[PATCH] bug in old-cache handling in mountd.

[PATCH] bug in old-cache handling in mountd.

[Neil Brown]

Hi,
 there is a bug in the old-cache behaviour of mountd.
 Hopefully no-one is using this but you never know....

 The bug was reported to redhat
    https://bugzilla.redhat.com/show_bug.cgi?id=164186

 It was fixed incorrectly and never (that I recall) reported
 upstream. :-(

 A SLES customer thinks they have hit the same problem, which is why
 I'm looking at it.  (They shouldn't really hit it because that
 version of sles uses 'new_cache' but maybe nfsdfs was disabled
 somehow - don't know yet).

 Anyway, when mountd gets a mount request, it tells the kernel about it
 and remembers that it told the kernel so it doesn't have to tell the
 kernel again.
 It is possible that "exportfs -r" will tell the kernel to forget that
 information if the client has since unmounted.
 So mountd really needs to tell the kernel even if it thinks the
 kernel should already know.

 The following patch does this if the kernel refuses to provide a
 valid filehandle.

 Does RHEL3 use old-cache or new-cache ??

Thanks,
NeilBrown

From: NeilBrown <neilb@suse.de>
Subject: retry export if getfh fails.

mountd tries to avoid telling the kernel to export something
when the kernel already knows to do that.
However sometimes (exportfs -r) the kernel can be told
to forget something without mountd realising.
So if mountd finds that it cannot get a  valid filehandle,
make sure it really has been exported to the kernel.

This only applies if the nfsd filesystem is not mounted.

Signed-off-by: NeilBrown <neilb@suse.de>

diff --git a/utils/mountd/mountd.c b/utils/mountd/mountd.c
index 25d292b..b59f939 100644
--- a/utils/mountd/mountd.c
+++ b/utils/mountd/mountd.c
@@ -467,8 +467,12 @@ get_rootfh(struct svc_req *rqstp, dirpath *path, nfs_export **expret,
 			return NULL;
 		}
 	} else {
-		if (exp->m_exported<1)
+		int did_export = 0;
+	retry:
+		if (exp->m_exported<1) {
 			export_export(exp);
+			did_export = 1;
+		}
 		if (!exp->m_xtabent)
 			xtab_append(exp);
 
@@ -482,6 +486,11 @@ get_rootfh(struct svc_req *rqstp, dirpath *path, nfs_export **expret,
 				fh = getfh_old ((struct sockaddr *) sin,
 						stb.st_dev, stb.st_ino);
 		}
+		if (fh == NULL && !did_export) {
+			exp->m_exported = 0;
+			goto retry;
+		}
+
 		if (fh == NULL) {
 			xlog(L_WARNING, "getfh failed: %s", strerror(errno));
 			*error = NFSERR_ACCES;

Re: [PATCH] bug in old-cache handling in mountd.

[Steve Dickson]

Neil Brown wrote:
> Hi,
>  there is a bug in the old-cache behaviour of mountd.
>  Hopefully no-one is using this but you never know....
> 
>  The bug was reported to redhat
>     https://bugzilla.redhat.com/show_bug.cgi?id=164186
> 
>  It was fixed incorrectly and never (that I recall) reported
>  upstream. :-(
> 
>  A SLES customer thinks they have hit the same problem, which is why
>  I'm looking at it.  (They shouldn't really hit it because that
>  version of sles uses 'new_cache' but maybe nfsdfs was disabled
>  somehow - don't know yet).
> 
>  Anyway, when mountd gets a mount request, it tells the kernel about it
>  and remembers that it told the kernel so it doesn't have to tell the
>  kernel again.
>  It is possible that "exportfs -r" will tell the kernel to forget that
>  information if the client has since unmounted.
>  So mountd really needs to tell the kernel even if it thinks the
>  kernel should already know.
> 
>  The following patch does this if the kernel refuses to provide a
>  valid filehandle.
> 
>  Does RHEL3 use old-cache or new-cache ??
The new cache, if it exists...

Committed..

steved.

Re: [PATCH] bug in old-cache handling in mountd.

[Neil Brown]

On Wednesday June 3, SteveD@redhat.com wrote:
> 
> 
> Neil Brown wrote:
> > Hi,
> >  there is a bug in the old-cache behaviour of mountd.
> >  Hopefully no-one is using this but you never know....
> > 
> >  The bug was reported to redhat
> >     https://bugzilla.redhat.com/show_bug.cgi?id=164186
> > 
> >  It was fixed incorrectly and never (that I recall) reported
> >  upstream. :-(
> > 
> >  A SLES customer thinks they have hit the same problem, which is why
> >  I'm looking at it.  (They shouldn't really hit it because that
> >  version of sles uses 'new_cache' but maybe nfsdfs was disabled
> >  somehow - don't know yet).
> > 
> >  Anyway, when mountd gets a mount request, it tells the kernel about it
> >  and remembers that it told the kernel so it doesn't have to tell the
> >  kernel again.
> >  It is possible that "exportfs -r" will tell the kernel to forget that
> >  information if the client has since unmounted.
> >  So mountd really needs to tell the kernel even if it thinks the
> >  kernel should already know.
> > 
> >  The following patch does this if the kernel refuses to provide a
> >  valid filehandle.
> > 
> >  Does RHEL3 use old-cache or new-cache ??
> The new cache, if it exists...

That's what I would have thought... which leave me wondering how both
your customer and our customer hit the bug...  I must be missing
something important.

> 
> Committed..

Thanks.

NeilBrown