[PATCH net-next] tcp: use __jhash_final() in inet6_ehashfn()

[PATCH net-next] tcp: use __jhash_final() in inet6_ehashfn()

[Eric Dumazet]

I misread jhash2() implementation.

Last round should use __jhash_final() instead of __jhash_mix().

Using __jhash_mix() here leaves entropy distributed across a, b, and c,
which might lead to incomplete diffusion of the faddr and fport bits
into the bucket index. Replacing this last __jhash_mix() with
__jhash_final() provides the correct avalanche properties
for the returned value in c.

$ scripts/bloat-o-meter -t vmlinux.0 vmlinux
add/remove: 0/0 grow/shrink: 0/1 up/down: 0/-4 (-4)
Function                                     old     new   delta
inet6_ehashfn                                306     302      -4
Total: Before=25155089, After=25155085, chg -0.00%

Fixes: 854587e69ef3 ("tcp: improve inet6_ehashfn() entropy")
Signed-off-by: Eric Dumazet <edumazet@google.com>
---
 net/ipv6/inet6_hashtables.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/net/ipv6/inet6_hashtables.c b/net/ipv6/inet6_hashtables.c
index 109fbf620ad7cde54300b937db58ada0d5c80664..0a781cba4085f2f74b01518cbdb33889900d99ec 100644
--- a/net/ipv6/inet6_hashtables.c
+++ b/net/ipv6/inet6_hashtables.c
@@ -63,7 +63,7 @@ u32 inet6_ehashfn(const struct net *net,
 	a += (__force u32)faddr->s6_addr32[2];
 	b += (__force u32)faddr->s6_addr32[3];
 	c += (__force u32)fport;
-	__jhash_mix(a, b, c);
+	__jhash_final(a, b, c);
 
 	/* Note: We need to add @lport instead of fully hashing it.
 	 * See commits 9544d60a2605 ("inet: change lport contribution
-- 
2.53.0.959.g497ff81fa9-goog

Re: [PATCH net-next] tcp: use __jhash_final() in inet6_ehashfn()

[Kuniyuki Iwashima]

On Sun, Mar 22, 2026 at 6:18 PM Eric Dumazet <edumazet@google.com> wrote:
>
> I misread jhash2() implementation.
>
> Last round should use __jhash_final() instead of __jhash_mix().
>
> Using __jhash_mix() here leaves entropy distributed across a, b, and c,
> which might lead to incomplete diffusion of the faddr and fport bits
> into the bucket index. Replacing this last __jhash_mix() with
> __jhash_final() provides the correct avalanche properties
> for the returned value in c.
>
> $ scripts/bloat-o-meter -t vmlinux.0 vmlinux
> add/remove: 0/0 grow/shrink: 0/1 up/down: 0/-4 (-4)
> Function                                     old     new   delta
> inet6_ehashfn                                306     302      -4
> Total: Before=25155089, After=25155085, chg -0.00%
>
> Fixes: 854587e69ef3 ("tcp: improve inet6_ehashfn() entropy")
> Signed-off-by: Eric Dumazet <edumazet@google.com>

Reviewed-by: Kuniyuki Iwashima <kuniyu@google.com>

Re: [net-next] tcp: use __jhash_final() in inet6_ehashfn()

[Jakub Kicinski]

This is an AI-generated review of your patch. The human sending this
email has considered the AI review valid, or at least plausible.

For local reproduction steps see https://netdev-ai.bots.linux.dev/ai-local.html
---
tcp: use __jhash_final() in inet6_ehashfn()

This commit fixes a bug in inet6_ehashfn() where the last round of
jhash incorrectly used __jhash_mix() instead of __jhash_final(),
which left entropy distributed incompletely and didn't provide proper
avalanche properties.

> diff --git a/net/ipv6/inet6_hashtables.c b/net/ipv6/inet6_hashtables.c
> index 109fbf620ad7..0a781cba4085 100644
> --- a/net/ipv6/inet6_hashtables.c
> +++ b/net/ipv6/inet6_hashtables.c
> @@ -63,7 +63,7 @@ u32 inet6_ehashfn(const struct net *net,
>  	a += (__force u32)faddr->s6_addr32[2];
>  	b += (__force u32)faddr->s6_addr32[3];
>  	c += (__force u32)fport;
> -	__jhash_mix(a, b, c);
> +	__jhash_final(a, b, c);
>
>  	/* Note: We need to add @lport instead of fully hashing it.
>  	 * See commits 9544d60a2605 ("inet: change lport contribution

This isn't a bug, but the comment earlier in the function needs updating.
The comment currently says:

	/*
	 * Please look at jhash() implementation for reference.
	 * Hash laddr + faddr + lport/fport + net_hash_mix.
	 * Notes:
	 * We combine laddr[0] (high order 32 bits of local address)
	 * with net_hash_mix() to avoid using __jhash_final(a, b, c).
	 *

The line "to avoid using __jhash_final(a, b, c)" is now misleading since
the code was just changed to use __jhash_final() in the last round. Should
this comment be updated to explain that we combine laddr[0] with
net_hash_mix() to avoid an extra __jhash_mix() round, and that the final
round correctly uses __jhash_final() for proper avalanche properties?
-- 
pw-bot: cr