ipv4: fix IPSKB_FRAG_PMTU handling with fragmentation

This patch removes the iph field from the state structure, which is not
properly initialized. Instead, add a new field to make the "do we want
to set DF" be the state bit and move the code to set the DF flag from
ip_frag_next().

Joint work with Pablo and Linus.

Fixes: 19c3401a917b ("net: ipv4: place control buffer handling away from fragmentation iterators")
Reported-by: Patrick Schönthaler <patrick@notvads.ovh>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

3 files changed, 9 insertions, 8 deletions

diff --git a/include/net/ip.h b/include/net/ip.h
index 95bb77f95bcc..a2c61c36dc4a 100644
--- a/include/net/ip.h
+++ b/include/net/ip.h
@@ -185,7 +185,7 @@ static inline struct sk_buff *ip_fraglist_next(struct ip_fraglist_iter *iter)
 }
 
 struct ip_frag_state {
-	struct iphdr	*iph;
+	bool		DF;
 	unsigned int	hlen;
 	unsigned int	ll_rs;
 	unsigned int	mtu;
@@ -196,7 +196,7 @@ struct ip_frag_state {
 };
 
 void ip_frag_init(struct sk_buff *skb, unsigned int hlen, unsigned int ll_rs,
-		  unsigned int mtu, struct ip_frag_state *state);
+		  unsigned int mtu, bool DF, struct ip_frag_state *state);
 struct sk_buff *ip_frag_next(struct sk_buff *skb,
 			     struct ip_frag_state *state);
 
diff --git a/net/bridge/netfilter/nf_conntrack_bridge.c b/net/bridge/netfilter/nf_conntrack_bridge.c
index 506d6141e44e..809673222382 100644
--- a/net/bridge/netfilter/nf_conntrack_bridge.c
+++ b/net/bridge/netfilter/nf_conntrack_bridge.c
@@ -95,7 +95,7 @@ slow_path:
 	 * This may also be a clone skbuff, we could preserve the geometry for
 	 * the copies but probably not worth the effort.
 	 */
-	ip_frag_init(skb, hlen, ll_rs, frag_max_size, &state);
+	ip_frag_init(skb, hlen, ll_rs, frag_max_size, false, &state);
 
 	while (state.left > 0) {
 		struct sk_buff *skb2;
diff --git a/net/ipv4/ip_output.c b/net/ipv4/ip_output.c
index 814b9b8882a0..3d8baaaf7086 100644
--- a/net/ipv4/ip_output.c
+++ b/net/ipv4/ip_output.c
@@ -645,11 +645,12 @@ void ip_fraglist_prepare(struct sk_buff *skb, struct ip_fraglist_iter *iter)
 EXPORT_SYMBOL(ip_fraglist_prepare);
 
 void ip_frag_init(struct sk_buff *skb, unsigned int hlen,
-		  unsigned int ll_rs, unsigned int mtu,
+		  unsigned int ll_rs, unsigned int mtu, bool DF,
 		  struct ip_frag_state *state)
 {
 	struct iphdr *iph = ip_hdr(skb);
 
+	state->DF = DF;
 	state->hlen = hlen;
 	state->ll_rs = ll_rs;
 	state->mtu = mtu;
@@ -668,9 +669,6 @@ static void ip_frag_ipcb(struct sk_buff *from, struct sk_buff *to,
 	/* Copy the flags to each fragment. */
 	IPCB(to)->flags = IPCB(from)->flags;
 
-	if (IPCB(from)->flags & IPSKB_FRAG_PMTU)
-		state->iph->frag_off |= htons(IP_DF);
-
 	/* ANK: dirty, but effective trick. Upgrade options only if
 	 * the segment to be fragmented was THE FIRST (otherwise,
 	 * options are already fixed) and make it ONCE
@@ -738,6 +736,8 @@ struct sk_buff *ip_frag_next(struct sk_buff *skb, struct ip_frag_state *state)
 	 */
 	iph = ip_hdr(skb2);
 	iph->frag_off = htons((state->offset >> 3));
+	if (state->DF)
+		iph->frag_off |= htons(IP_DF);
 
 	/*
 	 *	Added AC : If we are fragmenting a fragment that's not the
@@ -883,7 +883,8 @@ slow_path:
 	 *	Fragment the datagram.
 	 */
 
-	ip_frag_init(skb, hlen, ll_rs, mtu, &state);
+	ip_frag_init(skb, hlen, ll_rs, mtu, IPCB(skb)->flags & IPSKB_FRAG_PMTU,
+		     &state);
 
 	/*
 	 *	Keep copying data until we run out.