[PATCH Liburing 0/2] build and compiler warning fixes

[PATCH Liburing 0/2] build and compiler warning fixes

[Yang Xiuwei]

Fix two issues: (1) cbpf_filter test fails to build when kernel headers
lack openat2.h; (2) send-zerocopy triggers -Wstringop-truncation on
ifr.ifr_name.

Yang Xiuwei (2):
  examples/send-zerocopy: fix -Wstringop-truncation on ifr.ifr_name
  test/cbpf_filter: skip when openat2.h is not available

 examples/send-zerocopy.c |  2 +-
 test/cbpf_filter.c      |  8 ++++++++
 2 files changed, 9 insertions(+), 1 deletion(-)

-- 
2.25.1

[PATCH 1/2] examples/send-zerocopy: fix -Wstringop-truncation on ifr.ifr_name

[Yang Xiuwei]

strncpy(ifr.ifr_name, cfg_ifname, sizeof(ifr.ifr_name)) triggers
-Wstringop-truncation because the bound equals the destination size,
so the compiler assumes the result may not be null-terminated. Use
sizeof(ifr.ifr_name) - 1 as the bound so at most 15 bytes are copied.

Fixes: 3e4f05342662 ("examples/send-zerocopy: use strncpy() to copy interface name")
Signed-off-by: Yang Xiuwei <yangxiuwei@kylinos.cn>
---
 examples/send-zerocopy.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/examples/send-zerocopy.c b/examples/send-zerocopy.c
index f0248d20..b2821af6 100644
--- a/examples/send-zerocopy.c
+++ b/examples/send-zerocopy.c
@@ -353,7 +353,7 @@ static void do_tx(struct thread_data *td, int domain, int type, int protocol)
 		struct ifreq ifr;
 
 		memset(&ifr, 0, sizeof(ifr));
-		strncpy(ifr.ifr_name, cfg_ifname, sizeof(ifr.ifr_name));
+		strncpy(ifr.ifr_name, cfg_ifname, sizeof(ifr.ifr_name) - 1);
 
 		if (setsockopt(fd, SOL_SOCKET, SO_BINDTODEVICE, &ifr, sizeof(ifr)) < 0)
 			t_error(1, errno, "Binding to device failed\n");
-- 
2.25.1

[PATCH 2/2] test/cbpf_filter: skip when openat2.h is not available

[Yang Xiuwei]

cbpf_filter.c unconditionally includes <linux/openat2.h>, so the test
fails to build on systems whose kernel headers do not provide that
file (e.g. older distros or LTS). configure already sets
CONFIG_HAVE_OPEN_HOW=n and provides struct open_how in compat.h for
the library and other tests; only this test bypassed that by
including the kernel header directly.

Wrap the entire test in #ifdef CONFIG_HAVE_OPEN_HOW and add a stub
main that returns T_EXIT_SKIP in the #else branch. The test then
always compiles; on systems without openat2.h it skips at runtime.

Signed-off-by: Yang Xiuwei <yangxiuwei@kylinos.cn>
---
 test/cbpf_filter.c | 8 ++++++++
 1 file changed, 8 insertions(+)

diff --git a/test/cbpf_filter.c b/test/cbpf_filter.c
index 41fd284e..050b3ac7 100644
--- a/test/cbpf_filter.c
+++ b/test/cbpf_filter.c
@@ -6,6 +6,7 @@
  * Unlike eBPF which requires a separate compiled program, cBPF filters can
  * be defined inline as an array of sock_filter instructions.
  */
+#ifdef CONFIG_HAVE_OPEN_HOW
 #include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
@@ -1443,3 +1444,10 @@ int main(int argc, char *argv[])
 
 	return total_failed;
 }
+#else /* #ifdef CONFIG_HAVE_OPEN_HOW */
+#include "helpers.h"
+int main(void)
+{
+	return T_EXIT_SKIP;
+}
+#endif /* #ifdef CONFIG_HAVE_OPEN_HOW */
-- 
2.25.1

Re: [PATCH 2/2] test/cbpf_filter: skip when openat2.h is not available

[Jens Axboe]

On 3/14/26 2:35 AM, Yang Xiuwei wrote:
> cbpf_filter.c unconditionally includes <linux/openat2.h>, so the test
> fails to build on systems whose kernel headers do not provide that
> file (e.g. older distros or LTS). configure already sets
> CONFIG_HAVE_OPEN_HOW=n and provides struct open_how in compat.h for
> the library and other tests; only this test bypassed that by
> including the kernel header directly.
> 
> Wrap the entire test in #ifdef CONFIG_HAVE_OPEN_HOW and add a stub
> main that returns T_EXIT_SKIP in the #else branch. The test then
> always compiles; on systems without openat2.h it skips at runtime.

liburing defines open_how if it's not in the system headers. I feel
like all you need to do here is remove the openat2.h include, rather
than disable the test entirely?

-- 
Jens Axboe

Re: [PATCH 2/2] test/cbpf_filter: skip when openat2.h is not available

[Yang Xiuwei]

Hi Jens,

On 2026-03-14 13:35 UTC, Jens Axboe wrote:
> liburing defines open_how if it's not in the system headers. I feel
> like all you need to do here is remove the openat2.h include, rather
> than disable the test entirely?

Thanks for the suggestion. I had actually tried that first: removing only the
#include <linux/openat2.h> and relying on liburing's compat for struct open_how.
It turned out that this test also uses the RESOLVE_IN_ROOT macro (and the
filter logic is built around it). Without the header, the build fails with
'RESOLVE_IN_ROOT' undeclared. RESOLVE_IN_ROOT and the other RESOLVE_* flags
come from <linux/openat2.h>; compat.h only provides struct open_how when the
header is missing, not those constants.

Defining RESOLVE_IN_ROOT (and friends) ourselves in compat or in the test
would duplicate kernel UAPI and could get out of sync if the kernel ever
changes them. So I went back to the more conservative approach: wrap the
test in #ifdef CONFIG_HAVE_OPEN_HOW and provide a stub main that returns
T_EXIT_SKIP when openat2.h is not available. The test then always compiles;
on systems without the header it simply skips at runtime.

If you have a better approach in mind, I'd be glad to follow that instead.

Best regards,
Yang Xiuwei

Re: [PATCH 2/2] test/cbpf_filter: skip when openat2.h is not available

[Jens Axboe]

On 3/14/26 11:02 PM, Yang Xiuwei wrote:
> Hi Jens,
> 
> On 2026-03-14 13:35 UTC, Jens Axboe wrote:
>> liburing defines open_how if it's not in the system headers. I feel
>> like all you need to do here is remove the openat2.h include, rather
>> than disable the test entirely?
> 
> Thanks for the suggestion. I had actually tried that first: removing only the
> #include <linux/openat2.h> and relying on liburing's compat for struct open_how.
> It turned out that this test also uses the RESOLVE_IN_ROOT macro (and the
> filter logic is built around it). Without the header, the build fails with
> 'RESOLVE_IN_ROOT' undeclared. RESOLVE_IN_ROOT and the other RESOLVE_* flags
> come from <linux/openat2.h>; compat.h only provides struct open_how when the
> header is missing, not those constants.
> 
> Defining RESOLVE_IN_ROOT (and friends) ourselves in compat or in the test
> would duplicate kernel UAPI and could get out of sync if the kernel ever
> changes them. So I went back to the more conservative approach: wrap the
> test in #ifdef CONFIG_HAVE_OPEN_HOW and provide a stub main that returns
> T_EXIT_SKIP when openat2.h is not available. The test then always compiles;
> on systems without the header it simply skips at runtime.
> 
> If you have a better approach in mind, I'd be glad to follow that instead.

I think just defining RESOLVE_IN_ROOT if it's not available should be
fine. And yes the test will always compile when you stub everything out,
but it also won't do anything at all. This will prevent running this
test case on a host with old headers, but with a current kernel. How
about the below, hopefully that should do it. That'll keep the test
functional, rather than wrap it all in a define that just disables it
entirely.

diff --git a/test/cbpf_filter.c b/test/cbpf_filter.c
index 41fd284e4434..b80b15033662 100644
--- a/test/cbpf_filter.c
+++ b/test/cbpf_filter.c
@@ -15,12 +15,15 @@
 #include <sys/wait.h>
 #include <sys/prctl.h>
 #include <linux/filter.h>
-#include <linux/openat2.h>
 
 #include "liburing.h"
 #include "liburing/io_uring/bpf_filter.h"
 #include "helpers.h"
 
+#ifndef RESOLVE_IN_ROOT
+#define RESOLVE_IN_ROOT	0x10
+#endif
+
 /*
  * cBPF filter context layout (struct io_uring_bpf_ctx):
  *   offset 0:  user_data (u64)

-- 
Jens Axboe

Re: [PATCH 2/2] test/cbpf_filter: skip when openat2.h is not available

[Yang Xiuwei]

Hi Jens,

On 2026-03-15 14:45 UTC, Jens Axboe wrote:
> I think just defining RESOLVE_IN_ROOT if it's not available should be
> fine. And yes the test will always compile when you stub everything out,
> but it also won't do anything at all. This will prevent running this
> test case on a host with old headers, but with a current kernel. How
> about the below, hopefully that should do it. That'll keep the test
> functional, rather than wrap it all in a define that just disables it
> entirely.
>
> diff --git a/test/cbpf_filter.c b/test/cbpf_filter.c
> ...
> +#ifndef RESOLVE_IN_ROOT
> +#define RESOLVE_IN_ROOT	0x10
> +#endif

Thanks, that works well for me. I'll send v2 with your approach.

Best regards,
Yang Xiuwei