With -Werror=array-bounds, gcc v4.8.x warns that in tcp_sack_remove(), a
selective_acks[] "array subscript is above array bounds".
I don't understand how gcc figures this out, or why we don't see similar
problems many other places, but this is the only fix I can figure out.
Signed-off-by: Bjorn Helgaas <[email protected]>
---
net/ipv4/tcp_input.c | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)
diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c
index 65cf90e063d5..65133b108236 100644
--- a/net/ipv4/tcp_input.c
+++ b/net/ipv4/tcp_input.c
@@ -4047,7 +4047,8 @@ static void tcp_sack_remove(struct tcp_sock *tp)
/* Zap this SACK, by moving forward any other SACKS. */
for (i = this_sack+1; i < num_sacks; i++)
- tp->selective_acks[i-1] = tp->selective_acks[i];
+ if (i < ARRAY_SIZE(tp->selective_acks))
+ tp->selective_acks[i-1] = tp->selective_acks[i];
num_sacks--;
continue;
}
Bjorn Helgaas
> With -Werror=array-bounds, gcc v4.8.x warns that in tcp_sack_remove(), a
> selective_acks[] "array subscript is above array bounds".
>
> I don't understand how gcc figures this out, or why we don't see similar
> problems many other places, but this is the only fix I can figure out.
...
> diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c
> index 65cf90e063d5..65133b108236 100644
> --- a/net/ipv4/tcp_input.c
> +++ b/net/ipv4/tcp_input.c
> @@ -4047,7 +4047,8 @@ static void tcp_sack_remove(struct tcp_sock *tp)
>
> /* Zap this SACK, by moving forward any other SACKS. */
> for (i = this_sack+1; i < num_sacks; i++)
> - tp->selective_acks[i-1] = tp->selective_acks[i];
> + if (i < ARRAY_SIZE(tp->selective_acks))
> + tp->selective_acks[i-1] = tp->selective_acks[i];
> num_sacks--;
> continue;
> }
You really shouldn't add that test every time around the loop.
Try changing the loop so the assignment is:
tp->selective_acks[i] = tp->selective_acks[i + 1];
or the loop test to:
i <= num_sacks - 1;
Or beat up the gcc developers :-)
David
????{.n?+???????+%?????ݶ??w??{.n?+????{??G?????{ay? ʇڙ?,j??f???h?????????z_??(?階?ݢj"???m??????G????????????&???~???iO???z??v?^?m???? ????????I?
On Fri, Apr 11, 2014 at 2:59 AM, David Laight <[email protected]> wrote:
> Bjorn Helgaas
>> With -Werror=array-bounds, gcc v4.8.x warns that in tcp_sack_remove(), a
>> selective_acks[] "array subscript is above array bounds".
>>
>> I don't understand how gcc figures this out, or why we don't see similar
>> problems many other places, but this is the only fix I can figure out.
> ...
>> diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c
>> index 65cf90e063d5..65133b108236 100644
>> --- a/net/ipv4/tcp_input.c
>> +++ b/net/ipv4/tcp_input.c
>> @@ -4047,7 +4047,8 @@ static void tcp_sack_remove(struct tcp_sock *tp)
>>
>> /* Zap this SACK, by moving forward any other SACKS. */
>> for (i = this_sack+1; i < num_sacks; i++)
>> - tp->selective_acks[i-1] = tp->selective_acks[i];
>> + if (i < ARRAY_SIZE(tp->selective_acks))
>> + tp->selective_acks[i-1] = tp->selective_acks[i];
>> num_sacks--;
>> continue;
>> }
>
> You really shouldn't add that test every time around the loop.
>
> Try changing the loop so the assignment is:
> tp->selective_acks[i] = tp->selective_acks[i + 1];
>
> or the loop test to:
> i <= num_sacks - 1;
>
> Or beat up the gcc developers :-)
You're right, that *is* ugly. And it seems that gcc-v4.9.x doesn't
complain about this, and that's good enough for my purposes, so I
withdraw this patch.
Thanks and sorry for the noise,
Bjorn