From: Ben Hutchings <ben@decadent.org.uk>
Date: Mon, 25 Sep 2017 23:03:33 +0000 (+0100)
Subject: inet_frag: Limit ABI change in 4.9.51
X-Git-Tag: archive/raspbian/4.9.51-1+rpi1~8^2~1
X-Git-Url: https://dgit.raspbian.org/?a=commitdiff_plain;h=fdcd91b4911529e043a5eb146add4a9a44518c38;p=linux-4.9.git

inet_frag: Limit ABI change in 4.9.51

Commit fb452a1aa3fd 'Revert "net: use lib/percpu_counter API for
fragmentation mem accounting"' changed the type of inet_frag::mem
from struct percpu_counter to atomic_t.  There are few modules
that actually depend on this, but struct inet_frag is embedded in
various other structures that end up affecting symbol versions
for a huge number of networking APIs.

So add padding to keep the structure size unchanged, hide the type
change from genksyms, and rename a key inet_frag function so that
those modules that actually depend on the type of inet_frag::mem do
see an ABI change.


Gbp-Pq: Topic debian
Gbp-Pq: Name inet_frag-limit-abi-change-in-4.9.51.patch
---

diff --git a/include/net/inet_frag.h b/include/net/inet_frag.h
index 634d19203e7d..ac6f5b167c83 100644
--- a/include/net/inet_frag.h
+++ b/include/net/inet_frag.h
@@ -1,9 +1,22 @@
 #ifndef __NET_FRAG_H__
 #define __NET_FRAG_H__
 
+#include <linux/percpu_counter.h>
+
 struct netns_frags {
+	/*
+	 * bwh: This change is hidden from genksyms, but we still make
+	 * sure to avoid an ABI mismatch for the modules that access
+	 * it (see comment below).
+	 */
+#ifndef __GENKSYMS__
 	/* Keep atomic mem on separate cachelines in structs that include it */
 	atomic_t		mem ____cacheline_aligned_in_smp;
+	char			pad[sizeof(struct percpu_counter) -
+				    sizeof(atomic_t)];
+#else
+	struct percpu_counter   mem ____cacheline_aligned_in_smp;
+#endif
 	/* sysctls */
 	int			timeout;
 	int			high_thresh;
@@ -107,6 +120,14 @@ static inline void inet_frags_init_net(struct netns_frags *nf)
 {
 	atomic_set(&nf->mem, 0);
 }
+/*
+ * bwh: All modules accessing inet_frag::mem through the inline
+ * functions below also call inet_frags_exit_net().  Change the
+ * function name together with that field's type, so that all loaded
+ * code agrees on whether the type is atomic_t or struct
+ * percpu_counter.
+ */
+#define inet_frags_exit_net inet_frags_exit_net_atomic
 void inet_frags_exit_net(struct netns_frags *nf, struct inet_frags *f);
 
 void inet_frag_kill(struct inet_frag_queue *q, struct inet_frags *f);