Re: [PATCH 19/20] xfs: implement pNFS export operations

[Christoph Hellwig <hch@lst.de>]

On Thu, Feb 05, 2015 at 11:47:58AM +1100, Dave Chinner wrote:
> > +++ b/fs/xfs/Makefile
> > @@ -121,3 +121,4 @@ xfs-$(CONFIG_XFS_POSIX_ACL)	+= xfs_acl.o
> >  xfs-$(CONFIG_PROC_FS)		+= xfs_stats.o
> >  xfs-$(CONFIG_SYSCTL)		+= xfs_sysctl.o
> >  xfs-$(CONFIG_COMPAT)		+= xfs_ioctl32.o
> > +xfs-$(CONFIG_NFSD_PNFS)		+= xfs_pnfs.o
> 
> ... because I'll have to jump through hoops to get it to compile, I
> think.

Just get the whole git tree from

	git://git.infradead.org/users/hch/pnfs.git pnfsd-for-3.20-3

> > diff --git a/fs/xfs/xfs_fsops.c b/fs/xfs/xfs_fsops.c
> > index 74c6211..99465ba 100644
> > --- a/fs/xfs/xfs_fsops.c
> > +++ b/fs/xfs/xfs_fsops.c
> > @@ -602,6 +602,8 @@ xfs_growfs_data(
> >  	if (!mutex_trylock(&mp->m_growlock))
> >  		return -EWOULDBLOCK;
> >  	error = xfs_growfs_data_private(mp, in);
> > +	if (!error)
> > +		mp->m_generation++;
> >  	mutex_unlock(&mp->m_growlock);
> >  	return error;
> >  }
> 
> Even on error I think we should bump this. Errors can come from
> secondary superblock updates after the filesystem has been grown,
> hence an error is not a reliable indication of whether the layout
> has changed or not.

Ok.
> 
> > +int
> > +xfs_fs_get_uuid(
> > +	struct super_block	*sb,
> > +	u8			*buf,
> > +	u32			*len,
> > +	u64			*offset)
> > +{
> > +	struct xfs_mount	*mp = XFS_M(sb);
> > +
> > +	if (*len < sizeof(uuid_t))
> > +		return -EINVAL;
> > +
> > +	memcpy(buf, &mp->m_sb.sb_uuid, sizeof(uuid_t));
> > +	*len = sizeof(uuid_t);
> > +	*offset = offsetof(struct xfs_dsb, sb_uuid);
> 
> What purpose does the offset serve here? I can't tell from the usage
> in the PNFS code. Can we ignore offset - as it seems entirely
> arbitrary - and still have this work? Either way, comment please.

The get_uuid methods gets content and location of the uuid so that
the client can find the disks.  The offset simply is part of the wire
protocol.

> I'd like to see a IOMAP_NULL_BLOCK define here for the -1 value,
> say:
> 
> #define IOMAP_NULL_BLOCK	-1LL

Ok.

> OK, so we are not mapping realtime inodes here. Any specific reason?

The realtime device doesn't have a way for the client to find it, as
it doesn't have its own superblockuuids.

> FWIW, that also means you can use XFS_FSB_TO_DADDR() in the iomap
> mapping as xfs_fsb_to_db() is only needed if we might be mapping
> realtime extents...

ok.

> > +	tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_NOT_SIZE);
> > +	error = xfs_trans_reserve(tp, &M_RES(mp)->tr_ichange, 0, 0);
> > +	if (error)
> > +		goto out_drop_iolock;
> > +
> > +	xfs_ilock(ip, XFS_ILOCK_EXCL);
> > +	xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
> > +	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
> > +
> > +	xfs_setattr_time(ip, iattr);
> > +	if (iattr->ia_valid & ATTR_SIZE) {
> > +		if (iattr->ia_size > i_size_read(inode)) {
> > +			i_size_write(inode, iattr->ia_size);
> > +			ip->i_d.di_size = iattr->ia_size;
> > +		}
> > +	}
> 
> The concern I have about this is that extending the file size can
> expose uninitialised blocks beyond the old EOF. That can happen if
> delayed allocation has previously been done on the file and we
> haven't trimmed the excess beyond EOF back yet. I know the pnfs
> server is not aimed at mixed usage, but it still makes me
> uncomfortable in the case where you have normal NFS and PNFS clients
> accessing the same files...

The protocol only allows to commit to a size that we previously
returned a layout for, which means we already have allocated space
for it at same time.  For robustness reasons a sanity check might make
sense, though.

> > +	xfs_trans_set_sync(tp);
> > +	error = xfs_trans_commit(tp, 0);
> 
> I just had a thought about these sync transctions - could the NFS
> server handle persistence of the maps via ->commit_metadata?

It probably could, but that wouldn't buy us anything over just committing
the transaction synchronously.