Re: fix encode_entryplus_baggage stack usage

[Jeff Layton <jlayton@redhat.com>]

On Thu, 9 Jan 2014 16:33:35 -0500
"J. Bruce Fields" <bfields@fieldses.org> wrote:

> Eric ran across an excessive stack with this (and a bunch of xfs code)
> to blame.  Untested.  I don't know whether it's worth the minor extra
> work here to avoid the alloc/free on each entry.
> 
> nfsd4_encode_fattr will need the same treatment.
> 
> --b.
> 
> commit 5b35a1f99f999555e82420e687e9d079ae796547
> Author: J. Bruce Fields <bfields@redhat.com>
> Date:   Thu Jan 9 16:24:35 2014 -0500
> 
>     nfsd: fix encode_entryplus_baggage stack usage
>     
>     We stick an extra svc_fh in nfsd3_readdirres to save the need to
>     kmalloc, though maybe it would be fine to kmalloc instead.
>     
>     Signed-off-by: J. Bruce Fields <bfields@redhat.com>
> 
> diff --git a/fs/nfsd/nfs3xdr.c b/fs/nfsd/nfs3xdr.c
> index 1ee6bae..de6e39e 100644
> --- a/fs/nfsd/nfs3xdr.c
> +++ b/fs/nfsd/nfs3xdr.c
> @@ -842,21 +842,21 @@ out:
>  
>  static __be32 *encode_entryplus_baggage(struct nfsd3_readdirres *cd, __be32 *p, const char *name, int namlen)
>  {
> -	struct svc_fh	fh;
> +	struct svc_fh	*fh = &cd->scratch;
>  	__be32 err;
>  
> -	fh_init(&fh, NFS3_FHSIZE);
> -	err = compose_entry_fh(cd, &fh, name, namlen);
> +	fh_init(fh, NFS3_FHSIZE);
> +	err = compose_entry_fh(cd, fh, name, namlen);
>  	if (err) {
>  		*p++ = 0;
>  		*p++ = 0;
>  		goto out;
>  	}
> -	p = encode_post_op_attr(cd->rqstp, p, &fh);
> +	p = encode_post_op_attr(cd->rqstp, p, fh);
>  	*p++ = xdr_one;			/* yes, a file handle follows */
> -	p = encode_fh(p, &fh);
> +	p = encode_fh(p, fh);
>  out:
> -	fh_put(&fh);
> +	fh_put(fh);
>  	return p;
>  }
>  
> diff --git a/fs/nfsd/xdr3.h b/fs/nfsd/xdr3.h
> index b6d5542..335e04a 100644
> --- a/fs/nfsd/xdr3.h
> +++ b/fs/nfsd/xdr3.h
> @@ -174,6 +174,9 @@ struct nfsd3_linkres {
>  struct nfsd3_readdirres {
>  	__be32			status;
>  	struct svc_fh		fh;
> +	/* Just to save kmalloc on every readdirplus entry (svc_fh is a
> +	 * little large for the stack): */
> +	struct svc_fh		scratch;
>  	int			count;
>  	__be32			verf[2];
>  

I think this is probably the best solution...

Acked-by: Jeff Layton <jlayton@redhat.com>