[PATCH liburing v2 0/3] Bring back `CONFIG_HAVE_MEMFD_CREATE` to fix Android build error

[PATCH liburing v2 0/3] Bring back `CONFIG_HAVE_MEMFD_CREATE` to fix Android build error

[Alviro Iskandar Setiawan]

Hello,

This is the v2 revision of the patch series to address the Android
build error related to `memfd_create()`. The series consists of three
patches:

1) Bring back `CONFIG_HAVE_MEMFD_CREATE` to fix Android build error.

Partially reverts commit:

  732bf609b670 ("test/io_uring_register: kill old memfd test")

to bring back `CONFIG_HAVE_MEMFD_CREATE` to resolve Android build
failures caused by:

  93d3a7a70b4a ("examples/zcrx: udmabuf backed areas")

It added a call to `memfd_create()`, which is unavailable on some
Android toolchains, leading to the following build error:
```
  zcrx.c:111:10: error: call to undeclared function 'memfd_create'; ISO C99 and \
  later do not support implicit function declarations \
  [-Wimplicit-function-declaration]
    111 |         memfd = memfd_create("udmabuf-test", MFD_ALLOW_SEALING);
        |                 ^
```
This reversion is a preparation patch for a proper fix by ensuring
`memfd_create()` usage is guarded and portable. It's only a partial
revert because the test itself is not restored.

2) Move `memfd_create()` to helpers.c, make it accessible for all tests.

Previously, the static definition of `memfd_create()` was limited to
io_uring_register.c. Now, promote it to a shared location accessible
to all test cases, ensuring that future tests using `memfd_create()`
do not trigger build failures on Android targets where the syscall
is undefined in the standard headers. It improves portability and
prevents regressions across test builds.

3) Also, add `memfd_create()` helper in the examples directory.

Changelog:
v1 -> v2:
  - Omit the old memfd test because it's not needed anymore (Jens Axboe)
    Link: https://lore.kernel.org/io-uring/4bc75566-9cb5-42ec-a6b7-16e04062e0c6@kernel.dk

Signed-off-by: Ammar Faizi <ammarfaizi2@gnuweeb.org>
Signed-off-by: Alviro Iskandar Setiawan <alviro.iskandar@gnuweeb.org>
---

Alviro Iskandar Setiawan (3):
  Bring back `CONFIG_HAVE_MEMFD_CREATE` to fix Android build errors
  test: Move `memfd_create()` to helpers.c, make it accessible for all tests
  examples: Add `memfd_create()` helper

 configure          | 19 +++++++++++++++++++
 examples/helpers.c |  8 ++++++++
 examples/helpers.h |  5 +++++
 test/helpers.c     |  8 ++++++++
 test/helpers.h     |  5 +++++
 5 files changed, 45 insertions(+)


base-commit: 0272bfa96f02cc47c024ec510a764ef7e37b76bf
-- 
Alviro Iskandar Setiawan

[PATCH liburing v2 1/3] Bring back `CONFIG_HAVE_MEMFD_CREATE` to fix Android build error

[Alviro Iskandar Setiawan]

This partially reverts commit:

  732bf609b670 ("test/io_uring_register: kill old memfd test")

Bring back `CONFIG_HAVE_MEMFD_CREATE` to resolve Android build failures
caused by:

  93d3a7a70b4a ("examples/zcrx: udmabuf backed areas")

It added a call to `memfd_create()`, which is unavailable on some
Android toolchains, leading to the following build error:

```
  zcrx.c:111:10: error: call to undeclared function 'memfd_create'; ISO C99 and \
  later do not support implicit function declarations \
  [-Wimplicit-function-declaration]
    111 |         memfd = memfd_create("udmabuf-test", MFD_ALLOW_SEALING);
        |                 ^
```

This reversion is a preparation patch for a proper fix by ensuring
`memfd_create()` usage is guarded and portable. It's only a partial
revert because the test itself is not restored.

v1 -> v2:
  - Omit the old memfd test because it's not needed anymore.
    Link: https://lore.kernel.org/io-uring/4bc75566-9cb5-42ec-a6b7-16e04062e0c6@kernel.dk

Co-authored-by: Ammar Faizi <ammarfaizi2@gnuweeb.org>
Signed-off-by: Ammar Faizi <ammarfaizi2@gnuweeb.org>
Signed-off-by: Alviro Iskandar Setiawan <alviro.iskandar@gnuweeb.org>
---
 configure                | 19 +++++++++++++++++++
 test/io_uring_register.c | 11 +++++++++++
 2 files changed, 30 insertions(+)

diff --git a/configure b/configure
index 552f8ae..3c95214 100755
--- a/configure
+++ b/configure
@@ -379,6 +379,22 @@ if compile_prog "" "" "has_ucontext"; then
 fi
 print_config "has_ucontext" "$has_ucontext"
 
+##########################################
+# check for memfd_create(2)
+has_memfd_create="no"
+cat > $TMPC << EOF
+#include <sys/mman.h>
+int main(int argc, char **argv)
+{
+  int memfd = memfd_create("test", 0);
+  return 0;
+}
+EOF
+if compile_prog "-Werror=implicit-function-declaration" "" "has_memfd_create"; then
+  has_memfd_create="yes"
+fi
+print_config "has_memfd_create" "$has_memfd_create"
+
 ##########################################
 # Check NVME_URING_CMD support
 nvme_uring_cmd="no"
@@ -539,6 +555,9 @@ fi
 if test "$array_bounds" = "yes"; then
   output_sym "CONFIG_HAVE_ARRAY_BOUNDS"
 fi
+if test "$has_memfd_create" = "yes"; then
+  output_sym "CONFIG_HAVE_MEMFD_CREATE"
+fi
 if test "$nvme_uring_cmd" = "yes"; then
   output_sym "CONFIG_HAVE_NVME_URING"
 fi
diff --git a/test/io_uring_register.c b/test/io_uring_register.c
index b53a67d..405a812 100644
--- a/test/io_uring_register.c
+++ b/test/io_uring_register.c
@@ -32,6 +32,17 @@ static int pagesize;
 static rlim_t mlock_limit;
 static int devnull;
 
+#if !defined(CONFIG_HAVE_MEMFD_CREATE)
+#include <sys/syscall.h>
+#include <linux/memfd.h>
+
+int memfd_create(const char *name, unsigned int flags)
+{
+	return (int)syscall(SYS_memfd_create, name, flags);
+}
+#endif
+
+
 static int expect_fail(int fd, unsigned int opcode, void *arg,
 		       unsigned int nr_args, int error, int error2)
 {
-- 
Alviro Iskandar Setiawan

[PATCH liburing v2 2/3] test: Move `memfd_create()` to helpers.c, make it accessible for all tests

[Alviro Iskandar Setiawan]

Previously, the static definition of `memfd_create()` was limited to
io_uring_register.c. Now, promote it to a shared location accessible
to all test cases, ensuring that future tests using `memfd_create()`
do not trigger build failures on Android targets where the syscall
is undefined in the standard headers. It improves portability and
prevents regressions across test builds.

Co-authored-by: Ammar Faizi <ammarfaizi2@gnuweeb.org>
Signed-off-by: Ammar Faizi <ammarfaizi2@gnuweeb.org>
Signed-off-by: Alviro Iskandar Setiawan <alviro.iskandar@gnuweeb.org>
---
 test/helpers.c           |  8 ++++++++
 test/helpers.h           |  5 +++++
 test/io_uring_register.c | 11 -----------
 3 files changed, 13 insertions(+), 11 deletions(-)

diff --git a/test/helpers.c b/test/helpers.c
index 0589548..f1d4477 100644
--- a/test/helpers.c
+++ b/test/helpers.c
@@ -18,6 +18,14 @@
 #include "helpers.h"
 #include "liburing.h"
 
+#ifndef CONFIG_HAVE_MEMFD_CREATE
+#include <sys/syscall.h>
+int memfd_create(const char *name, unsigned int flags)
+{
+	return (int)syscall(SYS_memfd_create, name, flags);
+}
+#endif
+
 /*
  * Helper for allocating memory in tests.
  */
diff --git a/test/helpers.h b/test/helpers.h
index 3f0c11a..8ab0920 100644
--- a/test/helpers.h
+++ b/test/helpers.h
@@ -122,6 +122,11 @@ unsigned long long mtime_since_now(struct timeval *tv);
 unsigned long long utime_since(const struct timeval *s, const struct timeval *e);
 unsigned long long utime_since_now(struct timeval *tv);
 
+#ifndef CONFIG_HAVE_MEMFD_CREATE
+#include <linux/memfd.h>
+#endif
+int memfd_create(const char *name, unsigned int flags);
+
 #ifdef __cplusplus
 }
 #endif
diff --git a/test/io_uring_register.c b/test/io_uring_register.c
index 405a812..b53a67d 100644
--- a/test/io_uring_register.c
+++ b/test/io_uring_register.c
@@ -32,17 +32,6 @@ static int pagesize;
 static rlim_t mlock_limit;
 static int devnull;
 
-#if !defined(CONFIG_HAVE_MEMFD_CREATE)
-#include <sys/syscall.h>
-#include <linux/memfd.h>
-
-int memfd_create(const char *name, unsigned int flags)
-{
-	return (int)syscall(SYS_memfd_create, name, flags);
-}
-#endif
-
-
 static int expect_fail(int fd, unsigned int opcode, void *arg,
 		       unsigned int nr_args, int error, int error2)
 {
-- 
Alviro Iskandar Setiawan

[PATCH liburing v2 3/3] examples: Add `memfd_create()` helper

[Alviro Iskandar Setiawan]

Add `memfd_create()` helper to handle missing defintion on an environment
where `CONFIG_HAVE_MEMFD_CREATE` is not defined.

Fixes: 93d3a7a70b4a ("examples/zcrx: udmabuf backed areas")
Co-authored-by: Ammar Faizi <ammarfaizi2@gnuweeb.org>
Signed-off-by: Ammar Faizi <ammarfaizi2@gnuweeb.org>
Signed-off-by: Alviro Iskandar Setiawan <alviro.iskandar@gnuweeb.org>
---
 examples/helpers.c | 8 ++++++++
 examples/helpers.h | 5 +++++
 2 files changed, 13 insertions(+)

diff --git a/examples/helpers.c b/examples/helpers.c
index 483ddee..8c112f1 100644
--- a/examples/helpers.c
+++ b/examples/helpers.c
@@ -13,6 +13,14 @@
 
 #include "helpers.h"
 
+#ifndef CONFIG_HAVE_MEMFD_CREATE
+#include <sys/syscall.h>
+int memfd_create(const char *name, unsigned int flags)
+{
+	return (int)syscall(SYS_memfd_create, name, flags);
+}
+#endif
+
 int setup_listening_socket(int port, int ipv6)
 {
 	struct sockaddr_in srv_addr = { };
diff --git a/examples/helpers.h b/examples/helpers.h
index 44543e1..0b6f15f 100644
--- a/examples/helpers.h
+++ b/examples/helpers.h
@@ -17,4 +17,9 @@ void *t_aligned_alloc(size_t alignment, size_t size);
 
 void t_error(int status, int errnum, const char *format, ...);
 
+#ifndef CONFIG_HAVE_MEMFD_CREATE
+#include <linux/memfd.h>
+#endif
+int memfd_create(const char *name, unsigned int flags);
+
 #endif
-- 
Alviro Iskandar Setiawan

Re: [PATCH liburing v2 0/3] Bring back `CONFIG_HAVE_MEMFD_CREATE` to fix Android build error

[Jens Axboe]

On 7/15/25 6:43 PM, Alviro Iskandar Setiawan wrote:
> Hello,
> 
> This is the v2 revision of the patch series to address the Android
> build error related to `memfd_create()`. The series consists of three
> patches:

Took a closer look at this. A few comments:

For patch 1, maybe just bring back the configure test and not bother
with a revert style commit? There is nothing in test/ that uses
memfd_create, so there's no point bringing it back in there.

IOW, patch 2 can be dropped, as it's really just dropping bits
that patch 1 re-added for some reason.

All that's needed is to add it to the examples/ helpers. If it's
needed for test/ later, then it can get added at that time.

All of that to say, I'd just add the configure bit and the examples/
helper in a single patch and not worry about test/ at all.

-- 
Jens Axboe

Re: [PATCH liburing v2 0/3] Bring back `CONFIG_HAVE_MEMFD_CREATE` to fix Android build error

[Alviro Iskandar Setiawan]

On Wed, Jul 16, 2025 at 7:41 PM Jens Axboe wrote:
> For patch 1, maybe just bring back the configure test and not bother
> with a revert style commit? There is nothing in test/ that uses
> memfd_create, so there's no point bringing it back in there.

Ah yea. That'd be easier. I'll copy the configure part instead of
modifying the git revert result 😆

> IOW, patch 2 can be dropped, as it's really just dropping bits
> that patch 1 re-added for some reason.
>
> All that's needed is to add it to the examples/ helpers. If it's
> needed for test/ later, then it can get added at that time.
>
> All of that to say, I'd just add the configure bit and the examples/
> helper in a single patch and not worry about test/ at all.

Understandable. I'll send a v3 revision shortly.

-- Viro

Re: [PATCH liburing v2 0/3] Bring back `CONFIG_HAVE_MEMFD_CREATE` to fix Android build error

[Jens Axboe]

On 7/16/25 8:28 AM, Alviro Iskandar Setiawan wrote:
> On Wed, Jul 16, 2025 at 7:41?PM Jens Axboe wrote:
>> For patch 1, maybe just bring back the configure test and not bother
>> with a revert style commit? There is nothing in test/ that uses
>> memfd_create, so there's no point bringing it back in there.
> 
> Ah yea. That'd be easier. I'll copy the configure part instead of
> modifying the git revert result ?

Exactly, just make it that single patch.

>> IOW, patch 2 can be dropped, as it's really just dropping bits
>> that patch 1 re-added for some reason.
>>
>> All that's needed is to add it to the examples/ helpers. If it's
>> needed for test/ later, then it can get added at that time.
>>
>> All of that to say, I'd just add the configure bit and the examples/
>> helper in a single patch and not worry about test/ at all.
> 
> Understandable. I'll send a v3 revision shortly.

Thanks!

-- 
Jens Axboe