Re: [RFC PATCH] nfsd: allow nfsd to advertise multiple layout types

["Mkrtchyan, Tigran" <tigran.mkrtchyan@desy.de>]

----- Original Message -----
> From: "Jeff Layton" <jlayton@poochiereds.net>
> To: "Mkrtchyan, Tigran" <tigran.mkrtchyan@desy.de>
> Cc: linux-nfs@vger.kernel.org, "Tom Haynes" <thomas.haynes@primarydata.com>, "Christoph Hellwig" <hch@lst.de>
> Sent: Monday, May 30, 2016 6:51:38 PM
> Subject: Re: [RFC PATCH] nfsd: allow nfsd to advertise multiple layout types

> On Mon, 2016-05-30 at 17:47 +0200, Mkrtchyan, Tigran wrote:
>> 
>> ----- Original Message -----
>> > From: "Jeff Layton" <jlayton@poochiereds.net>
>> > To: linux-nfs@vger.kernel.org
>> > Cc: "Tom Haynes" <thomas.haynes@primarydata.com>, "Christoph Hellwig"
>> > <hch@lst.de>
>> > Sent: Sunday, May 29, 2016 11:06:14 PM
>> > Subject: Re: [RFC PATCH] nfsd: allow nfsd to advertise multiple layout types
>> 
>> > On Sun, 2016-05-29 at 16:59 -0400, Jeff Layton wrote:
>> > > If the underlying filesystem supports multiple layout types, then there
>> > > is little reason not to advertise that fact to clients and let them
>> > > choose what type to use.
>> > > 
>> > > Turn the ex_layout_type field into a bitfield. For each supported
>> > > layout type, we set a bit in that field. When the client requests a
>> > > layout, ensure that the bit for that layout type is set. When the
>> > > client requests attributes, send back a list of supported types.
>> > > 
>> > > Signed-off-by: Jeff Layton <jeff.layton@primarydata.com>
>> > > ---
>> > >  fs/nfsd/export.c      |  4 ++--
>> > >  fs/nfsd/export.h      |  2 +-
>> > >  fs/nfsd/nfs4layouts.c |  6 +++---
>> > >  fs/nfsd/nfs4proc.c    |  4 ++--
>> > >  fs/nfsd/nfs4xdr.c     | 30 ++++++++++++++----------------
>> > >  5 files changed, 22 insertions(+), 24 deletions(-)
>> > > 
>> > > diff --git a/fs/nfsd/export.c b/fs/nfsd/export.c
>> > > index b4d84b579f20..f97ba49d5e66 100644
>> > > --- a/fs/nfsd/export.c
>> > > +++ b/fs/nfsd/export.c
>> > > @@ -706,7 +706,7 @@ static void svc_export_init(struct cache_head *cnew, struct
>> > > cache_head *citem)
>> > >  	new->ex_fslocs.locations = NULL;
>> > >  	new->ex_fslocs.locations_count = 0;
>> > >  	new->ex_fslocs.migrated = 0;
>> > > -	new->ex_layout_type = 0;
>> > > +	new->ex_layout_types = 0;
>> > >  	new->ex_uuid = NULL;
>> > >  	new->cd = item->cd;
>> > >  }
>> > > @@ -731,7 +731,7 @@ static void export_update(struct cache_head *cnew, struct
>> > > cache_head *citem)
>> > >  	item->ex_fslocs.locations_count = 0;
>> > >  	new->ex_fslocs.migrated = item->ex_fslocs.migrated;
>> > >  	item->ex_fslocs.migrated = 0;
>> > > -	new->ex_layout_type = item->ex_layout_type;
>> > > +	new->ex_layout_types = item->ex_layout_types;
>> > >  	new->ex_nflavors = item->ex_nflavors;
>> > >  	for (i = 0; i < MAX_SECINFO_LIST; i++) {
>> > >  		new->ex_flavors[i] = item->ex_flavors[i];
>> > > diff --git a/fs/nfsd/export.h b/fs/nfsd/export.h
>> > > index 2e315072bf3f..730f15eeb7ed 100644
>> > > --- a/fs/nfsd/export.h
>> > > +++ b/fs/nfsd/export.h
>> > > @@ -57,7 +57,7 @@ struct svc_export {
>> > >  	struct nfsd4_fs_locations ex_fslocs;
>> > >  	uint32_t		ex_nflavors;
>> > >  	struct exp_flavor_info	ex_flavors[MAX_SECINFO_LIST];
>> > > -	enum pnfs_layouttype	ex_layout_type;
>> > > +	u32			ex_layout_types;
>> > >  	struct nfsd4_deviceid_map *ex_devid_map;
>> > >  	struct cache_detail	*cd;
>> > >  };
>> > > diff --git a/fs/nfsd/nfs4layouts.c b/fs/nfsd/nfs4layouts.c
>> > > index 6d98d16b3354..2be9602b0221 100644
>> > > --- a/fs/nfsd/nfs4layouts.c
>> > > +++ b/fs/nfsd/nfs4layouts.c
>> > > @@ -139,21 +139,21 @@ void nfsd4_setup_layout_type(struct svc_export *exp)
>> > >  	 * otherwise advertise the block layout.
>> > >  	 */
>> > >  #ifdef CONFIG_NFSD_FLEXFILELAYOUT
>> > > -	exp->ex_layout_type = LAYOUT_FLEX_FILES;
>> > > +	exp->ex_layout_types |= 1 << LAYOUT_FLEX_FILES;
>> > >  #endif
>> > >  #ifdef CONFIG_NFSD_BLOCKLAYOUT
>> > >  	/* overwrite flex file layout selection if needed */
>> > >  	if (sb->s_export_op->get_uuid &&
>> > >  	    sb->s_export_op->map_blocks &&
>> > >  	    sb->s_export_op->commit_blocks)
>> > > -		exp->ex_layout_type = LAYOUT_BLOCK_VOLUME;
>> > > +		exp->ex_layout_types |= 1 << LAYOUT_BLOCK_VOLUME;
>> > >  #endif
>> > >  #ifdef CONFIG_NFSD_SCSILAYOUT
>> > >  	/* overwrite block layout selection if needed */
>> > >  	if (sb->s_export_op->map_blocks &&
>> > >  	    sb->s_export_op->commit_blocks &&
>> > >  	    sb->s_bdev && sb->s_bdev->bd_disk->fops->pr_ops)
>> > > -		exp->ex_layout_type = LAYOUT_SCSI;
>> > > +		exp->ex_layout_types |= 1 << LAYOUT_SCSI;
>> > >  #endif
>> > >  }
>> > >  
>> > > diff --git a/fs/nfsd/nfs4proc.c b/fs/nfsd/nfs4proc.c
>> > > index 2ee2dc170327..719c620753e2 100644
>> > > --- a/fs/nfsd/nfs4proc.c
>> > > +++ b/fs/nfsd/nfs4proc.c
>> > > @@ -1219,12 +1219,12 @@ nfsd4_verify(struct svc_rqst *rqstp, struct
>> > > nfsd4_compound_state *cstate,
>> > >  static const struct nfsd4_layout_ops *
>> > >  nfsd4_layout_verify(struct svc_export *exp, unsigned int layout_type)
>> > >  {
>> > > -	if (!exp->ex_layout_type) {
>> > > +	if (!exp->ex_layout_types) {
>> > >  		dprintk("%s: export does not support pNFS\n", __func__);
>> > >  		return NULL;
>> > >  	}
>> > >  
>> > > -	if (exp->ex_layout_type != layout_type) {
>> > > +	if (!(exp->ex_layout_types & (1 << layout_type))) {
>> > >  		dprintk("%s: layout type %d not supported\n",
>> > >  			__func__, layout_type);
>> > >  		return NULL;
>> > > diff --git a/fs/nfsd/nfs4xdr.c b/fs/nfsd/nfs4xdr.c
>> > > index 9df898ba648f..4d3ae75ad4c8 100644
>> > > --- a/fs/nfsd/nfs4xdr.c
>> > > +++ b/fs/nfsd/nfs4xdr.c
>> > > @@ -2164,22 +2164,20 @@ nfsd4_encode_aclname(struct xdr_stream *xdr, struct
>> > > svc_rqst *rqstp,
>> > >  }
>> > >  
>> > >  static inline __be32
>> > > -nfsd4_encode_layout_type(struct xdr_stream *xdr, enum pnfs_layouttype
>> > > layout_type)
>> > > +nfsd4_encode_layout_types(struct xdr_stream *xdr, u32 layout_types)
>> > >  {
>> > > -	__be32 *p;
>> > > +	__be32		*p;
>> > > +	unsigned long	i = hweight_long(layout_types);
>> > >  
>> > > -	if (layout_type) {
>> > > -		p = xdr_reserve_space(xdr, 8);
>> > > -		if (!p)
>> > > -			return nfserr_resource;
>> > > -		*p++ = cpu_to_be32(1);
>> > > -		*p++ = cpu_to_be32(layout_type);
>> > > -	} else {
>> > > -		p = xdr_reserve_space(xdr, 4);
>> > > -		if (!p)
>> > > -			return nfserr_resource;
>> > > -		*p++ = cpu_to_be32(0);
>> > > -	}
>> > > +	p = xdr_reserve_space(xdr, 4 + 4 * i);
>> > > +	if (!p)
>> > > +		return nfserr_resource;
>> > > +
>> > > +	*p++ = cpu_to_be32(i);
>> > > +
>> > > +	for (i = LAYOUT_NFSV4_1_FILES; i < LAYOUT_TYPE_MAX; ++i)
>> > > +		if (layout_types & (1 << i))
>> > > +			*p++ = cpu_to_be32(i);
>> > >  
>> > >  	return 0;
>> > >  }
>> > > @@ -2754,13 +2752,13 @@ out_acl:
>> > >  	}
>> > >  #ifdef CONFIG_NFSD_PNFS
>> > >  	if (bmval1 & FATTR4_WORD1_FS_LAYOUT_TYPES) {
>> > > -		status = nfsd4_encode_layout_type(xdr, exp->ex_layout_type);
>> > > +		status = nfsd4_encode_layout_types(xdr, exp->ex_layout_types);
>> > >  		if (status)
>> > >  			goto out;
>> > >  	}
>> > >  
>> > >  	if (bmval2 & FATTR4_WORD2_LAYOUT_TYPES) {
>> > > -		status = nfsd4_encode_layout_type(xdr, exp->ex_layout_type);
>> > > +		status = nfsd4_encode_layout_types(xdr, exp->ex_layout_types);
>> > >  		if (status)
>> > >  			goto out;
>> > >  	}
>> > 
>> > 
>> > Just a (lightly-tested) RFC patch for now.
>> > 
>> > This seems to do the right thing WRT to GETATTR response that requests
>> > a layout types list. The current client code spews this into the ring
>> > buffer though:
>> > 
>> >     NFS: decode_first_pnfs_layout_type: Warning: Multiple pNFS layout drivers per
>> >     filesystem not supported
>> > 
>> > ...so that would need a little work. Wireshark also seems to not parse
>> > the layout types list correctly.
>> 
>> Hi Jeff,
>> 
>> I had an attempt to add multi-layout support into the client:
>> 
>> https://github.com/0day-ci/linux/commit/e2d792dc32b82ffc511b009c0a8db5313126888e
>> 
>> I can't find it in the list archive. This explains why Trond never
>> respond to it.
>> 
>> Tigran.
>> 
> 
> Thanks. I had already rolled one up that does a different approach of a
> hardcoded selection order in the client code and sent it as an RFC.
> 
> So...yours assumes that the server will send the "default" one first in
> the list, but AFAICT, the spec doesn't say anything about the
> fs_layout_type list being ordered in any way. I don't think we can make
> that assumption. Am I wrong there?

Imagine a muti-layout server in a deployment, where some client are old (RHEL6)
and some clients are new. RHEL6 will always take only the first one and fail, if
it wasn't nfs4_file_layout. But, if we always return 'well-known' as a first one,
then, then RHEL6 clients will be able to coexists with newer clients.

Our server can provide flexfile layouts, but 99% of our clients are RHEL6-clones.
This stops us from enable flexfile support, unless we add per-client based control
of layout type.

Tigran.

> 
> What we probably ought to do is to plumb the selection in at a
> different layer -- i.e., when you go to actually do the I/O. That way,
> you could get a SCSI or block layout, figure out that you don't
> actually have access to the device and fall back to using flexfiles or
> files.
> --
> Jeff Layton <jlayton@poochiereds.net>
> --
> To unsubscribe from this list: send the line "unsubscribe linux-nfs" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html