[PATCH] tcp: Fix build against musl, __sum16 comes from linux/types.h
...so we need to include that header explicitly. For some reason, it
works without it when building against glibc.
Reported-by: omni
On Wed, Oct 30, 2024 at 09:50:10PM +0100, Stefano Brivio wrote:
...so we need to include that header explicitly. For some reason, it works without it when building against glibc.
Presumably because glibc indirectly imports linux/types.h at some point. But, I think there's a better way to fix this. We use __sum16 because it's the type of the checksum field in the linux/tcp.h version of struct tcphdr. But we've recently changed to using the netinet/tcp.h version, which just uses a plain uint16_t. We should change to match, and this will go away.
Reported-by: omni
Fixes: 3d484aa37090 ("tcp: Update TCP checksum using an iovec array") Signed-off-by: Stefano Brivio --- tcp.c | 2 ++ 1 file changed, 2 insertions(+) diff --git a/tcp.c b/tcp.c index 4e0a17e..429634a 100644 --- a/tcp.c +++ b/tcp.c @@ -287,6 +287,8 @@ #include
#include +#include
/* For __sum16 */ + #include "checksum.h" #include "util.h" #include "iov.h"
-- David Gibson (he or they) | I'll have my music baroque, and my code david AT gibson.dropbear.id.au | minimalist, thank you, not the other way | around. http://www.ozlabs.org/~dgibson
On Thu, 31 Oct 2024 11:34:38 +1100
David Gibson
On Wed, Oct 30, 2024 at 09:50:10PM +0100, Stefano Brivio wrote:
...so we need to include that header explicitly. For some reason, it works without it when building against glibc.
Presumably because glibc indirectly imports linux/types.h at some point.
But, I think there's a better way to fix this. We use __sum16 because it's the type of the checksum field in the linux/tcp.h version of struct tcphdr. But we've recently changed to using the netinet/tcp.h version, which just uses a plain uint16_t. We should change to match, and this will go away.
True, but I thought we would want to keep the 'bitwise' attribute that comes with __sum16. On the other hand, I see it's only used by sparse(1), not by gcc itself: https://gcc.gnu.org/legacy-ml/gcc-help/2008-03/msg00267.html so it's rather pointless outside the kernel and we can drop it as well. I'll change checksum pointers to uint16_t *. -- Stefano
participants (2)
-
David Gibson
-
Stefano Brivio