_CHAIN_MAX, NFTA_FLOWTABLE_MAX),\
+				max(NFTA_OBJ_MAX, NFTA_RULE_MAX)),	\
+			    max(NFTA_TABLE_MAX,				\
+				max(NFTA_SET_ELEM_LIST_MAX, NFTA_SET_MAX)))
+
 static struct {
 	struct mutex				mutex;
 	const struct nfnetlink_subsystem __rcu	*subsys;
@@ -185,11 +191,17 @@ static int nfnetlink_rcv_msg(struct sk_buff *skb, struct nlmsghdr *nlh,
 	{
 		int min_len = nlmsg_total_size(sizeof(struct nfgenmsg));
 		u8 cb_id = NFNL_MSG_TYPE(nlh->nlmsg_type);
-		struct nlattr *cda[ss->cb[cb_id].attr_count + 1];
+		struct nlattr *cda[NFTA_MAX_ATTR + 1];
 		struct nlattr *attr = (void *)nlh + min_len;
 		int attrlen = nlh->nlmsg_len - min_len;
 		__u8 subsys_id = NFNL_SUBSYS_ID(type);
 
+		/* Sanity-check NFTA_MAX_ATTR */
+		if (ss->cb[cb_id].attr_count > NFTA_MAX_ATTR) {
+			rcu_read_unlock();
+			return -ENOMEM;
+		}
+
 		err = nla_parse(cda, ss->cb[cb_id].attr_count, attr, attrlen,
 				ss->cb[cb_id].policy, extack);
 		if (err < 0) {
@@ -379,10 +391,16 @@ static void nfnetlink_rcv_batch(struct sk_buff *skb, struct nlmsghdr *nlh,
 		{
 			int min_len = nlmsg_total_size(sizeof(struct nfgenmsg));
 			u8 cb_id = NFNL_MSG_TYPE(nlh->nlmsg_type);
-			struct nlattr *cda[ss->cb[cb_id].attr_count + 1];
+			struct nlattr *cda[NFTA_MAX_ATTR + 1];
 			struct nlattr *attr = (void *)nlh + min_len;
 			int attrlen = nlh->nlmsg_len - min_len;
 
+			/* Sanity-check NFTA_MAX_ATTR */
+			if (ss->cb[cb_id].attr_count > NFTA_MAX_ATTR) {
+				err = -ENOMEM;
+				goto ack;
+			}
+
 			err = nla_parse(cda, ss->cb[cb_id].attr_count, attr,
 					attrlen, ss->cb[cb_id].policy, NULL);
 			if (err < 0)
Signature · 6 lines
2.17.0


-- 
Kees Cook
Pixel Security
maintainer
#1 lore raw
Quoted 27 lines · Kees Cook

In the quest to remove all stack VLA usage from the kernel[1], this allocates the maximum size expected for all possible attrs and adds a sanity-check to make sure nothing gets out of sync.

[1] https://lkml.kernel.org/r/CA+55aFzCG-zNmZwX4A2FQpadafLfEzK6CC=qPXydAacU1RqZWA@mail.gmail.com

---
 net/netfilter/nfnetlink.c | 22 ++++++++++++++++++++--
 1 file changed, 20 insertions(+), 2 deletions(-)
net/netfilter/nfnetlink.c +1 -1
index 03ead8a9e90c..0cb395f9627e 100644
@@ -28,6 +28,7 @@
 
 #include <net/netlink.h>
 #include <linux/netfilter/nfnetlink.h>
+#include <linux/netfilter/nf_tables.h>
 
 	const struct nfnetlink_subsystem __rcu	*subsys;
@@ -185,11 +191,17 @@ static int nfnetlink_rcv_msg(struct sk_buff *skb, struct nlmsghdr *nlh,
 	{
 		int min_len = nlmsg_total_size(sizeof(struct nfgenmsg));
 		u8 cb_id = NFNL_MSG_TYPE(nlh->nlmsg_type);
-		struct nlattr *cda[ss->cb[cb_id].attr_count + 1];

net/netfilter/nfnetlink.c static int nfnetlink_rcv_msg(struct sk_buff *skb, struct nlmsghdr *nlh, · line 203

Quoted 10 lines
net/netfilter/nfnetlink.c +6 -0
+		struct nlattr *cda[NFTA_MAX_ATTR + 1];
 		struct nlattr *attr = (void *)nlh + min_len;
 		int attrlen = nlh->nlmsg_len - min_len;
 		__u8 subsys_id = NFNL_SUBSYS_ID(type);
 
+		/* Sanity-check NFTA_MAX_ATTR */
+		if (ss->cb[cb_id].attr_count > NFTA_MAX_ATTR) {
+			rcu_read_unlock();
+			return -ENOMEM;
+		}

Would you mind also adding check to nfnetlink_subsys_register, plus WARN()?

That way we'll see that NFTA_MAX_ATTR isn't large enough by virtue of registration rather than actual use.

Other than that this looks fine to me.

#2 lore raw
In reply to Kees Cook
Quoted 24 lines · Kees Cook

In the quest to remove all stack VLA usage from the kernel[1], this allocates the maximum size expected for all possible attrs and adds a sanity-check to make sure nothing gets out of sync.

[1] https://lkml.kernel.org/r/CA+55aFzCG-zNmZwX4A2FQpadafLfEzK6CC=qPXydAacU1RqZWA@mail.gmail.com

---
 net/netfilter/nfnetlink.c | 22 ++++++++++++++++++++--
 1 file changed, 20 insertions(+), 2 deletions(-)
net/netfilter/nfnetlink.c +1 -0
index 03ead8a9e90c..0cb395f9627e 100644
@@ -28,6 +28,7 @@
 
 #include <net/netlink.h>
 #include <linux/netfilter/nfnetlink.h>
+#include <linux/netfilter/nf_tables.h>
 
 MODULE_LICENSE("GPL");
 MODULE_AUTHOR("Harald Welte <[email protected]>");

net/netfilter/nfnetlink.c MODULE_ALIAS_NET_PF_PROTO(PF_NETLINK, NETLINK_NETFILTER); · line 44

Quoted 7 lines
net/netfilter/nfnetlink.c +4 -0
@@ -37,6 +38,11 @@ MODULE_ALIAS_NET_PF_PROTO(PF_NETLINK, NETLINK_NETFILTER);
 	rcu_dereference_protected(table[(id)].subsys, \
 				  lockdep_nfnl_is_held((id)))
 
+#define NFTA_MAX_ATTR	max(max(max(NFTA_CHAIN_MAX, NFTA_FLOWTABLE_MAX),\
+				max(NFTA_OBJ_MAX, NFTA_RULE_MAX)),	\
+			    max(NFTA_TABLE_MAX,				\
+				max(NFTA_SET_ELEM_LIST_MAX, NFTA_SET_MAX)))

This is very specific of nftables, there are other nf subsystems using nfnetlink that may go over this maximum attribute value (grep from "struct nfnetlink_subsystem").

To remove the VLA, I think we need an artificial maximum attribute that reasonably large enough.

author
#3 lore raw
Quoted 28 lines · Pablo Neira Ayuso
Quoted 27 lines · Kees Cook

In the quest to remove all stack VLA usage from the kernel[1], this allocates the maximum size expected for all possible attrs and adds a sanity-check to make sure nothing gets out of sync.

[1] https://lkml.kernel.org/r/CA+55aFzCG-zNmZwX4A2FQpadafLfEzK6CC=qPXydAacU1RqZWA@mail.gmail.com

---
 net/netfilter/nfnetlink.c | 22 ++++++++++++++++++++--
 1 file changed, 20 insertions(+), 2 deletions(-)
net/netfilter/nfnetlink.c +1 -0
index 03ead8a9e90c..0cb395f9627e 100644
@@ -28,6 +28,7 @@
 
 #include <net/netlink.h>
 #include <linux/netfilter/nfnetlink.h>
+#include <linux/netfilter/nf_tables.h>
 
 MODULE_LICENSE("GPL");
 MODULE_AUTHOR("Harald Welte <[email protected]>");
@@ -37,6 +38,11 @@ MODULE_ALIAS_NET_PF_PROTO(PF_NETLINK, NETLINK_NETFILTER);
      rcu_dereference_protected(table[(id)].subsys, \
                                lockdep_nfnl_is_held((id)))
 
Quoted 8 lines

net/netfilter/nfnetlink.c MODULE_ALIAS_NET_PF_PROTO(PF_NETLINK, NETLINK_NETFILTER); · line 44

Quoted 4 lines
net/netfilter/nfnetlink.c +4 -0
+#define NFTA_MAX_ATTR        max(max(max(NFTA_CHAIN_MAX, NFTA_FLOWTABLE_MAX),\
+                             max(NFTA_OBJ_MAX, NFTA_RULE_MAX)),      \
+                         max(NFTA_TABLE_MAX,                         \
+                             max(NFTA_SET_ELEM_LIST_MAX, NFTA_SET_MAX)))

This is very specific of nftables, there are other nf subsystems using nfnetlink that may go over this maximum attribute value (grep from "struct nfnetlink_subsystem").

Oops, yes. I see that now.

Quoted 2 lines

To remove the VLA, I think we need an artificial maximum attribute that reasonably large enough.

git grep insanity:

$ for i in $(git grep -A10 'static .*struct nfnetlink_subsystem' | \
                 grep '\.cb\b' | awk '{print $NF}' | cut -d, -f1)
do git grep -A100 'static .*'"$i\b"; done | \
    grep '\.attr_count\b' | awk -F= '{print $NF}' | \
    sed -e 's/ //g' | cut -d, -f1 | sort -u

and manual counting gets me:

CTA_EXPECT_MAX 12
CTA_MAX 25
CTA_TIMEOUT_MAX 6
IPSET_ATTR_CMD_MAX 11
NFACCT_MAX 9
NFCTH_MAX 7
NFQA_CFG_MAX 6
NFQA_MAX 21
NFTA_CHAIN_MAX 10
NFTA_COMPAT_MAX 4
NFTA_OBJ_MAX 8
NFTA_RULE_MAX 10
NFTA_SET_ELEM_LIST_MAX 5
NFTA_SET_MAX 17
NFTA_TABLE_MAX 6
NFULA_CFG_MAX 7
NFULA_MAX 20
OSF_ATTR_MAX 2

How about 32?

As Florian suggested, I'll add the check in
nfnetlink_subsys_register() with a WARN().

-Kees

Signature · 2 lines
Kees Cook
Pixel Security