[musl][PATCH 1/2] include/netinet/if_ether.h: define __UAPI_DEF_ETHHDR early

[musl][PATCH 1/2] include/netinet/if_ether.h: define __UAPI_DEF_ETHHDR early

[Qi.Chen]

From: Chen Qi <Qi.Chen@windriver.com>

After 'struct ethhdr' is defined, we should immediately define
__UAPI_DEF_ETHHDR. Because the following lines are header inclusion
lines which might introduce header files checking __UAPI_DEF_ETHHDR.
If we don't define this macro early, we will get redefinition error.

Signed-off-by: Chen Qi <Qi.Chen@windriver.com>
---
 include/netinet/if_ether.h | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)

diff --git a/include/netinet/if_ether.h b/include/netinet/if_ether.h
index 3479f511..b1646f6f 100644
--- a/include/netinet/if_ether.h
+++ b/include/netinet/if_ether.h
@@ -117,6 +117,7 @@ struct ethhdr {
 	uint8_t h_source[ETH_ALEN];
 	uint16_t h_proto;
 };
+#define __UAPI_DEF_ETHHDR       0
 
 #include <net/ethernet.h>
 #include <net/if_arp.h>
@@ -144,6 +145,4 @@ do { \
 	(enaddr)[5] = ((uint8_t *)ipaddr)[3]; \
 } while(0)
 
-#define __UAPI_DEF_ETHHDR       0
-
 #endif
-- 
2.34.1

[musl][PATCH 2/2] include/netinet/if_ether.h: guard ethhdr definition to better work with other header files

[Qi.Chen]

From: Chen Qi <Qi.Chen@windriver.com>

The unconditional definition of 'struct ethhdr' has the implicit
requirement on this header file, that is, it needs to be included
before other header files defining ethhdr.

Signed-off-by: Chen Qi <Qi.Chen@windriver.com>
---
 include/netinet/if_ether.h | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/include/netinet/if_ether.h b/include/netinet/if_ether.h
index b1646f6f..e5a1ebd0 100644
--- a/include/netinet/if_ether.h
+++ b/include/netinet/if_ether.h
@@ -112,12 +112,14 @@
 #define ETH_P_XDSA	0x00F8
 #define ETH_P_MAP	0x00F9
 
+#ifndef __UAPI_DEF_ETHHDR
 struct ethhdr {
 	uint8_t h_dest[ETH_ALEN];
 	uint8_t h_source[ETH_ALEN];
 	uint16_t h_proto;
 };
 #define __UAPI_DEF_ETHHDR       0
+#endif
 
 #include <net/ethernet.h>
 #include <net/if_arp.h>
-- 
2.34.1

Re: [musl][PATCH 2/2] include/netinet/if_ether.h: guard ethhdr definition to better work with other header files

[Rich Felker]

On Tue, Sep 30, 2025 at 10:45:38AM +0800, Qi.Chen@windriver.com wrote:
> From: Chen Qi <Qi.Chen@windriver.com>
> 
> The unconditional definition of 'struct ethhdr' has the implicit
> requirement on this header file, that is, it needs to be included
> before other header files defining ethhdr.
> 
> Signed-off-by: Chen Qi <Qi.Chen@windriver.com>
> ---
>  include/netinet/if_ether.h | 2 ++
>  1 file changed, 2 insertions(+)
> 
> diff --git a/include/netinet/if_ether.h b/include/netinet/if_ether.h
> index b1646f6f..e5a1ebd0 100644
> --- a/include/netinet/if_ether.h
> +++ b/include/netinet/if_ether.h
> @@ -112,12 +112,14 @@
>  #define ETH_P_XDSA	0x00F8
>  #define ETH_P_MAP	0x00F9
>  
> +#ifndef __UAPI_DEF_ETHHDR
>  struct ethhdr {
>  	uint8_t h_dest[ETH_ALEN];
>  	uint8_t h_source[ETH_ALEN];
>  	uint16_t h_proto;
>  };
>  #define __UAPI_DEF_ETHHDR       0
> +#endif
>  
>  #include <net/ethernet.h>
>  #include <net/if_arp.h>
> -- 
> 2.34.1

Defining a macro multiple times with the exact same token sequence as
its definition each time is not an error.

RE: [musl][PATCH 2/2] include/netinet/if_ether.h: guard ethhdr definition to better work with other header files

[Chen, Qi]

Hi Rich,

The problem is about the re-definition of 'struct ethhdr'.

For example, a header also defines 'struct ethhdr' and __UAPI_DEF_ETHHDR, and it's included *before* our <netinet/if_ether.h>, then the unconditional definition of 'struct ethhdr' in this header will be treated as an error.

Some background info:
I was making systemd to compile with musl. I got this re-definition error. I checked the headers and thought that for different headers to work together, the definition of this 'struct ethhdr' needs to be guarded by the same macro.

Regards,
Qi

-----Original Message-----
From: Rich Felker <dalias@libc.org> 
Sent: Tuesday, September 30, 2025 9:34 PM
To: Chen, Qi <Qi.Chen@windriver.com>
Cc: musl@lists.openwall.com
Subject: Re: [musl][PATCH 2/2] include/netinet/if_ether.h: guard ethhdr definition to better work with other header files

On Tue, Sep 30, 2025 at 10:45:38AM +0800, Qi.Chen@windriver.com wrote:
> From: Chen Qi <Qi.Chen@windriver.com>
> 
> The unconditional definition of 'struct ethhdr' has the implicit 
> requirement on this header file, that is, it needs to be included 
> before other header files defining ethhdr.
> 
> Signed-off-by: Chen Qi <Qi.Chen@windriver.com>
> ---
>  include/netinet/if_ether.h | 2 ++
>  1 file changed, 2 insertions(+)
> 
> diff --git a/include/netinet/if_ether.h b/include/netinet/if_ether.h 
> index b1646f6f..e5a1ebd0 100644
> --- a/include/netinet/if_ether.h
> +++ b/include/netinet/if_ether.h
> @@ -112,12 +112,14 @@
>  #define ETH_P_XDSA	0x00F8
>  #define ETH_P_MAP	0x00F9
>  
> +#ifndef __UAPI_DEF_ETHHDR
>  struct ethhdr {
>  	uint8_t h_dest[ETH_ALEN];
>  	uint8_t h_source[ETH_ALEN];
>  	uint16_t h_proto;
>  };
>  #define __UAPI_DEF_ETHHDR       0
> +#endif
>  
>  #include <net/ethernet.h>
>  #include <net/if_arp.h>
> --
> 2.34.1

Defining a macro multiple times with the exact same token sequence as its definition each time is not an error.

Re: [musl][PATCH 2/2] include/netinet/if_ether.h: guard ethhdr definition to better work with other header files

[A. Wilcox]

On Oct 8, 2025, at 21:53, Chen, Qi <Qi.Chen@windriver.com> wrote:
> 
> Hi Rich,
> 
> The problem is about the re-definition of 'struct ethhdr'.
> 
> For example, a header also defines 'struct ethhdr' and __UAPI_DEF_ETHHDR, and it's included *before* our <netinet/if_ether.h>, then the unconditional definition of 'struct ethhdr' in this header will be treated as an error.
> 
> Some background info:
> I was making systemd to compile with musl. I got this re-definition error. I checked the headers and thought that for different headers to work together, the definition of this 'struct ethhdr' needs to be guarded by the same macro.
> 
> Regards,
> Qi


Hi Qi,

You might consider patching systemd to have libc includes instead of using the Linux UAPI headers.  We did this at WTI with this commit: https://code.atwilcox.tech/sphen/scaly/systemd/-/commit/f423321481330c5f2f9158c490162b2e62b4c8bd

Best,
-Anna

--
Anna Wilcox (she/her)
SW Engineering: C++/Rust, DevOps, POSIX, Py/Ruby
Wilcox Technologies Inc.

RE: [musl][PATCH 1/2] include/netinet/if_ether.h: define __UAPI_DEF_ETHHDR early

[Chen, Qi]

ping

-----Original Message-----
From: Chen, Qi <Qi.Chen@windriver.com> 
Sent: Tuesday, September 30, 2025 10:46 AM
To: musl@lists.openwall.com
Cc: Chen, Qi <Qi.Chen@windriver.com>
Subject: [musl][PATCH 1/2] include/netinet/if_ether.h: define __UAPI_DEF_ETHHDR early

From: Chen Qi <Qi.Chen@windriver.com>

After 'struct ethhdr' is defined, we should immediately define __UAPI_DEF_ETHHDR. Because the following lines are header inclusion lines which might introduce header files checking __UAPI_DEF_ETHHDR.
If we don't define this macro early, we will get redefinition error.

Signed-off-by: Chen Qi <Qi.Chen@windriver.com>
---
 include/netinet/if_ether.h | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)

diff --git a/include/netinet/if_ether.h b/include/netinet/if_ether.h index 3479f511..b1646f6f 100644
--- a/include/netinet/if_ether.h
+++ b/include/netinet/if_ether.h
@@ -117,6 +117,7 @@ struct ethhdr {
 	uint8_t h_source[ETH_ALEN];
 	uint16_t h_proto;
 };
+#define __UAPI_DEF_ETHHDR       0
 
 #include <net/ethernet.h>
 #include <net/if_arp.h>
@@ -144,6 +145,4 @@ do { \
 	(enaddr)[5] = ((uint8_t *)ipaddr)[3]; \  } while(0)
 
-#define __UAPI_DEF_ETHHDR       0
-
 #endif
--
2.34.1

Re: [musl][PATCH 2/2] include/netinet/if_ether.h: guard ethhdr definition to better work with other header files

[Rich Felker]

On Thu, Oct 09, 2025 at 02:53:13AM +0000, Chen, Qi wrote:
> Hi Rich,
> 
> The problem is about the re-definition of 'struct ethhdr'.
> 
> For example, a header also defines 'struct ethhdr' and
> __UAPI_DEF_ETHHDR, and it's included *before* our
> <netinet/if_ether.h>, then the unconditional definition of 'struct
> ethhdr' in this header will be treated as an error.
> 
> Some background info:
> I was making systemd to compile with musl. I got this re-definition
> error. I checked the headers and thought that for different headers
> to work together, the definition of this 'struct ethhdr' needs to be
> guarded by the same macro.

It's intentional that we don't support getting definitions for the
libc types from kernel headers. They may have subtle mismatches that
could break things. For example some types in the kernel are
known-wrong, like struct msghdr, and others have members that are the
same size/representation but different type (like long long vs
int64_t).

For the most part conflicting kernel headers shouldn't be used in
userspace. In cases where they should/need to be used, they should be
used either in separate translation units or included after the libc
headers.

This is documented in the original commit which added the __UAPI
macros, 04983f2272382af92eb8f8838964ff944fbb8258.

Rich

Re: [musl][PATCH 2/2] include/netinet/if_ether.h: guard ethhdr definition to better work with other header files

[ChenQi]

On 10/15/25 00:53, Rich Felker wrote:
> On Thu, Oct 09, 2025 at 02:53:13AM +0000, Chen, Qi wrote:
>> Hi Rich,
>>
>> The problem is about the re-definition of 'struct ethhdr'.
>>
>> For example, a header also defines 'struct ethhdr' and
>> __UAPI_DEF_ETHHDR, and it's included *before* our
>> <netinet/if_ether.h>, then the unconditional definition of 'struct
>> ethhdr' in this header will be treated as an error.
>>
>> Some background info:
>> I was making systemd to compile with musl. I got this re-definition
>> error. I checked the headers and thought that for different headers
>> to work together, the definition of this 'struct ethhdr' needs to be
>> guarded by the same macro.
> It's intentional that we don't support getting definitions for the
> libc types from kernel headers. They may have subtle mismatches that
> could break things. For example some types in the kernel are
> known-wrong, like struct msghdr, and others have members that are the
> same size/representation but different type (like long long vs
> int64_t).
>
> For the most part conflicting kernel headers shouldn't be used in
> userspace. In cases where they should/need to be used, they should be
> used either in separate translation units or included after the libc
> headers.
>
> This is documented in the original commit which added the __UAPI
> macros, 04983f2272382af92eb8f8838964ff944fbb8258.
>
> Rich

Thanks a lot for the information.

I'll drop this patch and fix things in systemd recipe in Yocto.

Regards,
Qi