[PATCH 3.2 22/27] tcp: md5: remove spinlock usage in fast path

From: Ben Hutchings
Date: Sun Dec 28 2014 - 20:17:30 EST


3.2.66-rc1 review patch. If anyone has any objections, please let me know.

------------------

From: Eric Dumazet <edumazet@xxxxxxxxxx>

commit 71cea17ed39fdf1c0634f530ddc6a2c2fc601c2b upstream.

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@xxxxxxxxxx>
Cc: Herbert Xu <herbert@xxxxxxxxxxxxxxxxxxx>
Signed-off-by: David S. Miller <davem@xxxxxxxxxxxxx>
[bwh: Backported to 3.2:
- Adjust context
- Conditions for alloc/free are quite different]
Signed-off-by: Ben Hutchings <ben@xxxxxxxxxxxxxxx>
---
--- a/include/net/tcp.h
+++ b/include/net/tcp.h
@@ -1214,11 +1214,13 @@ extern int tcp_v4_md5_do_del(struct sock
#define tcp_twsk_md5_key(twsk) NULL
#endif

-extern struct tcp_md5sig_pool __percpu *tcp_alloc_md5sig_pool(struct sock *);
-extern void tcp_free_md5sig_pool(void);
+extern bool tcp_alloc_md5sig_pool(void);

extern struct tcp_md5sig_pool *tcp_get_md5sig_pool(void);
-extern void tcp_put_md5sig_pool(void);
+static inline void tcp_put_md5sig_pool(void)
+{
+ local_bh_enable();
+}

extern int tcp_md5_hash_header(struct tcp_md5sig_pool *, const struct tcphdr *);
extern int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *,
--- a/net/ipv4/tcp.c
+++ b/net/ipv4/tcp.c
@@ -2863,9 +2863,8 @@ int tcp_gro_complete(struct sk_buff *skb
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)
{
@@ -2880,30 +2879,14 @@ static void __tcp_free_md5sig_pool(struc
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;
@@ -2914,53 +2897,27 @@ __tcp_alloc_md5sig_pool(struct sock *sk)

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;
- int alloc = 0;
+ if (unlikely(!tcp_md5sig_pool)) {
+ mutex_lock(&tcp_md5sig_mutex);
+
+ if (!tcp_md5sig_pool)
+ __tcp_alloc_md5sig_pool();

-retry:
- spin_lock_bh(&tcp_md5sig_pool_lock);
- pool = tcp_md5sig_pool;
- if (tcp_md5sig_users++ == 0) {
- alloc = 1;
- 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 (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);

@@ -2977,28 +2934,15 @@ struct tcp_md5sig_pool *tcp_get_md5sig_p
struct tcp_md5sig_pool __percpu *p;

local_bh_disable();
-
- spin_lock(&tcp_md5sig_pool_lock);
- p = tcp_md5sig_pool;
- if (p)
- tcp_md5sig_users++;
- spin_unlock(&tcp_md5sig_pool_lock);
-
+ p = ACCESS_ONCE(tcp_md5sig_pool);
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)
{
--- a/net/ipv4/tcp_ipv4.c
+++ b/net/ipv4/tcp_ipv4.c
@@ -938,8 +938,7 @@ int tcp_v4_md5_do_add(struct sock *sk, _
}

md5sig = tp->md5sig_info;
- if (md5sig->entries4 == 0 &&
- tcp_alloc_md5sig_pool(sk) == NULL) {
+ if (md5sig->entries4 == 0 && !tcp_alloc_md5sig_pool()) {
kfree(newkey);
return -ENOMEM;
}
@@ -949,8 +948,6 @@ int tcp_v4_md5_do_add(struct sock *sk, _
(md5sig->entries4 + 1)), GFP_ATOMIC);
if (!keys) {
kfree(newkey);
- if (md5sig->entries4 == 0)
- tcp_free_md5sig_pool();
return -ENOMEM;
}

@@ -994,7 +991,6 @@ int tcp_v4_md5_do_del(struct sock *sk, _
kfree(tp->md5sig_info->keys4);
tp->md5sig_info->keys4 = NULL;
tp->md5sig_info->alloced4 = 0;
- tcp_free_md5sig_pool();
} else if (tp->md5sig_info->entries4 != i) {
/* Need to do some manipulation */
memmove(&tp->md5sig_info->keys4[i],
@@ -1022,7 +1018,6 @@ static void tcp_v4_clear_md5_list(struct
for (i = 0; i < tp->md5sig_info->entries4; i++)
kfree(tp->md5sig_info->keys4[i].base.key);
tp->md5sig_info->entries4 = 0;
- tcp_free_md5sig_pool();
}
if (tp->md5sig_info->keys4) {
kfree(tp->md5sig_info->keys4);
--- a/net/ipv4/tcp_minisocks.c
+++ b/net/ipv4/tcp_minisocks.c
@@ -365,7 +365,7 @@ void tcp_time_wait(struct sock *sk, int
if (key != NULL) {
memcpy(&tcptw->tw_md5_key, key->key, key->keylen);
tcptw->tw_md5_keylen = key->keylen;
- if (tcp_alloc_md5sig_pool(sk) == NULL)
+ if (!tcp_alloc_md5sig_pool())
BUG();
}
} while (0);
@@ -403,11 +403,6 @@ void tcp_time_wait(struct sock *sk, int

void tcp_twsk_destructor(struct sock *sk)
{
-#ifdef CONFIG_TCP_MD5SIG
- struct tcp_timewait_sock *twsk = tcp_twsk(sk);
- if (twsk->tw_md5_keylen)
- tcp_free_md5sig_pool();
-#endif
}
EXPORT_SYMBOL_GPL(tcp_twsk_destructor);

--- a/net/ipv6/tcp_ipv6.c
+++ b/net/ipv6/tcp_ipv6.c
@@ -592,7 +592,7 @@ static int tcp_v6_md5_do_add(struct sock
sk_nocaps_add(sk, NETIF_F_GSO_MASK);
}
if (tp->md5sig_info->entries6 == 0 &&
- tcp_alloc_md5sig_pool(sk) == NULL) {
+ !tcp_alloc_md5sig_pool()) {
kfree(newkey);
return -ENOMEM;
}
@@ -602,8 +602,6 @@ static int tcp_v6_md5_do_add(struct sock

if (!keys) {
kfree(newkey);
- if (tp->md5sig_info->entries6 == 0)
- tcp_free_md5sig_pool();
return -ENOMEM;
}

@@ -649,7 +647,6 @@ static int tcp_v6_md5_do_del(struct sock
kfree(tp->md5sig_info->keys6);
tp->md5sig_info->keys6 = NULL;
tp->md5sig_info->alloced6 = 0;
- tcp_free_md5sig_pool();
} else {
/* shrink the database */
if (tp->md5sig_info->entries6 != i)
@@ -673,7 +670,6 @@ static void tcp_v6_clear_md5_list (struc
for (i = 0; i < tp->md5sig_info->entries6; i++)
kfree(tp->md5sig_info->keys6[i].base.key);
tp->md5sig_info->entries6 = 0;
- tcp_free_md5sig_pool();
}

kfree(tp->md5sig_info->keys6);
@@ -684,7 +680,6 @@ static void tcp_v6_clear_md5_list (struc
for (i = 0; i < tp->md5sig_info->entries4; i++)
kfree(tp->md5sig_info->keys4[i].base.key);
tp->md5sig_info->entries4 = 0;
- tcp_free_md5sig_pool();
}

kfree(tp->md5sig_info->keys4);

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/