tcp: md5: remove spinlock usage in fast path
TCP md5 code uses per cpu variables but protects access to them with
a shared spinlock, which is a contention point.
[ tcp_md5sig_pool_lock is locked twice per incoming packet ]
Makes things much simpler, by allocating crypto structures once, first
time a socket needs md5 keys, and not deallocating them as they are
really small.
Next step would be to allow crypto allocations being done in a NUMA
aware way.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
diff --git a/net/ipv4/tcp.c b/net/ipv4/tcp.c
index dcb116d..53d9c12 100644
--- a/net/ipv4/tcp.c
+++ b/net/ipv4/tcp.c
@@ -3095,9 +3095,8 @@
EXPORT_SYMBOL(tcp_gro_complete);
#ifdef CONFIG_TCP_MD5SIG
-static unsigned long tcp_md5sig_users;
-static struct tcp_md5sig_pool __percpu *tcp_md5sig_pool;
-static DEFINE_SPINLOCK(tcp_md5sig_pool_lock);
+static struct tcp_md5sig_pool __percpu *tcp_md5sig_pool __read_mostly;
+static DEFINE_MUTEX(tcp_md5sig_mutex);
static void __tcp_free_md5sig_pool(struct tcp_md5sig_pool __percpu *pool)
{
@@ -3112,30 +3111,14 @@
free_percpu(pool);
}
-void tcp_free_md5sig_pool(void)
-{
- struct tcp_md5sig_pool __percpu *pool = NULL;
-
- spin_lock_bh(&tcp_md5sig_pool_lock);
- if (--tcp_md5sig_users == 0) {
- pool = tcp_md5sig_pool;
- tcp_md5sig_pool = NULL;
- }
- spin_unlock_bh(&tcp_md5sig_pool_lock);
- if (pool)
- __tcp_free_md5sig_pool(pool);
-}
-EXPORT_SYMBOL(tcp_free_md5sig_pool);
-
-static struct tcp_md5sig_pool __percpu *
-__tcp_alloc_md5sig_pool(struct sock *sk)
+static void __tcp_alloc_md5sig_pool(void)
{
int cpu;
struct tcp_md5sig_pool __percpu *pool;
pool = alloc_percpu(struct tcp_md5sig_pool);
if (!pool)
- return NULL;
+ return;
for_each_possible_cpu(cpu) {
struct crypto_hash *hash;
@@ -3146,53 +3129,27 @@
per_cpu_ptr(pool, cpu)->md5_desc.tfm = hash;
}
- return pool;
+ /* before setting tcp_md5sig_pool, we must commit all writes
+ * to memory. See ACCESS_ONCE() in tcp_get_md5sig_pool()
+ */
+ smp_wmb();
+ tcp_md5sig_pool = pool;
+ return;
out_free:
__tcp_free_md5sig_pool(pool);
- return NULL;
}
-struct tcp_md5sig_pool __percpu *tcp_alloc_md5sig_pool(struct sock *sk)
+bool tcp_alloc_md5sig_pool(void)
{
- struct tcp_md5sig_pool __percpu *pool;
- bool alloc = false;
+ if (unlikely(!tcp_md5sig_pool)) {
+ mutex_lock(&tcp_md5sig_mutex);
-retry:
- spin_lock_bh(&tcp_md5sig_pool_lock);
- pool = tcp_md5sig_pool;
- if (tcp_md5sig_users++ == 0) {
- alloc = true;
- spin_unlock_bh(&tcp_md5sig_pool_lock);
- } else if (!pool) {
- tcp_md5sig_users--;
- spin_unlock_bh(&tcp_md5sig_pool_lock);
- cpu_relax();
- goto retry;
- } else
- spin_unlock_bh(&tcp_md5sig_pool_lock);
+ if (!tcp_md5sig_pool)
+ __tcp_alloc_md5sig_pool();
- if (alloc) {
- /* we cannot hold spinlock here because this may sleep. */
- struct tcp_md5sig_pool __percpu *p;
-
- p = __tcp_alloc_md5sig_pool(sk);
- spin_lock_bh(&tcp_md5sig_pool_lock);
- if (!p) {
- tcp_md5sig_users--;
- spin_unlock_bh(&tcp_md5sig_pool_lock);
- return NULL;
- }
- pool = tcp_md5sig_pool;
- if (pool) {
- /* oops, it has already been assigned. */
- spin_unlock_bh(&tcp_md5sig_pool_lock);
- __tcp_free_md5sig_pool(p);
- } else {
- tcp_md5sig_pool = pool = p;
- spin_unlock_bh(&tcp_md5sig_pool_lock);
- }
+ mutex_unlock(&tcp_md5sig_mutex);
}
- return pool;
+ return tcp_md5sig_pool != NULL;
}
EXPORT_SYMBOL(tcp_alloc_md5sig_pool);
@@ -3209,28 +3166,15 @@
struct tcp_md5sig_pool __percpu *p;
local_bh_disable();
-
- spin_lock(&tcp_md5sig_pool_lock);
- p = tcp_md5sig_pool;
+ p = ACCESS_ONCE(tcp_md5sig_pool);
if (p)
- tcp_md5sig_users++;
- spin_unlock(&tcp_md5sig_pool_lock);
-
- if (p)
- return this_cpu_ptr(p);
+ return __this_cpu_ptr(p);
local_bh_enable();
return NULL;
}
EXPORT_SYMBOL(tcp_get_md5sig_pool);
-void tcp_put_md5sig_pool(void)
-{
- local_bh_enable();
- tcp_free_md5sig_pool();
-}
-EXPORT_SYMBOL(tcp_put_md5sig_pool);
-
int tcp_md5_hash_header(struct tcp_md5sig_pool *hp,
const struct tcphdr *th)
{
diff --git a/net/ipv4/tcp_ipv4.c b/net/ipv4/tcp_ipv4.c
index 7196523..d20ede0 100644
--- a/net/ipv4/tcp_ipv4.c
+++ b/net/ipv4/tcp_ipv4.c
@@ -1026,7 +1026,7 @@
key = sock_kmalloc(sk, sizeof(*key), gfp);
if (!key)
return -ENOMEM;
- if (hlist_empty(&md5sig->head) && !tcp_alloc_md5sig_pool(sk)) {
+ if (!tcp_alloc_md5sig_pool()) {
sock_kfree_s(sk, key, sizeof(*key));
return -ENOMEM;
}
@@ -1044,9 +1044,7 @@
int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, int family)
{
- struct tcp_sock *tp = tcp_sk(sk);
struct tcp_md5sig_key *key;
- struct tcp_md5sig_info *md5sig;
key = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&addr, AF_INET);
if (!key)
@@ -1054,10 +1052,6 @@
hlist_del_rcu(&key->node);
atomic_sub(sizeof(*key), &sk->sk_omem_alloc);
kfree_rcu(key, rcu);
- md5sig = rcu_dereference_protected(tp->md5sig_info,
- sock_owned_by_user(sk));
- if (hlist_empty(&md5sig->head))
- tcp_free_md5sig_pool();
return 0;
}
EXPORT_SYMBOL(tcp_md5_do_del);
@@ -1071,8 +1065,6 @@
md5sig = rcu_dereference_protected(tp->md5sig_info, 1);
- if (!hlist_empty(&md5sig->head))
- tcp_free_md5sig_pool();
hlist_for_each_entry_safe(key, n, &md5sig->head, node) {
hlist_del_rcu(&key->node);
atomic_sub(sizeof(*key), &sk->sk_omem_alloc);
diff --git a/net/ipv4/tcp_minisocks.c b/net/ipv4/tcp_minisocks.c
index 0f01788..ab1c086 100644
--- a/net/ipv4/tcp_minisocks.c
+++ b/net/ipv4/tcp_minisocks.c
@@ -317,7 +317,7 @@
key = tp->af_specific->md5_lookup(sk, sk);
if (key != NULL) {
tcptw->tw_md5_key = kmemdup(key, sizeof(*key), GFP_ATOMIC);
- if (tcptw->tw_md5_key && tcp_alloc_md5sig_pool(sk) == NULL)
+ if (tcptw->tw_md5_key && !tcp_alloc_md5sig_pool())
BUG();
}
} while (0);
@@ -358,10 +358,8 @@
#ifdef CONFIG_TCP_MD5SIG
struct tcp_timewait_sock *twsk = tcp_twsk(sk);
- if (twsk->tw_md5_key) {
- tcp_free_md5sig_pool();
+ if (twsk->tw_md5_key)
kfree_rcu(twsk->tw_md5_key, rcu);
- }
#endif
}
EXPORT_SYMBOL_GPL(tcp_twsk_destructor);