2014-08-07 03:10:46

by Kinglong Mee

[permalink] [raw]
Subject: [PATCH 1/2] NFSD: Treat rq_reserved when initializing xdr_buf

After calling svc_reserve() with a small length,
nfsd does't treat it in svcxdr_init_encode().

So, when testing conflock, got message as,

[ 970.127216] RPC request reserved 88 but used 116

Signed-off-by: Kinglong Mee <[email protected]>
---
fs/nfsd/nfs4proc.c | 7 +++++--
1 file changed, 5 insertions(+), 2 deletions(-)

diff --git a/fs/nfsd/nfs4proc.c b/fs/nfsd/nfs4proc.c
index 5e0dc52..084e46e 100644
--- a/fs/nfsd/nfs4proc.c
+++ b/fs/nfsd/nfs4proc.c
@@ -1245,17 +1245,20 @@ static void svcxdr_init_encode(struct svc_rqst *rqstp,
struct xdr_stream *xdr = &resp->xdr;
struct xdr_buf *buf = &rqstp->rq_res;
struct kvec *head = buf->head;
+ int tlen = 0;

xdr->buf = buf;
xdr->iov = head;
xdr->p = head->iov_base + head->iov_len;
- xdr->end = head->iov_base + PAGE_SIZE - rqstp->rq_auth_slack;
+ tlen = PAGE_SIZE - rqstp->rq_auth_slack;
+ xdr->end = head->iov_base + min(tlen, rqstp->rq_reserved);
/* Tail and page_len should be zero at this point: */
buf->len = buf->head[0].iov_len;
xdr->scratch.iov_len = 0;
xdr->page_ptr = buf->pages - 1;
- buf->buflen = PAGE_SIZE * (1 + rqstp->rq_page_end - buf->pages)
+ tlen = PAGE_SIZE * (1 + rqstp->rq_page_end - buf->pages)
- rqstp->rq_auth_slack;
+ buf->buflen = min(tlen, rqstp->rq_reserved);
}

/*
--
1.9.3



2014-08-11 19:15:51

by J. Bruce Fields

[permalink] [raw]
Subject: Re: [PATCH 1/2] NFSD: Treat rq_reserved when initializing xdr_buf

On Thu, Aug 07, 2014 at 11:10:31AM +0800, Kinglong Mee wrote:
> After calling svc_reserve() with a small length,
> nfsd does't treat it in svcxdr_init_encode().
>
> So, when testing conflock, got message as,
>
> [ 970.127216] RPC request reserved 88 but used 116

rq_reserved should always be set no smaller than the maximum possible
size of the reply, given everything we know about the request.

So the bug is that rq_reserved was set to small.

The risk of returning a reply larger than rq_reserved is a relatively
rare deadlock, so I'd rather just stick to the log warning than convert
it to a hard error.

The solution in this case is to fix the incorrect estimate.

--b.

>
> Signed-off-by: Kinglong Mee <[email protected]>
> ---
> fs/nfsd/nfs4proc.c | 7 +++++--
> 1 file changed, 5 insertions(+), 2 deletions(-)
>
> diff --git a/fs/nfsd/nfs4proc.c b/fs/nfsd/nfs4proc.c
> index 5e0dc52..084e46e 100644
> --- a/fs/nfsd/nfs4proc.c
> +++ b/fs/nfsd/nfs4proc.c
> @@ -1245,17 +1245,20 @@ static void svcxdr_init_encode(struct svc_rqst *rqstp,
> struct xdr_stream *xdr = &resp->xdr;
> struct xdr_buf *buf = &rqstp->rq_res;
> struct kvec *head = buf->head;
> + int tlen = 0;
>
> xdr->buf = buf;
> xdr->iov = head;
> xdr->p = head->iov_base + head->iov_len;
> - xdr->end = head->iov_base + PAGE_SIZE - rqstp->rq_auth_slack;
> + tlen = PAGE_SIZE - rqstp->rq_auth_slack;
> + xdr->end = head->iov_base + min(tlen, rqstp->rq_reserved);
> /* Tail and page_len should be zero at this point: */
> buf->len = buf->head[0].iov_len;
> xdr->scratch.iov_len = 0;
> xdr->page_ptr = buf->pages - 1;
> - buf->buflen = PAGE_SIZE * (1 + rqstp->rq_page_end - buf->pages)
> + tlen = PAGE_SIZE * (1 + rqstp->rq_page_end - buf->pages)
> - rqstp->rq_auth_slack;
> + buf->buflen = min(tlen, rqstp->rq_reserved);
> }
>
> /*
> --
> 1.9.3
>