Skip to contentSkip to conversation

->migratepage aops for nfs missing

DISCUSSION Andi Kleen 4 participants 8 messages Aug 4, 2009 – Aug 5, 2009 6 messages not in the archive

Subsystems (2)
lore
Share & tools
lore

Andi Kleen

author
#0 lore raw

Trond,

Any particular reason ->migratepage is not enabled for the nfs file aops?

Is it known to be unsafe to move NFS buffer pages at runtime?

I have a similar requirement for memory error handling.

Thanks,

-Andi

Signature · 1 line
[email protected] -- Speaking for myself only.

A message that is not in the archive (sent off-list or not yet received): 20090804202648.GA12484-u0/[email protected]

#2 lore raw
Quoted 5 lines · Andi Kleen

Trond,

Any particular reason ->migratepage is not enabled for the nfs file aops?

Quoted 7 lines

Is it known to be unsafe to move NFS buffer pages at runtime?

I have a similar requirement for memory error handling.

Thanks,

-Andi

Hi Andi,

I don't think there are any fundamental problems with migrating NFS page cache pages. In order to handle dirty pages, we'd need to write the equivalent of buffer_migrate_page() in order to transfer the private data, lock and update the 'struct nfs_page' that tracks the page's dirty area.

The only thing I'm not 100% sure about is fscache, but I think the correct thing is just to call nfs_fscache_release_page(). David?

Cheers
Trond

Signature · 6 lines
Trond Myklebust
Linux NFS client maintainer

NetApp
[email protected]
www.netapp.com

A message that is not in the archive (sent off-list or not yet received): 1249419186.5377.20.camel-rJ7iovZKK19ZJLDQqaL3InhyD016LWXt@public.gmane.org

#4 lore raw
Quoted 2 lines · Trond Myklebust

The only thing I'm not 100% sure about is fscache, but I think the correct thing is just to call nfs_fscache_release_page(). David?

Yes, probably. Note that that will only release the page unconditionally if gfp includes __GFP_WAIT.

David

#5 lore raw
Quoted 7 lines · David Howells
Quoted 2 lines · Trond Myklebust

The only thing I'm not 100% sure about is fscache, but I think the correct thing is just to call nfs_fscache_release_page(). David?

Yes, probably. Note that that will only release the page unconditionally if gfp includes __GFP_WAIT.

Right, but I believe that migratepage does allow you to sleep, so that shouldn't be a problem.

Cheers
Trond

Signature · 6 lines
Trond Myklebust
Linux NFS client maintainer

NetApp
[email protected]
www.netapp.com

A message that is not in the archive (sent off-list or not yet received): 1249423594.12510.0.camel-rJ7iovZKK19ZJLDQqaL3InhyD016LWXt@public.gmane.org

#7 lore raw
Quoted 6 lines · Trond Myklebust
Quoted 5 lines · David Howells
Quoted 2 lines · Trond Myklebust

The only thing I'm not 100% sure about is fscache, but I think the correct thing is just to call nfs_fscache_release_page(). David?

Quoted 8 lines
Quoted 2 lines

Yes, probably. Note that that will only release the page unconditionally if gfp includes __GFP_WAIT.

Right, but I believe that migratepage does allow you to sleep, so that shouldn't be a problem.

Cheers
Trond

Something like this ought to do it then...

Cheers
Trond
-----------------------------------------------------------------------
From: Trond Myklebust <[email protected]>
Subject: NFS: Add a ->migratepage() aop for NFS

Make NFS a bit more NUMA-friendly...

---

 fs/nfs/file.c     |    1 +
 fs/nfs/internal.h |    6 +++
 fs/nfs/write.c    |   92 ++++++++++++++++++++++++++++++++++++++++-------------
 3 files changed, 77 insertions(+), 22 deletions(-)
fs/nfs/file.c +1 -0
index 0506232..dfc8967 100644
@@ -479,6 +479,7 @@ const struct address_space_operations nfs_file_aops = {
 	.invalidatepage = nfs_invalidate_page,
 	.releasepage = nfs_release_page,
 	.direct_IO = nfs_direct_IO,
+	.migratepage = nfs_migrate_page,
 	.launder_page = nfs_launder_page,
 };
 
fs/nfs/internal.h +6 -0
index 7dd90a6..e2ccb4a 100644
@@ -248,6 +248,12 @@ extern void nfs_read_prepare(struct rpc_task *task, void *calldata);
 
 /* write.c */
 extern void nfs_write_prepare(struct rpc_task *task, void *calldata);
+#ifdef CONFIG_MIGRATION
+extern int nfs_migrate_page(struct address_space *,
+		struct page *, struct page *);
+#else
+#define nfs_migrate_page NULL
+#endif
 
 /* nfs4proc.c */
 extern int _nfs4_call_sync(struct nfs_server *server,
fs/nfs/write.c +70 -22

135 lines not on this page · raw message

A message that is not in the archive (sent off-list or not yet received): 1249430567.12510.3.camel-rJ7iovZKK19ZJLDQqaL3InhyD016LWXt@public.gmane.org

author
#9 lore raw
Quoted 1 line

Make NFS a bit more NUMA-friendly...

Minor nit :- page migration is not only used for NUMA.

-Andi

A message that is not in the archive (sent off-list or not yet received): 20090805073036.GC11385-u0/[email protected]

#11 lore raw
Quoted 5 lines · Andi Kleen
Quoted 1 line

Make NFS a bit more NUMA-friendly...

Minor nit :- page migration is not only used for NUMA.

-Andi

How about "Make NFS a bit more NUMA and memory hot removal-friendly..."? :-)

A message that is not in the archive (sent off-list or not yet received): 1249474573.5385.0.camel-rJ7iovZKK19ZJLDQqaL3InhyD016LWXt@public.gmane.org

author
#13 lore raw
Quoted 9 lines · Trond Myklebust
Quoted 5 lines · Andi Kleen
Quoted 1 line

Make NFS a bit more NUMA-friendly...

Minor nit :- page migration is not only used for NUMA.

-Andi

How about "Make NFS a bit more NUMA and memory hot removal-friendly..."? :-)

Thanks. Looks better.

But I'm actually adding a third user soon which is neither.

-Andi

Signature · 1 line
[email protected] -- Speaking for myself only.