[PATCH 3/5] tcp/md5: Alloc tcp_md5sig_pool only in setsockopt()

[Dmitry Safonov]

Besides setsockopt() tcp_md5_do_add() can be called from
tcp_v4_syn_recv_sock()/tcp_v6_syn_recv_sock().
If it is called from there, tcp_md5_do_lookup() has succeeded, which
means that md5 static key is enabled. Which makes this
tcp_alloc_md5sig_pool() call to be nop for any caller, but
tcp_v4_parse_md5_keys()/tcp_v6_parse_md5_keys().

tcp_alloc_md5sig_pool() can sleep if tcp_md5sig_pool hasn't been
populated, so if anything changes tcp_md5_do_add() may start sleeping in
atomic context.

Let's leave the check for tcp_md5sig_pool in tcp_md5_do_add(), but
intentionally call tcp_alloc_md5sig_pool() only from sleepable
setsockopt() syscall context.

Signed-off-by: Dmitry Safonov <dima@xxxxxxxxxx>
---
 net/ipv4/tcp_ipv4.c | 5 ++++-
 net/ipv6/tcp_ipv6.c | 3 +++
 2 files changed, 7 insertions(+), 1 deletion(-)

diff --git a/net/ipv4/tcp_ipv4.c b/net/ipv4/tcp_ipv4.c
index 13d868c43284..6a8ff9ab1cbc 100644
--- a/net/ipv4/tcp_ipv4.c
+++ b/net/ipv4/tcp_ipv4.c
@@ -1190,7 +1190,7 @@ int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
 	key = sock_kmalloc(sk, sizeof(*key), gfp | __GFP_ZERO);
 	if (!key)
 		return -ENOMEM;
-	if (!tcp_alloc_md5sig_pool()) {
+	if (WARN_ON_ONCE(!tcp_md5sig_pool_ready())) {
 		sock_kfree_s(sk, key, sizeof(*key));
 		return -ENOMEM;
 	}
@@ -1294,6 +1294,9 @@ static int tcp_v4_parse_md5_keys(struct sock *sk, int optname,
 	if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
 		return -EINVAL;
 
+	if (!tcp_alloc_md5sig_pool())
+		return -ENOMEM;
+
 	return tcp_md5_do_add(sk, addr, AF_INET, prefixlen, l3index, flags,
 			      cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
 }
diff --git a/net/ipv6/tcp_ipv6.c b/net/ipv6/tcp_ipv6.c
index 2cc9b0e53ad1..3af13bd6fed0 100644
--- a/net/ipv6/tcp_ipv6.c
+++ b/net/ipv6/tcp_ipv6.c
@@ -654,6 +654,9 @@ static int tcp_v6_parse_md5_keys(struct sock *sk, int optname,
 	if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
 		return -EINVAL;
 
+	if (!tcp_alloc_md5sig_pool())
+		return -ENOMEM;
+
 	if (ipv6_addr_v4mapped(&sin6->sin6_addr))
 		return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
 				      AF_INET, prefixlen, l3index, flags,
-- 
2.33.1