Re: [RFC 09/27] pnfs: support for non-rpc layout drivers

[Benny Halevy <bhalevy@panasas.com>]

On 2011-04-20 23:34, Trond Myklebust wrote:
> On Wed, 2011-04-20 at 20:27 +0300, Benny Halevy wrote:
>> Non-rpc layout driver such as for objects and blocks
>> implement their own I/O path and error handling logic.
>> Therefore bypass NFS-based error handling for these layout drivers.
>>
>> Signed-off-by: Benny Halevy <bhalevy@panasas.com>
>> ---
>>  fs/nfs/internal.h       |    2 +
>>  fs/nfs/nfs4filelayout.c |    1 +
>>  fs/nfs/nfs4proc.c       |   14 +++++++++++-
>>  fs/nfs/pnfs.c           |   48 +++++++++++++++++++++++++++++++++++++++++++++++
>>  fs/nfs/pnfs.h           |    7 +++++-
>>  include/linux/nfs_xdr.h |    2 +
>>  6 files changed, 71 insertions(+), 3 deletions(-)
>>
>> diff --git a/fs/nfs/internal.h b/fs/nfs/internal.h
>> index ce118ce..1914d2f 100644
>> --- a/fs/nfs/internal.h
>> +++ b/fs/nfs/internal.h
>> @@ -310,6 +310,8 @@ extern int nfs_migrate_page(struct address_space *,
>>  #endif
>>  
>>  /* nfs4proc.c */
>> +extern void __nfs4_read_done_cb(struct nfs_read_data *);
>> +extern void __nfs4_write_done_cb(struct nfs_write_data *);
>>  extern void nfs4_reset_read(struct rpc_task *task, struct nfs_read_data *data);
>>  extern int nfs4_init_client(struct nfs_client *clp,
>>  			    const struct rpc_timeout *timeparms,
>> diff --git a/fs/nfs/nfs4filelayout.c b/fs/nfs/nfs4filelayout.c
>> index 2feab7f..e67a0d4 100644
>> --- a/fs/nfs/nfs4filelayout.c
>> +++ b/fs/nfs/nfs4filelayout.c
>> @@ -859,6 +859,7 @@ static struct pnfs_layoutdriver_type filelayout_type = {
>>  	.id			= LAYOUT_NFSV4_1_FILES,
>>  	.name			= "LAYOUT_NFSV4_1_FILES",
>>  	.owner			= THIS_MODULE,
>> +	.flags			= PNFS_USE_RPC_CODE,
> 
> This isn't being used anywhere, so why do I need it in this patch?
> 

Sorry, it's just leftovers from the previous version.
I'll get rid of it.

>>  	.alloc_lseg		= filelayout_alloc_lseg,
>>  	.free_lseg		= filelayout_free_lseg,
>>  	.pg_test		= filelayout_pg_test,
>> diff --git a/fs/nfs/nfs4proc.c b/fs/nfs/nfs4proc.c
>> index d0eb50b..cc2cdcd 100644
>> --- a/fs/nfs/nfs4proc.c
>> +++ b/fs/nfs/nfs4proc.c
>> @@ -3149,6 +3149,11 @@ static int nfs4_proc_pathconf(struct nfs_server *server, struct nfs_fh *fhandle,
>>  	return err;
>>  }
>>  
>> +void __nfs4_read_done_cb(struct nfs_read_data *data)
> ^^^^^^^^^^^^^^^^^^^^^^^^^^^ why the wrapper?

To be called from nfs4_read_done_cb and from pnfs_read_done.
I can just call nfs_invalidate_atime from there but since
this is common logic I think the code, small as it is,
should be kept common.

By the way, what about filelayout_read_done_cb()?
shouldn't it invalidate the inode's atime too?

>> +{
>> +	nfs_invalidate_atime(data->inode);
>> +}
>> +
>>  static int nfs4_read_done_cb(struct rpc_task *task, struct nfs_read_data *data)
>>  {
>>  	struct nfs_server *server = NFS_SERVER(data->inode);
>> @@ -3158,7 +3163,7 @@ static int nfs4_read_done_cb(struct rpc_task *task, struct nfs_read_data *data)
>>  		return -EAGAIN;
>>  	}
>>  
>> -	nfs_invalidate_atime(data->inode);
>> +	__nfs4_read_done_cb(data);
>>  	if (task->tk_status > 0)
>>  		renew_lease(server, data->timestamp);
>>  	return 0;
>> @@ -3198,6 +3203,11 @@ void nfs4_reset_read(struct rpc_task *task, struct nfs_read_data *data)
>>  }
>>  EXPORT_SYMBOL_GPL(nfs4_reset_read);
>>  
>> +void __nfs4_write_done_cb(struct nfs_write_data *data)
> ^^^^^^^^^^^^^^^^^^^^^^^^^^^^ Again, why the wrapper?
> 
>> +{
>> +	nfs_post_op_update_inode_force_wcc(data->inode, data->res.fattr);

Sorry, my bad, the pnfs done path needs just pnfs_set_layoutcommit()

>> +}
>> +
>>  static int nfs4_write_done_cb(struct rpc_task *task, struct nfs_write_data *data)
>>  {
>>  	struct inode *inode = data->inode;
>> @@ -3208,7 +3218,7 @@ static int nfs4_write_done_cb(struct rpc_task *task, struct nfs_write_data *data
>>  	}
>>  	if (task->tk_status >= 0) {
>>  		renew_lease(NFS_SERVER(inode), data->timestamp);
>> -		nfs_post_op_update_inode_force_wcc(inode, data->res.fattr);
>> +		__nfs4_write_done_cb(data);
>>  	}
>>  	return 0;
>>  }
>> diff --git a/fs/nfs/pnfs.c b/fs/nfs/pnfs.c
>> index a5050d2..18ae397 100644
>> --- a/fs/nfs/pnfs.c
>> +++ b/fs/nfs/pnfs.c
>> @@ -1130,6 +1130,30 @@ pnfs_pageio_init_write(struct nfs_pageio_descriptor *pgio, struct inode *inode)
>>  	pgio->pg_test = (ld && ld->pg_test) ? pnfs_write_pg_test : NULL;
>>  }
>>  
>> +/*
>> + * Called by non rpc-based layout drivers
>> + */
>> +int
>> +pnfs_write_done(struct nfs_write_data *data)
> ^^^^^^^^^^^^^^^^^^ If this is not generic to all pnfs layout drivers,
> then why the apparently generic name?
> 

Makes sense, how about pnfs_ld_write_done?

> Why isn't this being introduced together with a driver that actually
> uses the functionality? There is no way to review it outside of that
> context.
> 

OK, will do in next version of the patch series.

Benny

>> +{
>> +	int status;
>> +
>> +	put_lseg(data->lseg);
>> +	data->lseg = NULL;
>> +	if (!data->pnfs_error) {
>> +		__nfs4_write_done_cb(data);
>> +		data->mds_ops->rpc_call_done(NULL, data);
>> +		data->mds_ops->rpc_release(data);
>> +		return 0;
>> +	}
>> +
>> +	dprintk("%s: pnfs_error=%d, retry via MDS\n", __func__,
>> +		data->pnfs_error);
>> +	status = nfs_initiate_write(data, NFS_CLIENT(data->inode), data->mds_ops, NFS_FILE_SYNC);
>> +	return status ? : -EAGAIN;
>> +}
>> +EXPORT_SYMBOL_GPL(pnfs_write_done);
>> +
>>  enum pnfs_try_status
>>  pnfs_try_to_write_data(struct nfs_write_data *wdata,
>>  			const struct rpc_call_ops *call_ops, int how)
>> @@ -1155,6 +1179,30 @@ pnfs_try_to_write_data(struct nfs_write_data *wdata,
>>  }
>>  
>>  /*
>> + * Called by non rpc-based layout drivers
>> + */
>> +int
>> +pnfs_read_done(struct nfs_read_data *data)
>> +{
>> +	int status;
>> +
>> +	put_lseg(data->lseg);
>> +	data->lseg = NULL;
>> +	if (!data->pnfs_error) {
>> +		__nfs4_read_done_cb(data);
>> +		data->mds_ops->rpc_call_done(NULL, data);
>> +		data->mds_ops->rpc_release(data);
>> +		return 0;
>> +	}
>> +
>> +	dprintk("%s: pnfs_error=%d, retry via MDS\n", __func__,
>> +		data->pnfs_error);
>> +	status = nfs_initiate_read(data, NFS_CLIENT(data->inode), data->mds_ops);
>> +	return status ? : -EAGAIN;
>> +}
>> +EXPORT_SYMBOL_GPL(pnfs_read_done);
>> +
>> +/*
>>   * Call the appropriate parallel I/O subsystem read function.
>>   */
>>  enum pnfs_try_status
>> diff --git a/fs/nfs/pnfs.h b/fs/nfs/pnfs.h
>> index 9f8e970..18b84ce 100644
>> --- a/fs/nfs/pnfs.h
>> +++ b/fs/nfs/pnfs.h
>> @@ -65,8 +65,11 @@ enum {
>>  };
>>  
>>  enum layoutdriver_policy_flags {
>> +	/* Should the full nfs rpc cleanup code be used after io */
>> +	PNFS_USE_RPC_CODE		= 1 << 0,
>> +
>>  	/* Should the pNFS client commit and return the layout upon a setattr */
>> -	PNFS_LAYOUTRET_ON_SETATTR	= 1 << 0,
>> +	PNFS_LAYOUTRET_ON_SETATTR	= 1 << 1,
>>  };
>>  
>>  /* Per-layout driver specific registration structure */
>> @@ -182,6 +185,8 @@ bool pnfs_roc_drain(struct inode *ino, u32 *barrier);
>>  void pnfs_set_layoutcommit(struct nfs_write_data *wdata);
>>  int pnfs_layoutcommit_inode(struct inode *inode, bool sync);
>>  int _pnfs_return_layout(struct inode *, struct pnfs_layout_range *, bool wait);
>> +int pnfs_write_done(struct nfs_write_data *);
>> +int pnfs_read_done(struct nfs_read_data *);
>>  
>>  static inline int lo_fail_bit(u32 iomode)
>>  {
>> diff --git a/include/linux/nfs_xdr.h b/include/linux/nfs_xdr.h
>> index 01eb1ae..41f896a 100644
>> --- a/include/linux/nfs_xdr.h
>> +++ b/include/linux/nfs_xdr.h
>> @@ -1108,6 +1108,7 @@ struct nfs_read_data {
>>  	const struct rpc_call_ops *mds_ops;
>>  	int (*read_done_cb) (struct rpc_task *task, struct nfs_read_data *data);
>>  	__u64			mds_offset;
>> +	int			pnfs_error;
>>  	struct page		*page_array[NFS_PAGEVEC_SIZE];
>>  };
>>  
>> @@ -1133,6 +1134,7 @@ struct nfs_write_data {
>>  	unsigned long		timestamp;	/* For lease renewal */
>>  #endif
>>  	__u64			mds_offset;	/* Filelayout dense stripe */
>> +	int			pnfs_error;
>>  	struct page		*page_array[NFS_PAGEVEC_SIZE];
>>  };
>>  
>