tcp: replace hard coded GFP_KERNEL with sk_allocation

This fixed a lockdep warning which appeared when doing stress
memory tests over NFS:

	inconsistent {RECLAIM_FS-ON-W} -> {IN-RECLAIM_FS-W} usage.

	page reclaim => nfs_writepage => tcp_sendmsg => lock sk_lock

	mount_root => nfs_root_data => tcp_close => lock sk_lock =>
			tcp_send_fin => alloc_skb_fclone => page reclaim

David raised a concern that if the allocation fails in tcp_send_fin(), and it's
GFP_ATOMIC, we are going to yield() (which sleeps) and loop endlessly waiting
for the allocation to succeed.

But fact is, the original GFP_KERNEL also sleeps. GFP_ATOMIC+yield() looks
weird, but it is no worse the implicit sleep inside GFP_KERNEL. Both could
loop endlessly under memory pressure.

CC: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
CC: David S. Miller <davem@davemloft.net>
CC: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

6 files changed, 15 insertions, 13 deletions

diff --git a/include/net/tcp.h b/include/net/tcp.h
index df50bc40b5f..b71a446d58f 100644
--- a/include/net/tcp.h
+++ b/include/net/tcp.h
@@ -1186,7 +1186,7 @@ extern int			tcp_v4_md5_do_del(struct sock *sk,
 #define tcp_twsk_md5_key(twsk)	NULL
 #endif
 
-extern struct tcp_md5sig_pool	**tcp_alloc_md5sig_pool(void);
+extern struct tcp_md5sig_pool	**tcp_alloc_md5sig_pool(struct sock *);
 extern void			tcp_free_md5sig_pool(void);
 
 extern struct tcp_md5sig_pool	*__tcp_get_md5sig_pool(int cpu);
diff --git a/net/ipv4/tcp.c b/net/ipv4/tcp.c
index 59f69a6c586..edeea060db4 100644
--- a/net/ipv4/tcp.c
+++ b/net/ipv4/tcp.c
@@ -1839,7 +1839,7 @@ void tcp_close(struct sock *sk, long timeout)
 		/* Unread data was tossed, zap the connection. */
 		NET_INC_STATS_USER(sock_net(sk), LINUX_MIB_TCPABORTONCLOSE);
 		tcp_set_state(sk, TCP_CLOSE);
-		tcp_send_active_reset(sk, GFP_KERNEL);
+		tcp_send_active_reset(sk, sk->sk_allocation);
 	} else if (sock_flag(sk, SOCK_LINGER) && !sk->sk_lingertime) {
 		/* Check zero linger _after_ checking for unread data. */
 		sk->sk_prot->disconnect(sk, 0);
@@ -2658,7 +2658,7 @@ void tcp_free_md5sig_pool(void)
 
 EXPORT_SYMBOL(tcp_free_md5sig_pool);
 
-static struct tcp_md5sig_pool **__tcp_alloc_md5sig_pool(void)
+static struct tcp_md5sig_pool **__tcp_alloc_md5sig_pool(struct sock *sk)
 {
 	int cpu;
 	struct tcp_md5sig_pool **pool;
@@ -2671,7 +2671,7 @@ static struct tcp_md5sig_pool **__tcp_alloc_md5sig_pool(void)
 		struct tcp_md5sig_pool *p;
 		struct crypto_hash *hash;
 
-		p = kzalloc(sizeof(*p), GFP_KERNEL);
+		p = kzalloc(sizeof(*p), sk->sk_allocation);
 		if (!p)
 			goto out_free;
 		*per_cpu_ptr(pool, cpu) = p;
@@ -2688,7 +2688,7 @@ out_free:
 	return NULL;
 }
 
-struct tcp_md5sig_pool **tcp_alloc_md5sig_pool(void)
+struct tcp_md5sig_pool **tcp_alloc_md5sig_pool(struct sock *sk)
 {
 	struct tcp_md5sig_pool **pool;
 	int alloc = 0;
@@ -2709,7 +2709,7 @@ retry:
 
 	if (alloc) {
 		/* we cannot hold spinlock here because this may sleep. */
-		struct tcp_md5sig_pool **p = __tcp_alloc_md5sig_pool();
+		struct tcp_md5sig_pool **p = __tcp_alloc_md5sig_pool(sk);
 		spin_lock_bh(&tcp_md5sig_pool_lock);
 		if (!p) {
 			tcp_md5sig_users--;
diff --git a/net/ipv4/tcp_ipv4.c b/net/ipv4/tcp_ipv4.c
index ce7d3b021ff..0543561da99 100644
--- a/net/ipv4/tcp_ipv4.c
+++ b/net/ipv4/tcp_ipv4.c
@@ -886,7 +886,7 @@ int tcp_v4_md5_do_add(struct sock *sk, __be32 addr,
 			}
 			sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
 		}
-		if (tcp_alloc_md5sig_pool() == NULL) {
+		if (tcp_alloc_md5sig_pool(sk) == NULL) {
 			kfree(newkey);
 			return -ENOMEM;
 		}
@@ -1007,8 +1007,9 @@ static int tcp_v4_parse_md5_keys(struct sock *sk, char __user *optval,
 
 	if (!tcp_sk(sk)->md5sig_info) {
 		struct tcp_sock *tp = tcp_sk(sk);
-		struct tcp_md5sig_info *p = kzalloc(sizeof(*p), GFP_KERNEL);
+		struct tcp_md5sig_info *p;
 
+		p = kzalloc(sizeof(*p), sk->sk_allocation);
 		if (!p)
 			return -EINVAL;
 
@@ -1016,7 +1017,7 @@ static int tcp_v4_parse_md5_keys(struct sock *sk, char __user *optval,
 		sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
 	}
 
-	newkey = kmemdup(cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
+	newkey = kmemdup(cmd.tcpm_key, cmd.tcpm_keylen, sk->sk_allocation);
 	if (!newkey)
 		return -ENOMEM;
 	return tcp_v4_md5_do_add(sk, sin->sin_addr.s_addr,
diff --git a/net/ipv4/tcp_minisocks.c b/net/ipv4/tcp_minisocks.c
index 6c8b42299d9..e48c37d74d7 100644
--- a/net/ipv4/tcp_minisocks.c
+++ b/net/ipv4/tcp_minisocks.c
@@ -322,7 +322,7 @@ void tcp_time_wait(struct sock *sk, int state, int timeo)
 			if (key != NULL) {
 				memcpy(&tcptw->tw_md5_key, key->key, key->keylen);
 				tcptw->tw_md5_keylen = key->keylen;
-				if (tcp_alloc_md5sig_pool() == NULL)
+				if (tcp_alloc_md5sig_pool(sk) == NULL)
 					BUG();
 			}
 		} while (0);
diff --git a/net/ipv4/tcp_output.c b/net/ipv4/tcp_output.c
index 4e004424d40..5200aab0ca9 100644
--- a/net/ipv4/tcp_output.c
+++ b/net/ipv4/tcp_output.c
@@ -2135,7 +2135,8 @@ void tcp_send_fin(struct sock *sk)
 	} else {
 		/* Socket is locked, keep trying until memory is available. */
 		for (;;) {
-			skb = alloc_skb_fclone(MAX_TCP_HEADER, GFP_KERNEL);
+			skb = alloc_skb_fclone(MAX_TCP_HEADER,
+					       sk->sk_allocation);
 			if (skb)
 				break;
 			yield();
@@ -2388,7 +2389,7 @@ int tcp_connect(struct sock *sk)
 	sk->sk_wmem_queued += buff->truesize;
 	sk_mem_charge(sk, buff->truesize);
 	tp->packets_out += tcp_skb_pcount(buff);
-	tcp_transmit_skb(sk, buff, 1, GFP_KERNEL);
+	tcp_transmit_skb(sk, buff, 1, sk->sk_allocation);
 
 	/* We change tp->snd_nxt after the tcp_transmit_skb() call
 	 * in order to make this packet get counted in tcpOutSegs.
diff --git a/net/ipv6/tcp_ipv6.c b/net/ipv6/tcp_ipv6.c
index d73617e9708..65aecf27f2c 100644
--- a/net/ipv6/tcp_ipv6.c
+++ b/net/ipv6/tcp_ipv6.c
@@ -591,7 +591,7 @@ static int tcp_v6_md5_do_add(struct sock *sk, struct in6_addr *peer,
 			}
 			sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
 		}
-		if (tcp_alloc_md5sig_pool() == NULL) {
+		if (tcp_alloc_md5sig_pool(sk) == NULL) {
 			kfree(newkey);
 			return -ENOMEM;
 		}