tcp: drop the hash_32() part from the index calculation
authorWilly Tarreau <w@1wt.eu>
Mon, 2 May 2022 08:46:14 +0000 (10:46 +0200)
committerJakub Kicinski <kuba@kernel.org>
Thu, 5 May 2022 02:22:33 +0000 (19:22 -0700)
In commit 190cc82489f4 ("tcp: change source port randomizarion at
connect() time"), the table_perturb[] array was introduced and an
index was taken from the port_offset via hash_32(). But it turns
out that hash_32() performs a multiplication while the input here
comes from the output of SipHash in secure_seq, that is well
distributed enough to avoid the need for yet another hash.

Suggested-by: Amit Klein <aksecurity@gmail.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Willy Tarreau <w@1wt.eu>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
net/ipv4/inet_hashtables.c

index cc5f66328b47428eec19e145a357511c29df59bd..a5d57fa679caa47ec31ea4b1de3c45f93be4cd13 100644 (file)
@@ -778,7 +778,7 @@ int __inet_hash_connect(struct inet_timewait_death_row *death_row,
 
        net_get_random_once(table_perturb,
                            INET_TABLE_PERTURB_SIZE * sizeof(*table_perturb));
-       index = hash_32(port_offset, INET_TABLE_PERTURB_SHIFT);
+       index = port_offset & (INET_TABLE_PERTURB_SIZE - 1);
 
        offset = READ_ONCE(table_perturb[index]) + (port_offset >> 32);
        offset %= remaining;