Re: [PATCH v1 1/2] fstress: add suport for using liburing setxattr

[Dave Chinner <david@fromorbit.com>]

On Mon, Nov 29, 2021 at 02:15:41PM -0800, Stefan Roesch wrote:
> Summary:
> 
> Liburing added support for setxattr. This change adds
> support for this this in fsstress when fsstress is built
> with liburing support.
> 
> Signed-off-by: Stefan Roesch <shr@fb.com>
> ---
>  ltp/fsstress.c | 43 ++++++++++++++++++++++++++++++++++++++++++-
>  1 file changed, 42 insertions(+), 1 deletion(-)
> 
> diff --git a/ltp/fsstress.c b/ltp/fsstress.c
> index 003e0e49..4a5c4afe 100644
> --- a/ltp/fsstress.c
> +++ b/ltp/fsstress.c
> @@ -4779,6 +4779,43 @@ setattr_f(opnum_t opno, long r)
>  	close(fd);
>  }
>  
> +static int
> +io_uring_setxattr(const char *path, const char *name, const void *value, size_t size,
> +	  	  int flags)
> +{
> +	struct io_uring_sqe *sqe;
> +	struct io_uring_cqe *cqe;
> +	int ret;
> +
> +	sqe = io_uring_get_sqe(&ring);
> +	if (!sqe) {
> +		printf("io_uring_get_sqe failed\n");
> +		ret = -1;
> +		goto out;
> +	}
> +
> +	io_uring_prep_setxattr(sqe, name, value, path, flags, size);
> +
> +	ret = io_uring_submit_and_wait(&ring, 1);
> +	if (ret != 1) {
> +		printf("io_uring_submit_and_wait failed, ret=%d\n", ret);
> +		ret = -1;
> +		goto out;
> +    	}
> +
> +	ret = io_uring_wait_cqe(&ring, &cqe);
> +	if (ret < 0) {
> +		printf("io_uring_wait_cqe failed, ret=%d\n", ret);
> +		goto out;
> +	}
> +
> +	ret = cqe->res;
> +	io_uring_cqe_seen(&ring, cqe);
> +
> +out:
> +	return ret;
> +}
> +
>  void
>  setfattr_f(opnum_t opno, long r)
>  {
> @@ -4842,7 +4879,11 @@ setfattr_f(opnum_t opno, long r)
>  		goto out;
>  	}
>  
> -	e = setxattr(f.path, name, value, value_len, flag) < 0 ? errno : 0;
> +	if (have_io_uring)
> +		e = io_uring_setxattr(f.path, name, value, value_len, flag);
> +	else
> +		e = setxattr(f.path, name, value, value_len, flag) < 0 ? errno : 0;

While this is technically correct, it is architecturally wrong.
This replaces testing of the existing setxattr() syscall path on
systems that have io_uring enabled (which is most modern, upstream
test instances). That's a significant regression in test coverage,
especially given that most applications using xattrs do not use
io_uring...

The io_uring functionality should be added in the same way that
OP_URING_READ/OP_URING_WRITE were added. That is, new
operations were added in addition to the existing syscall based
OP_READ/OP_WRITE, OP_READV/OP_WRITEV and the AIO based versions
OP_AREAD/OP_AWRITE.

This way fsstress adds the io_uring mechanisms in addition to all
the normal syscall methods it already uses rather than replacing
them. This also allow io_uring operations to race with existing
syscall operations running the same operations on the same files
concurrently...

Cheers,

Dave.
-- 
Dave Chinner
david@fromorbit.com