Re: [PATCH] nfsd: call op_release, even when op_func returns an error

[Jeff Layton <jlayton@kernel.org>]

On Mon, 2023-03-27 at 13:14 +0000, Chuck Lever III wrote:
> 
> > On Mar 27, 2023, at 6:21 AM, Jeff Layton <jlayton@kernel.org> wrote:
> > 
> > For ops with "trivial" replies, nfsd4_encode_operation will shortcut
> > most of the encoding work and skip to just marshalling up the status.
> > One of the things it skips is calling op_release. This could cause a
> > memory leak in the layoutget codepath if there is an error at an
> > inopportune time.
> > 
> > Have the compound processing engine always call op_release, even when
> > op_func sets an error in op->status. With this change, we also need
> > nfsd4_block_get_device_info_scsi to set the gd_device pointer to NULL
> > on error to avoid a double free.
> > 
> > Reported-by: Zhi Li <yieli@redhat.com>
> > Link: https://bugzilla.redhat.com/show_bug.cgi?id=2181403
> > Signed-off-by: Jeff Layton <jlayton@kernel.org>
> 
> Thanks, Jeff.
> 
> May I add: Fixes: 34b1744c91cc ("nfsd4: define ->op_release for
> compound ops") ?
> 

I've seen some problems with this patch in testing and I have a fix
forthcoming (once I finish testing it):

The root cause is the OPDESC() function which can walk off the end of
the nfsd4_ops array when passed a large value (like OP_ILLEGAL). I think
we'll want to fix that to do something more sane before merging this
patch.

> 
> > ---
> > fs/nfsd/blocklayout.c |  1 +
> > fs/nfsd/nfs4xdr.c     | 13 +++++++------
> > 2 files changed, 8 insertions(+), 6 deletions(-)
> > 
> > diff --git a/fs/nfsd/blocklayout.c b/fs/nfsd/blocklayout.c
> > index 04697f8dc37d..01d7fd108cf3 100644
> > --- a/fs/nfsd/blocklayout.c
> > +++ b/fs/nfsd/blocklayout.c
> > @@ -297,6 +297,7 @@ nfsd4_block_get_device_info_scsi(struct super_block *sb,
> > 
> > out_free_dev:
> > 	kfree(dev);
> > +	gdp->gd_device = NULL;
> > 	return ret;
> > }
> > 
> > diff --git a/fs/nfsd/nfs4xdr.c b/fs/nfsd/nfs4xdr.c
> > index e12e5a4ad502..6b675fbdabd0 100644
> > --- a/fs/nfsd/nfs4xdr.c
> > +++ b/fs/nfsd/nfs4xdr.c
> > @@ -5402,7 +5402,7 @@ nfsd4_encode_operation(struct nfsd4_compoundres *resp, struct nfsd4_op *op)
> > 	p = xdr_reserve_space(xdr, 8);
> > 	if (!p) {
> > 		WARN_ON_ONCE(1);
> > -		return;
> > +		goto release;
> > 	}
> > 	*p++ = cpu_to_be32(op->opnum);
> > 	post_err_offset = xdr->buf->len;
> > @@ -5418,8 +5418,6 @@ nfsd4_encode_operation(struct nfsd4_compoundres *resp, struct nfsd4_op *op)
> > 	op->status = encoder(resp, op->status, &op->u);
> > 	if (op->status)
> > 		trace_nfsd_compound_encode_err(rqstp, op->opnum, op->status);
> > -	if (opdesc && opdesc->op_release)
> > -		opdesc->op_release(&op->u);
> > 	xdr_commit_encode(xdr);
> > 
> > 	/* nfsd4_check_resp_size guarantees enough room for error status */
> > @@ -5460,11 +5458,14 @@ nfsd4_encode_operation(struct nfsd4_compoundres *resp, struct nfsd4_op *op)
> > 	}
> > status:
> > 	*p = op->status;
> > +release:
> > +	if (opdesc && opdesc->op_release)
> > +		opdesc->op_release(&op->u);
> > }
> > 
> > -/* 
> > - * Encode the reply stored in the stateowner reply cache 
> > - * 
> > +/*
> > + * Encode the reply stored in the stateowner reply cache
> > + *
> >  * XDR note: do not encode rp->rp_buflen: the buffer contains the
> >  * previously sent already encoded operation.
> >  */
> > -- 
> > 2.39.2
> > 
> 
> --
> Chuck Lever
> 
> 

-- 
Jeff Layton <jlayton@kernel.org>