create_session bugfix

create_session bugfix

[J. Bruce Fields]

I'm queueing this up for 2.6.33 if there's no objections.

--b.

[PATCH 1/2] nfsd4.1: fix session memory use calculation

[J. Bruce Fields]

Unbalanced calculations on creation and destruction of sessions could
cause our estimate of cache memory used to become negative, sometimes
resulting in spurious SERVERFAULT returns to client CREATE_SESSION
requests.

Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu>
---
 fs/nfsd/nfs4state.c |    5 ++++-
 1 files changed, 4 insertions(+), 1 deletions(-)

diff --git a/fs/nfsd/nfs4state.c b/fs/nfsd/nfs4state.c
index fcb9817..c171371 100644
--- a/fs/nfsd/nfs4state.c
+++ b/fs/nfsd/nfs4state.c
@@ -629,10 +629,13 @@ void
 free_session(struct kref *kref)
 {
 	struct nfsd4_session *ses;
+	int mem;
 
 	ses = container_of(kref, struct nfsd4_session, se_ref);
 	spin_lock(&nfsd_drc_lock);
-	nfsd_drc_mem_used -= ses->se_fchannel.maxreqs * NFSD_SLOT_CACHE_SIZE;
+	mem = ses->se_fchannel.maxreqs
+		* (ses->se_fchannel.maxresp_cached - NFSD_MIN_HDR_SEQ_SZ);
+	nfsd_drc_mem_used -= mem;
 	spin_unlock(&nfsd_drc_lock);
 	free_session_slots(ses);
 	kfree(ses);
-- 
1.6.3.3

[PATCH 2/2] nfsd4.1: common slot allocation size calculation

[J. Bruce Fields]

We do the same calculation in a couple places; use a helper function,
and add a little documentation, in the hopes of preventing bugs like
that fixed in the last patch.

Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu>
---
 fs/nfsd/nfs4state.c |   14 +++++++++++---
 1 files changed, 11 insertions(+), 3 deletions(-)

diff --git a/fs/nfsd/nfs4state.c b/fs/nfsd/nfs4state.c
index c171371..42dab95 100644
--- a/fs/nfsd/nfs4state.c
+++ b/fs/nfsd/nfs4state.c
@@ -524,6 +524,15 @@ free_session_slots(struct nfsd4_session *ses)
 		kfree(ses->se_slots[i]);
 }
 
+/*
+ * We don't actually need to cache the rpc and session headers, so we
+ * can allocate a little less for each slot:
+ */
+static inline int slot_bytes(struct nfsd4_channel_attrs *ca)
+{
+	return ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
+}
+
 static int
 alloc_init_session(struct svc_rqst *rqstp, struct nfs4_client *clp,
 		   struct nfsd4_create_session *cses)
@@ -555,7 +564,7 @@ alloc_init_session(struct svc_rqst *rqstp, struct nfs4_client *clp,
 	memcpy(new, &tmp, sizeof(*new));
 
 	/* allocate each struct nfsd4_slot and data cache in one piece */
-	cachesize = new->se_fchannel.maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
+	cachesize = slot_bytes(&new->se_fchannel);
 	for (i = 0; i < new->se_fchannel.maxreqs; i++) {
 		sp = kzalloc(sizeof(*sp) + cachesize, GFP_KERNEL);
 		if (!sp)
@@ -633,8 +642,7 @@ free_session(struct kref *kref)
 
 	ses = container_of(kref, struct nfsd4_session, se_ref);
 	spin_lock(&nfsd_drc_lock);
-	mem = ses->se_fchannel.maxreqs
-		* (ses->se_fchannel.maxresp_cached - NFSD_MIN_HDR_SEQ_SZ);
+	mem = ses->se_fchannel.maxreqs * slot_bytes(&ses->se_fchannel);
 	nfsd_drc_mem_used -= mem;
 	spin_unlock(&nfsd_drc_lock);
 	free_session_slots(ses);
-- 
1.6.3.3

Re: [pnfs] create_session bugfix

[J. Bruce Fields]

On Tue, Oct 27, 2009 at 07:07:40PM -0400, J. Bruce Fields wrote:
> 
> I'm queueing this up for 2.6.33 if there's no objections.

Err.  I meant to send that to linux-nfs, not linux-fsdevel, apologies.
Oh well, pnfs@linux-nfs.org should have anyone that would care anyway.

Perhaps I need more coffee.  And fewer mailing lists.

--b.

Re: [pnfs] create_session bugfix

[Benny Halevy]

On Oct. 28, 2009, 1:07 +0200, "J. Bruce Fields" <bfields@citi.umich.edu> wrote:
> I'm queueing this up for 2.6.33 if there's no objections.

I've merged both patches also into the pnfs tree so they
get more soak time.

Thanks!

Benny

> 
> --b.
> _______________________________________________
> pNFS mailing list
> pNFS@linux-nfs.org
> http://linux-nfs.org/cgi-bin/mailman/listinfo/pnfs