2011-05-19 19:11:07

by Vitaliy Gusev

Subject: [PATCH] nfs41: Correct offset for LAYOUTCOMMIT

A client sends offset to MDS as it was seen by DS. As result,
file size after copy is only half of original file size in case
of 2 DS.

Signed-off-by: Vitaliy Gusev <[email protected]>
---
fs/nfs/nfs4filelayout.c | 1 +
1 files changed, 1 insertions(+), 0 deletions(-)

diff --git a/fs/nfs/nfs4filelayout.c b/fs/nfs/nfs4filelayout.c
index be79dc9..790eb1d 100644
--- a/fs/nfs/nfs4filelayout.c
+++ b/fs/nfs/nfs4filelayout.c
@@ -167,6 +167,7 @@ filelayout_set_layoutcommit(struct nfs_write_data *wdata)
wdata->res.verf->committed == NFS_FILE_SYNC)
return;

+ wdata->args.offset = wdata->mds_offset;
pnfs_set_layoutcommit(wdata);
dprintk("%s ionde %lu pls_end_pos %lu\n", __func__, wdata->inode->i_ino,
(unsigned long) wdata->lseg->pls_end_pos);
--
1.7.4.1

2011-05-19 20:02:15

by Vitaliy Gusev

Subject: Re: [PATCH] nfs41: Correct offset for LAYOUTCOMMIT

On 05/19/2011 11:38 PM, Trond Myklebust wrote:
> On Thu, 2011-05-19 at 23:10 +0400, Vitaliy Gusev wrote:
>> A client sends offset to MDS as it was seen by DS. As result,
>> file size after copy is only half of original file size in case
>> of 2 DS.
>>
>> Signed-off-by: Vitaliy Gusev<[email protected]>
>> ---
>> fs/nfs/nfs4filelayout.c | 1 +
>> 1 files changed, 1 insertions(+), 0 deletions(-)
>>
>> diff --git a/fs/nfs/nfs4filelayout.c b/fs/nfs/nfs4filelayout.c
>> index be79dc9..790eb1d 100644
>> --- a/fs/nfs/nfs4filelayout.c
>> +++ b/fs/nfs/nfs4filelayout.c
>> @@ -167,6 +167,7 @@ filelayout_set_layoutcommit(struct nfs_write_data *wdata)
>> wdata->res.verf->committed == NFS_FILE_SYNC)
>> return;
>>
>> + wdata->args.offset = wdata->mds_offset;
>> pnfs_set_layoutcommit(wdata);
>> dprintk("%s ionde %lu pls_end_pos %lu\n", __func__, wdata->inode->i_ino,
>> (unsigned long) wdata->lseg->pls_end_pos);
>
> Err... Shouldn't we rather be fixing pnfs_set_layoutcommit() to use the
> mds_offset instead of hacking args.offset?

I thought about it... pnfs_set_layoutcommit is a generic code for all
type of layouts, but mds_offset is set only for file layout.



>

2011-05-19 19:38:12

by Myklebust, Trond

Subject: Re: [PATCH] nfs41: Correct offset for LAYOUTCOMMIT

On Thu, 2011-05-19 at 23:10 +0400, Vitaliy Gusev wrote:
> A client sends offset to MDS as it was seen by DS. As result,
> file size after copy is only half of original file size in case
> of 2 DS.
>
> Signed-off-by: Vitaliy Gusev <[email protected]>
> ---
> fs/nfs/nfs4filelayout.c | 1 +
> 1 files changed, 1 insertions(+), 0 deletions(-)
>
> diff --git a/fs/nfs/nfs4filelayout.c b/fs/nfs/nfs4filelayout.c
> index be79dc9..790eb1d 100644
> --- a/fs/nfs/nfs4filelayout.c
> +++ b/fs/nfs/nfs4filelayout.c
> @@ -167,6 +167,7 @@ filelayout_set_layoutcommit(struct nfs_write_data *wdata)
> wdata->res.verf->committed == NFS_FILE_SYNC)
> return;
>
> + wdata->args.offset = wdata->mds_offset;
> pnfs_set_layoutcommit(wdata);
> dprintk("%s ionde %lu pls_end_pos %lu\n", __func__, wdata->inode->i_ino,
> (unsigned long) wdata->lseg->pls_end_pos);

Err... Shouldn't we rather be fixing pnfs_set_layoutcommit() to use the
mds_offset instead of hacking args.offset?

--
Trond Myklebust
Linux NFS client maintainer

NetApp
[email protected]
http://www.netapp.com

2011-05-19 19:56:14

by Myklebust, Trond

Subject: Re: [PATCH] nfs41: Correct offset for LAYOUTCOMMIT

On Thu, 2011-05-19 at 23:45 +0400, Vitaliy Gusev wrote:
> On 05/19/2011 11:38 PM, Trond Myklebust wrote:
> > On Thu, 2011-05-19 at 23:10 +0400, Vitaliy Gusev wrote:
> >> A client sends offset to MDS as it was seen by DS. As result,
> >> file size after copy is only half of original file size in case
> >> of 2 DS.
> >>
> >> Signed-off-by: Vitaliy Gusev<[email protected]>
> >> ---
> >> fs/nfs/nfs4filelayout.c | 1 +
> >> 1 files changed, 1 insertions(+), 0 deletions(-)
> >>
> >> diff --git a/fs/nfs/nfs4filelayout.c b/fs/nfs/nfs4filelayout.c
> >> index be79dc9..790eb1d 100644
> >> --- a/fs/nfs/nfs4filelayout.c
> >> +++ b/fs/nfs/nfs4filelayout.c
> >> @@ -167,6 +167,7 @@ filelayout_set_layoutcommit(struct nfs_write_data *wdata)
> >> wdata->res.verf->committed == NFS_FILE_SYNC)
> >> return;
> >>
> >> + wdata->args.offset = wdata->mds_offset;
> >> pnfs_set_layoutcommit(wdata);
> >> dprintk("%s ionde %lu pls_end_pos %lu\n", __func__, wdata->inode->i_ino,
> >> (unsigned long) wdata->lseg->pls_end_pos);
> >
> > Err... Shouldn't we rather be fixing pnfs_set_layoutcommit() to use the
> > mds_offset instead of hacking args.offset?
>
> I thought about it... pnfs_set_layoutcommit is a generic code for all
> type of layouts, but mds_offset is set only for file layout.

If we're going to use the struct nfs_write_data as the parameter to
pnfs_set_layoutcommit, then we can and should require that the fields be
correctly initialised. It's another matter altogether that the choice of
nfs_write_data may have been a poor one if other layouts don't use it.

--
Trond Myklebust
Linux NFS client maintainer

NetApp
[email protected]
http://www.netapp.com