From: Eric Dumazet <eduma...@google.com>

tcp_snd_una_update() and tcp_rcv_nxt_update() call
u64_stats_update_begin() either from process context or BH handler.

This triggers a lockdep splat on 32bit & SMP builds.

We could add u64_stats_update_begin_bh() variant but this would
slow down 32bit builds with useless local_disable_bh() and
local_enable_bh() pairs, since we own the socket lock at this point.

I add sock_owned_by_me() helper to have proper lockdep support
even on 64bit builds, and new u64_stats_update_begin_raw()
and u64_stats_update_end_raw methods.

Fixes: c10d9310edf5 ("tcp: do not assume TCP code is non preemptible")
Reported-by: Fabio Estevam <feste...@gmail.com>
Diagnosed-by: Francois Romieu <rom...@fr.zoreil.com>
Signed-off-by: Eric Dumazet <eduma...@google.com>
---
 include/linux/u64_stats_sync.h |   14 ++++++++++++++
 include/net/sock.h             |    7 ++++++-
 net/ipv4/tcp_input.c           |   10 ++++++----
 3 files changed, 26 insertions(+), 5 deletions(-)

diff --git a/include/linux/u64_stats_sync.h b/include/linux/u64_stats_sync.h
index 
df89c9bcba7db8dbde3bbf2b99f9af6ed562b112..d3a2bb712af3b9613b98ef9c3219f8dcd31568a5
 100644
--- a/include/linux/u64_stats_sync.h
+++ b/include/linux/u64_stats_sync.h
@@ -89,6 +89,20 @@ static inline void u64_stats_update_end(struct 
u64_stats_sync *syncp)
 #endif
 }
 
+static inline void u64_stats_update_begin_raw(struct u64_stats_sync *syncp)
+{
+#if BITS_PER_LONG==32 && defined(CONFIG_SMP)
+       raw_write_seqcount_begin(&syncp->seq);
+#endif
+}
+
+static inline void u64_stats_update_end_raw(struct u64_stats_sync *syncp)
+{
+#if BITS_PER_LONG==32 && defined(CONFIG_SMP)
+       raw_write_seqcount_end(&syncp->seq);
+#endif
+}
+
 static inline unsigned int u64_stats_fetch_begin(const struct u64_stats_sync 
*syncp)
 {
 #if BITS_PER_LONG==32 && defined(CONFIG_SMP)
diff --git a/include/net/sock.h b/include/net/sock.h
index 
45f5b492c65883cd22e2f615e019fe0d0ba31167..c9c8b19df27c558354687119db60c0716909ea3f
 100644
--- a/include/net/sock.h
+++ b/include/net/sock.h
@@ -1421,11 +1421,16 @@ static inline void unlock_sock_fast(struct sock *sk, 
bool slow)
  * accesses from user process context.
  */
 
-static inline bool sock_owned_by_user(const struct sock *sk)
+static inline void sock_owned_by_me(const struct sock *sk)
 {
 #ifdef CONFIG_LOCKDEP
        WARN_ON_ONCE(!lockdep_sock_is_held(sk) && debug_locks);
 #endif
+}
+
+static inline bool sock_owned_by_user(const struct sock *sk)
+{
+       sock_owned_by_me(sk);
        return sk->sk_lock.owned;
 }
 
diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c
index 
6171f92be0903f5a5d17f027dbe6b31829bcc043..a914e0607895dd9321559f93c1008f8de13b73ad
 100644
--- a/net/ipv4/tcp_input.c
+++ b/net/ipv4/tcp_input.c
@@ -3355,9 +3355,10 @@ static void tcp_snd_una_update(struct tcp_sock *tp, u32 
ack)
 {
        u32 delta = ack - tp->snd_una;
 
-       u64_stats_update_begin(&tp->syncp);
+       sock_owned_by_me((struct sock *)tp);
+       u64_stats_update_begin_raw(&tp->syncp);
        tp->bytes_acked += delta;
-       u64_stats_update_end(&tp->syncp);
+       u64_stats_update_end_raw(&tp->syncp);
        tp->snd_una = ack;
 }
 
@@ -3366,9 +3367,10 @@ static void tcp_rcv_nxt_update(struct tcp_sock *tp, u32 
seq)
 {
        u32 delta = seq - tp->rcv_nxt;
 
-       u64_stats_update_begin(&tp->syncp);
+       sock_owned_by_me((struct sock *)tp);
+       u64_stats_update_begin_raw(&tp->syncp);
        tp->bytes_received += delta;
-       u64_stats_update_end(&tp->syncp);
+       u64_stats_update_end_raw(&tp->syncp);
        tp->rcv_nxt = seq;
 }
 


Reply via email to