[v2] linux-user: allow null `pathname` for statx()/fstatat()

[PATCH v2] linux-user: allow null `pathname` for statx()/fstatat()

Posted by Jean-Christian CÎRSTEA 1 month, 2 weeks ago

Since Linux 6.11, statx() and fstatat() syscalls accept a null `pathname`.

Before this patch, qemu-*-linux-user failed with EFAULT when `pathname` was
specified as NULL, even for Linux kernel hosts > 6.10. This patch fixes this
issue by checking whether `arg2` is 0. If so, don't return EFAULT, but instead
perform the appropiate syscall and let the host's kernel handle null `pathname`.

Signed-off-by: Jean-Christian CÎRSTEA <jean.christian.cirstea@gmail.com>
---
 linux-user/syscall.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/linux-user/syscall.c b/linux-user/syscall.c
index 2060e561a2..e1b61f6dc5 100644
--- a/linux-user/syscall.c
+++ b/linux-user/syscall.c
@@ -12142,7 +12142,8 @@ static abi_long do_syscall1(CPUArchState *cpu_env, int num, abi_long arg1,
             int flags = arg3;
 
             p = lock_user_string(arg2);
-            if (p == NULL) {
+            /* Since Linux 6.11, the path argument may be NULL */
+            if (arg2 != 0 && p == NULL) {
                 return -TARGET_EFAULT;
             }
 #if defined(__NR_statx)
-- 
2.51.0

Re: [PATCH v2] linux-user: allow null `pathname` for statx()/fstatat()

Posted by Richard Henderson 1 month, 1 week ago

On 12/26/25 02:39, Jean-Christian CÎRSTEA wrote:
> Since Linux 6.11, statx() and fstatat() syscalls accept a null `pathname`.
> 
> Before this patch, qemu-*-linux-user failed with EFAULT when `pathname` was
> specified as NULL, even for Linux kernel hosts > 6.10. This patch fixes this
> issue by checking whether `arg2` is 0. If so, don't return EFAULT, but instead
> perform the appropiate syscall and let the host's kernel handle null `pathname`.
> 
> Signed-off-by: Jean-Christian CÎRSTEA <jean.christian.cirstea@gmail.com>
> ---
>   linux-user/syscall.c | 3 ++-
>   1 file changed, 2 insertions(+), 1 deletion(-)
> 
> diff --git a/linux-user/syscall.c b/linux-user/syscall.c
> index 2060e561a2..e1b61f6dc5 100644
> --- a/linux-user/syscall.c
> +++ b/linux-user/syscall.c
> @@ -12142,7 +12142,8 @@ static abi_long do_syscall1(CPUArchState *cpu_env, int num, abi_long arg1,
>               int flags = arg3;
>   
>               p = lock_user_string(arg2);
> -            if (p == NULL) {
> +            /* Since Linux 6.11, the path argument may be NULL */
> +            if (arg2 != 0 && p == NULL) {
>                   return -TARGET_EFAULT;
>               }
>   #if defined(__NR_statx)

No need to call lock_user_string if arg2 == 0.

Better as

	p = NULL;
	if (arg2) {
	    p = lock_user_string(arg2);
	    if (p == NULL) {
		return -TARGET_EFAULT;
	    }
	}


r~

Re: [PATCH v2] linux-user: allow null `pathname` for statx()/fstatat()

Posted by Laurent Vivier 1 month, 2 weeks ago

Le 25/12/2025 à 16:39, Jean-Christian CÎRSTEA a écrit :
> Since Linux 6.11, statx() and fstatat() syscalls accept a null `pathname`.
> 
> Before this patch, qemu-*-linux-user failed with EFAULT when `pathname` was
> specified as NULL, even for Linux kernel hosts > 6.10. This patch fixes this
> issue by checking whether `arg2` is 0. If so, don't return EFAULT, but instead
> perform the appropiate syscall and let the host's kernel handle null `pathname`.
> 
> Signed-off-by: Jean-Christian CÎRSTEA <jean.christian.cirstea@gmail.com>
> ---
>   linux-user/syscall.c | 3 ++-
>   1 file changed, 2 insertions(+), 1 deletion(-)
> 
> diff --git a/linux-user/syscall.c b/linux-user/syscall.c
> index 2060e561a2..e1b61f6dc5 100644
> --- a/linux-user/syscall.c
> +++ b/linux-user/syscall.c
> @@ -12142,7 +12142,8 @@ static abi_long do_syscall1(CPUArchState *cpu_env, int num, abi_long arg1,
>               int flags = arg3;
>   
>               p = lock_user_string(arg2);
> -            if (p == NULL) {
> +            /* Since Linux 6.11, the path argument may be NULL */
> +            if (arg2 != 0 && p == NULL) {
>                   return -TARGET_EFAULT;
>               }
>   #if defined(__NR_statx)

Reviewed-by: Laurent Vivier <laurent@vivier.eu>