Re: [PATCH 2/4] NFSD: Rervert "knfsd: locks: flag NFSv4-owned locks"

[Kinglong Mee <kinglongmee@gmail.com>]

On 7/8/2014 19:03, Jeff Layton wrote:
> On Tue, 08 Jul 2014 11:23:49 +0800
> Kinglong Mee <kinglongmee@gmail.com> wrote:
> 
>> On 7/8/2014 01:26, Jeff Layton wrote:
>>> On Mon, 07 Jul 2014 22:12:21 +0800
>>> Kinglong Mee <kinglongmee@gmail.com> wrote:
>>>
> -------------------------[snip]------------------------------
> 
> [PATCH] NFSD: Rervert "knfsd: locks: flag NFSv4-owned locks"
> 
> Commit d5b9026a67 ([PATCH] knfsd: locks: flag NFSv4-owned locks) using
> fl_lmops field in file_lock for checking nfsd4 lockowner.
> 
> But, commit 1a747ee0cc (locks: don't call ->copy_lock methods on return of conflicting locks) causes the fl_lmops of conflock for nfsd4_lock always be NULL.
> 
> Also, commit 0996905f93 (lockd: posix_test_lock() should not call locks_copy_lock()) caused the fl_lmops of conflock for nfsd4_lockt always be NULL too.
> 
> So that, using fl_lmops cannot checking nfsd4 lockowner for it always be NULL.
> 
> Signed-off-by: Kinglong Mee <kinglongmee@gmail.com>
> Signed-off-by: Jeff Layton <jlayton@primarydata.com>
> ---
>  fs/nfsd/nfs4state.c | 34 ++++++++++++++++++++++------------
>  1 file changed, 22 insertions(+), 12 deletions(-)
> 
> diff --git a/fs/nfsd/nfs4state.c b/fs/nfsd/nfs4state.c
> index c400ec17915e..00cd7fd17b0f 100644
> --- a/fs/nfsd/nfs4state.c
> +++ b/fs/nfsd/nfs4state.c
> @@ -4312,18 +4312,29 @@ nfs4_transform_lock_offset(struct file_lock *lock)
>  		lock->fl_end = OFFSET_MAX;
>  }
>  
> -/* Hack!: For now, we're defining this just so we can use a pointer to it
> - * as a unique cookie to identify our (NFSv4's) posix locks. */
> -static const struct lock_manager_operations nfsd_posix_mng_ops  = {
> -};
> +static bool
> +nfs4_verify_lock_stateowner(struct nfs4_lockowner *lo, struct nfs4_client *clp)
> +{
> +	struct nfs4_stateowner *local = NULL;
> +	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
> +	unsigned int hashval = ownerstr_hashval(clp->cl_clientid.cl_id,
> +						&lo->lo_owner.so_owner);

Needs the conflicting client which holds the lock,
Not the new client which try to locking the file has be locked.

> +
> +	list_for_each_entry(local, &nn->ownerstr_hashtbl[hashval], so_strhash) {
> +		if (local == &lo->lo_owner)
> +			return true;
> +	}
> +
> +	return false;
> +}
>  
>  static inline void
> -nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
> +nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny,
> +			struct nfs4_client *clp)
>  {
> -	struct nfs4_lockowner *lo;
> +	struct nfs4_lockowner *lo = (struct nfs4_lockowner *)fl->fl_owner;
>  
> -	if (fl->fl_lmops == &nfsd_posix_mng_ops) {
> -		lo = (struct nfs4_lockowner *) fl->fl_owner;
> +	if (nfs4_verify_lock_stateowner(lo, clp)) {
>  		deny->ld_owner.data = kmemdup(lo->lo_owner.so_owner.data,
>  					lo->lo_owner.so_owner.len, GFP_KERNEL);
>  		if (!deny->ld_owner.data)
> @@ -4594,7 +4605,6 @@ nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
>  	file_lock->fl_pid = current->tgid;
>  	file_lock->fl_file = filp;
>  	file_lock->fl_flags = FL_POSIX;
> -	file_lock->fl_lmops = &nfsd_posix_mng_ops;
>  	file_lock->fl_start = lock->lk_offset;
>  	file_lock->fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
>  	nfs4_transform_lock_offset(file_lock);
> @@ -4617,7 +4627,8 @@ nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
>  	case (EAGAIN):		/* conflock holds conflicting lock */
>  		status = nfserr_denied;
>  		dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
> -		nfs4_set_lock_denied(conflock, &lock->lk_denied);
> +		nfs4_set_lock_denied(conflock, &lock->lk_denied,
> +					cstate->clp);
>  		break;
>  	case (EDEADLK):
>  		status = nfserr_deadlock;
> @@ -4725,7 +4736,7 @@ nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
>  
>  	if (file_lock->fl_type != F_UNLCK) {
>  		status = nfserr_denied;
> -		nfs4_set_lock_denied(file_lock, &lockt->lt_denied);
> +		nfs4_set_lock_denied(file_lock, &lockt->lt_denied, cstate->clp);

The clp is the client which try to lock the file here,
but the file have be locked by other client or server.

With this patch, the hashval is wrong.
              *correct*                    *this patch*
clid        conflicting client        new request lock client
owner      from vfs_lock_file()        from vfs_lock_file()  
            or vfs_test_lock()           or vfs_test_lock()
hashval       right hashval           bad hashval (never exist)

With this patch, LOCK/LOCKT compound gets client with all zero, and owner with all zero.

thanks,
Kinglong Mee