diff mbox series

[v2,1/2] NFSD: handle GETATTR conflict with write delegation

Message ID 1685429537-11855-2-git-send-email-dai.ngo@oracle.com (mailing list archive)
State New, archived
Headers show
Series NFSD: recall write delegation on GETATTR conflict | expand

Commit Message

Dai Ngo May 30, 2023, 6:52 a.m. UTC
If the GETATTR request on a file that has write delegation in effect
and the request attributes include the change info and size attribute
then the write delegation is recalled. If the delegation is returned
within 30ms then the GETATTR is serviced as normal otherwise the
NFS4ERR_DELAY error is returned for the GETATTR.

Signed-off-by: Dai Ngo <dai.ngo@oracle.com>
---
 fs/nfsd/nfs4state.c | 50 ++++++++++++++++++++++++++++++++++++++++++++++++++
 fs/nfsd/nfs4xdr.c   |  5 +++++
 fs/nfsd/state.h     |  3 +++
 3 files changed, 58 insertions(+)

Comments

Jeff Layton May 30, 2023, 10:29 a.m. UTC | #1
On Mon, 2023-05-29 at 23:52 -0700, Dai Ngo wrote:
> If the GETATTR request on a file that has write delegation in effect
> and the request attributes include the change info and size attribute
> then the write delegation is recalled. If the delegation is returned
> within 30ms then the GETATTR is serviced as normal otherwise the
> NFS4ERR_DELAY error is returned for the GETATTR.
> 
> Signed-off-by: Dai Ngo <dai.ngo@oracle.com>
> ---
>  fs/nfsd/nfs4state.c | 50 ++++++++++++++++++++++++++++++++++++++++++++++++++
>  fs/nfsd/nfs4xdr.c   |  5 +++++
>  fs/nfsd/state.h     |  3 +++
>  3 files changed, 58 insertions(+)
> 
> diff --git a/fs/nfsd/nfs4state.c b/fs/nfsd/nfs4state.c
> index b90b74a5e66e..7826483e8421 100644
> --- a/fs/nfsd/nfs4state.c
> +++ b/fs/nfsd/nfs4state.c
> @@ -8353,3 +8353,53 @@ nfsd4_get_writestateid(struct nfsd4_compound_state *cstate,
>  {
>  	get_stateid(cstate, &u->write.wr_stateid);
>  }
> +
> +/**
> + * nfsd4_deleg_getattr_conflict - Trigger recall if GETATTR causes conflict
> + * @rqstp: RPC transaction context
> + * @inode: file to be checked for a conflict
> + *
> + * This function is called when there is a conflict between a write
> + * delegation and a change/size GETATR from another client. The server
> + * must either use the CB_GETATTR to get the current values of the
> + * attributes from the client that hold the delegation or recall the
> + * delegation before replying to the GETATTR. See RFC 8881 section
> + * 18.7.4.
> + *
> + * Returns 0 if there is no conflict; otherwise an nfs_stat
> + * code is returned.
> + */
> +__be32
> +nfsd4_deleg_getattr_conflict(struct svc_rqst *rqstp, struct inode *inode)
> +{
> +	__be32 status;
> +	struct file_lock_context *ctx;
> +	struct file_lock *fl;
> +	struct nfs4_delegation *dp;
> +
> +	ctx = locks_inode_context(inode);
> +	if (!ctx)
> +		return 0;
> +	spin_lock(&ctx->flc_lock);
> +	list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
> +		if (fl->fl_flags == FL_LAYOUT ||
> +				fl->fl_lmops != &nfsd_lease_mng_ops)
> +			continue;
> +		if (fl->fl_type == F_WRLCK) {
> +			dp = fl->fl_owner;
> +			if (dp->dl_recall.cb_clp == *(rqstp->rq_lease_breaker)) {
> +				spin_unlock(&ctx->flc_lock);
> +				return 0;
> +			}
> +			spin_unlock(&ctx->flc_lock);
> +			status = nfserrno(nfsd_open_break_lease(inode, NFSD_MAY_READ));
> +			if (status != nfserr_jukebox ||
> +					!nfsd_wait_for_delegreturn(rqstp, inode))
> +				return status;
> +			return 0;
> +		}
> +		break;
> +	}
> +	spin_unlock(&ctx->flc_lock);
> +	return 0;
> +}


If there is a lease held by a userland program (e.g. Samba), why don't
you want to break it here? Shouldn't it also be broken in this case?

I think this logic may be wrong. ISTM that you want to basically always
call nfsd_open_break_lease, unless it's a delegation held by the same
client.


> diff --git a/fs/nfsd/nfs4xdr.c b/fs/nfsd/nfs4xdr.c
> index b83954fc57e3..4590b893dbc8 100644
> --- a/fs/nfsd/nfs4xdr.c
> +++ b/fs/nfsd/nfs4xdr.c
> @@ -2970,6 +2970,11 @@ nfsd4_encode_fattr(struct xdr_stream *xdr, struct svc_fh *fhp,
>  		if (status)
>  			goto out;
>  	}
> +	if (bmval0 & (FATTR4_WORD0_CHANGE | FATTR4_WORD0_SIZE)) {
> +		status = nfsd4_deleg_getattr_conflict(rqstp, d_inode(dentry));
> +		if (status)
> +			goto out;
> +	}
>  
>  	err = vfs_getattr(&path, &stat,
>  			  STATX_BASIC_STATS | STATX_BTIME | STATX_CHANGE_COOKIE,
> diff --git a/fs/nfsd/state.h b/fs/nfsd/state.h
> index d49d3060ed4f..cbddcf484dba 100644
> --- a/fs/nfsd/state.h
> +++ b/fs/nfsd/state.h
> @@ -732,4 +732,7 @@ static inline bool try_to_expire_client(struct nfs4_client *clp)
>  	cmpxchg(&clp->cl_state, NFSD4_COURTESY, NFSD4_EXPIRABLE);
>  	return clp->cl_state == NFSD4_EXPIRABLE;
>  }
> +
> +extern __be32 nfsd4_deleg_getattr_conflict(struct svc_rqst *rqstp,
> +				struct inode *inode);
>  #endif   /* NFSD4_STATE_H */
Dai Ngo May 30, 2023, 10:04 p.m. UTC | #2
On 5/30/23 3:29 AM, Jeff Layton wrote:
> On Mon, 2023-05-29 at 23:52 -0700, Dai Ngo wrote:
>> If the GETATTR request on a file that has write delegation in effect
>> and the request attributes include the change info and size attribute
>> then the write delegation is recalled. If the delegation is returned
>> within 30ms then the GETATTR is serviced as normal otherwise the
>> NFS4ERR_DELAY error is returned for the GETATTR.
>>
>> Signed-off-by: Dai Ngo <dai.ngo@oracle.com>
>> ---
>>   fs/nfsd/nfs4state.c | 50 ++++++++++++++++++++++++++++++++++++++++++++++++++
>>   fs/nfsd/nfs4xdr.c   |  5 +++++
>>   fs/nfsd/state.h     |  3 +++
>>   3 files changed, 58 insertions(+)
>>
>> diff --git a/fs/nfsd/nfs4state.c b/fs/nfsd/nfs4state.c
>> index b90b74a5e66e..7826483e8421 100644
>> --- a/fs/nfsd/nfs4state.c
>> +++ b/fs/nfsd/nfs4state.c
>> @@ -8353,3 +8353,53 @@ nfsd4_get_writestateid(struct nfsd4_compound_state *cstate,
>>   {
>>   	get_stateid(cstate, &u->write.wr_stateid);
>>   }
>> +
>> +/**
>> + * nfsd4_deleg_getattr_conflict - Trigger recall if GETATTR causes conflict
>> + * @rqstp: RPC transaction context
>> + * @inode: file to be checked for a conflict
>> + *
>> + * This function is called when there is a conflict between a write
>> + * delegation and a change/size GETATR from another client. The server
>> + * must either use the CB_GETATTR to get the current values of the
>> + * attributes from the client that hold the delegation or recall the
>> + * delegation before replying to the GETATTR. See RFC 8881 section
>> + * 18.7.4.
>> + *
>> + * Returns 0 if there is no conflict; otherwise an nfs_stat
>> + * code is returned.
>> + */
>> +__be32
>> +nfsd4_deleg_getattr_conflict(struct svc_rqst *rqstp, struct inode *inode)
>> +{
>> +	__be32 status;
>> +	struct file_lock_context *ctx;
>> +	struct file_lock *fl;
>> +	struct nfs4_delegation *dp;
>> +
>> +	ctx = locks_inode_context(inode);
>> +	if (!ctx)
>> +		return 0;
>> +	spin_lock(&ctx->flc_lock);
>> +	list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
>> +		if (fl->fl_flags == FL_LAYOUT ||
>> +				fl->fl_lmops != &nfsd_lease_mng_ops)
>> +			continue;
>> +		if (fl->fl_type == F_WRLCK) {
>> +			dp = fl->fl_owner;
>> +			if (dp->dl_recall.cb_clp == *(rqstp->rq_lease_breaker)) {
>> +				spin_unlock(&ctx->flc_lock);
>> +				return 0;
>> +			}
>> +			spin_unlock(&ctx->flc_lock);
>> +			status = nfserrno(nfsd_open_break_lease(inode, NFSD_MAY_READ));
>> +			if (status != nfserr_jukebox ||
>> +					!nfsd_wait_for_delegreturn(rqstp, inode))
>> +				return status;
>> +			return 0;
>> +		}
>> +		break;
>> +	}
>> +	spin_unlock(&ctx->flc_lock);
>> +	return 0;
>> +}
>
> If there is a lease held by a userland program (e.g. Samba), why don't
> you want to break it here? Shouldn't it also be broken in this case?

okay, I will make the change to also break non-nfs lease with F_WRLCK.

>
> I think this logic may be wrong. ISTM that you want to basically always
> call nfsd_open_break_lease, unless it's a delegation held by the same
> client.

I don't think we need to break any lease with F_RDLCK.

-Dai

>
>
>> diff --git a/fs/nfsd/nfs4xdr.c b/fs/nfsd/nfs4xdr.c
>> index b83954fc57e3..4590b893dbc8 100644
>> --- a/fs/nfsd/nfs4xdr.c
>> +++ b/fs/nfsd/nfs4xdr.c
>> @@ -2970,6 +2970,11 @@ nfsd4_encode_fattr(struct xdr_stream *xdr, struct svc_fh *fhp,
>>   		if (status)
>>   			goto out;
>>   	}
>> +	if (bmval0 & (FATTR4_WORD0_CHANGE | FATTR4_WORD0_SIZE)) {
>> +		status = nfsd4_deleg_getattr_conflict(rqstp, d_inode(dentry));
>> +		if (status)
>> +			goto out;
>> +	}
>>   
>>   	err = vfs_getattr(&path, &stat,
>>   			  STATX_BASIC_STATS | STATX_BTIME | STATX_CHANGE_COOKIE,
>> diff --git a/fs/nfsd/state.h b/fs/nfsd/state.h
>> index d49d3060ed4f..cbddcf484dba 100644
>> --- a/fs/nfsd/state.h
>> +++ b/fs/nfsd/state.h
>> @@ -732,4 +732,7 @@ static inline bool try_to_expire_client(struct nfs4_client *clp)
>>   	cmpxchg(&clp->cl_state, NFSD4_COURTESY, NFSD4_EXPIRABLE);
>>   	return clp->cl_state == NFSD4_EXPIRABLE;
>>   }
>> +
>> +extern __be32 nfsd4_deleg_getattr_conflict(struct svc_rqst *rqstp,
>> +				struct inode *inode);
>>   #endif   /* NFSD4_STATE_H */
Jeff Layton May 30, 2023, 11:48 p.m. UTC | #3
On Tue, 2023-05-30 at 15:04 -0700, dai.ngo@oracle.com wrote:
> On 5/30/23 3:29 AM, Jeff Layton wrote:
> > On Mon, 2023-05-29 at 23:52 -0700, Dai Ngo wrote:
> > > If the GETATTR request on a file that has write delegation in effect
> > > and the request attributes include the change info and size attribute
> > > then the write delegation is recalled. If the delegation is returned
> > > within 30ms then the GETATTR is serviced as normal otherwise the
> > > NFS4ERR_DELAY error is returned for the GETATTR.
> > > 
> > > Signed-off-by: Dai Ngo <dai.ngo@oracle.com>
> > > ---
> > >   fs/nfsd/nfs4state.c | 50 ++++++++++++++++++++++++++++++++++++++++++++++++++
> > >   fs/nfsd/nfs4xdr.c   |  5 +++++
> > >   fs/nfsd/state.h     |  3 +++
> > >   3 files changed, 58 insertions(+)
> > > 
> > > diff --git a/fs/nfsd/nfs4state.c b/fs/nfsd/nfs4state.c
> > > index b90b74a5e66e..7826483e8421 100644
> > > --- a/fs/nfsd/nfs4state.c
> > > +++ b/fs/nfsd/nfs4state.c
> > > @@ -8353,3 +8353,53 @@ nfsd4_get_writestateid(struct nfsd4_compound_state *cstate,
> > >   {
> > >   	get_stateid(cstate, &u->write.wr_stateid);
> > >   }
> > > +
> > > +/**
> > > + * nfsd4_deleg_getattr_conflict - Trigger recall if GETATTR causes conflict
> > > + * @rqstp: RPC transaction context
> > > + * @inode: file to be checked for a conflict
> > > + *
> > > + * This function is called when there is a conflict between a write
> > > + * delegation and a change/size GETATR from another client. The server
> > > + * must either use the CB_GETATTR to get the current values of the
> > > + * attributes from the client that hold the delegation or recall the
> > > + * delegation before replying to the GETATTR. See RFC 8881 section
> > > + * 18.7.4.
> > > + *
> > > + * Returns 0 if there is no conflict; otherwise an nfs_stat
> > > + * code is returned.
> > > + */
> > > +__be32
> > > +nfsd4_deleg_getattr_conflict(struct svc_rqst *rqstp, struct inode *inode)
> > > +{
> > > +	__be32 status;
> > > +	struct file_lock_context *ctx;
> > > +	struct file_lock *fl;
> > > +	struct nfs4_delegation *dp;
> > > +
> > > +	ctx = locks_inode_context(inode);
> > > +	if (!ctx)
> > > +		return 0;
> > > +	spin_lock(&ctx->flc_lock);
> > > +	list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
> > > +		if (fl->fl_flags == FL_LAYOUT ||
> > > +				fl->fl_lmops != &nfsd_lease_mng_ops)
> > > +			continue;
> > > +		if (fl->fl_type == F_WRLCK) {
> > > +			dp = fl->fl_owner;
> > > +			if (dp->dl_recall.cb_clp == *(rqstp->rq_lease_breaker)) {
> > > +				spin_unlock(&ctx->flc_lock);
> > > +				return 0;
> > > +			}
> > > +			spin_unlock(&ctx->flc_lock);
> > > +			status = nfserrno(nfsd_open_break_lease(inode, NFSD_MAY_READ));
> > > +			if (status != nfserr_jukebox ||
> > > +					!nfsd_wait_for_delegreturn(rqstp, inode))
> > > +				return status;
> > > +			return 0;
> > > +		}
> > > +		break;
> > > +	}
> > > +	spin_unlock(&ctx->flc_lock);
> > > +	return 0;
> > > +}
> > 
> > If there is a lease held by a userland program (e.g. Samba), why don't
> > you want to break it here? Shouldn't it also be broken in this case?
> 
> okay, I will make the change to also break non-nfs lease with F_WRLCK.
> 

Sounds good.

> > 
> > I think this logic may be wrong. ISTM that you want to basically always
> > call nfsd_open_break_lease, unless it's a delegation held by the same
> > client.
> 
> I don't think we need to break any lease with F_RDLCK.
> 

Correct. To be clear: break any write lease unless it's a write
delegation held by the same client.

> 
> > 
> > 
> > > diff --git a/fs/nfsd/nfs4xdr.c b/fs/nfsd/nfs4xdr.c
> > > index b83954fc57e3..4590b893dbc8 100644
> > > --- a/fs/nfsd/nfs4xdr.c
> > > +++ b/fs/nfsd/nfs4xdr.c
> > > @@ -2970,6 +2970,11 @@ nfsd4_encode_fattr(struct xdr_stream *xdr, struct svc_fh *fhp,
> > >   		if (status)
> > >   			goto out;
> > >   	}
> > > +	if (bmval0 & (FATTR4_WORD0_CHANGE | FATTR4_WORD0_SIZE)) {
> > > +		status = nfsd4_deleg_getattr_conflict(rqstp, d_inode(dentry));
> > > +		if (status)
> > > +			goto out;
> > > +	}
> > >   
> > >   	err = vfs_getattr(&path, &stat,
> > >   			  STATX_BASIC_STATS | STATX_BTIME | STATX_CHANGE_COOKIE,
> > > diff --git a/fs/nfsd/state.h b/fs/nfsd/state.h
> > > index d49d3060ed4f..cbddcf484dba 100644
> > > --- a/fs/nfsd/state.h
> > > +++ b/fs/nfsd/state.h
> > > @@ -732,4 +732,7 @@ static inline bool try_to_expire_client(struct nfs4_client *clp)
> > >   	cmpxchg(&clp->cl_state, NFSD4_COURTESY, NFSD4_EXPIRABLE);
> > >   	return clp->cl_state == NFSD4_EXPIRABLE;
> > >   }
> > > +
> > > +extern __be32 nfsd4_deleg_getattr_conflict(struct svc_rqst *rqstp,
> > > +				struct inode *inode);
> > >   #endif   /* NFSD4_STATE_H */
diff mbox series

Patch

diff --git a/fs/nfsd/nfs4state.c b/fs/nfsd/nfs4state.c
index b90b74a5e66e..7826483e8421 100644
--- a/fs/nfsd/nfs4state.c
+++ b/fs/nfsd/nfs4state.c
@@ -8353,3 +8353,53 @@  nfsd4_get_writestateid(struct nfsd4_compound_state *cstate,
 {
 	get_stateid(cstate, &u->write.wr_stateid);
 }
+
+/**
+ * nfsd4_deleg_getattr_conflict - Trigger recall if GETATTR causes conflict
+ * @rqstp: RPC transaction context
+ * @inode: file to be checked for a conflict
+ *
+ * This function is called when there is a conflict between a write
+ * delegation and a change/size GETATR from another client. The server
+ * must either use the CB_GETATTR to get the current values of the
+ * attributes from the client that hold the delegation or recall the
+ * delegation before replying to the GETATTR. See RFC 8881 section
+ * 18.7.4.
+ *
+ * Returns 0 if there is no conflict; otherwise an nfs_stat
+ * code is returned.
+ */
+__be32
+nfsd4_deleg_getattr_conflict(struct svc_rqst *rqstp, struct inode *inode)
+{
+	__be32 status;
+	struct file_lock_context *ctx;
+	struct file_lock *fl;
+	struct nfs4_delegation *dp;
+
+	ctx = locks_inode_context(inode);
+	if (!ctx)
+		return 0;
+	spin_lock(&ctx->flc_lock);
+	list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
+		if (fl->fl_flags == FL_LAYOUT ||
+				fl->fl_lmops != &nfsd_lease_mng_ops)
+			continue;
+		if (fl->fl_type == F_WRLCK) {
+			dp = fl->fl_owner;
+			if (dp->dl_recall.cb_clp == *(rqstp->rq_lease_breaker)) {
+				spin_unlock(&ctx->flc_lock);
+				return 0;
+			}
+			spin_unlock(&ctx->flc_lock);
+			status = nfserrno(nfsd_open_break_lease(inode, NFSD_MAY_READ));
+			if (status != nfserr_jukebox ||
+					!nfsd_wait_for_delegreturn(rqstp, inode))
+				return status;
+			return 0;
+		}
+		break;
+	}
+	spin_unlock(&ctx->flc_lock);
+	return 0;
+}
diff --git a/fs/nfsd/nfs4xdr.c b/fs/nfsd/nfs4xdr.c
index b83954fc57e3..4590b893dbc8 100644
--- a/fs/nfsd/nfs4xdr.c
+++ b/fs/nfsd/nfs4xdr.c
@@ -2970,6 +2970,11 @@  nfsd4_encode_fattr(struct xdr_stream *xdr, struct svc_fh *fhp,
 		if (status)
 			goto out;
 	}
+	if (bmval0 & (FATTR4_WORD0_CHANGE | FATTR4_WORD0_SIZE)) {
+		status = nfsd4_deleg_getattr_conflict(rqstp, d_inode(dentry));
+		if (status)
+			goto out;
+	}
 
 	err = vfs_getattr(&path, &stat,
 			  STATX_BASIC_STATS | STATX_BTIME | STATX_CHANGE_COOKIE,
diff --git a/fs/nfsd/state.h b/fs/nfsd/state.h
index d49d3060ed4f..cbddcf484dba 100644
--- a/fs/nfsd/state.h
+++ b/fs/nfsd/state.h
@@ -732,4 +732,7 @@  static inline bool try_to_expire_client(struct nfs4_client *clp)
 	cmpxchg(&clp->cl_state, NFSD4_COURTESY, NFSD4_EXPIRABLE);
 	return clp->cl_state == NFSD4_EXPIRABLE;
 }
+
+extern __be32 nfsd4_deleg_getattr_conflict(struct svc_rqst *rqstp,
+				struct inode *inode);
 #endif   /* NFSD4_STATE_H */