Re: ppp/pppoe, still panic 4.15.3 in ppp_push

On Thu, Mar 01, 2018 at 10:07:05PM +0200, Denys Fedoryshchenko wrote:

On 2018-03-01 22:01, Guillaume Nault wrote:

quoted

diff --git a/drivers/net/ppp/ppp_generic.c

b/drivers/net/ppp/ppp_generic.c
index 255a5def56e9..2acf4b0eabd1 100644

--- a/drivers/net/ppp/ppp_generic.c
+++ b/drivers/net/ppp/ppp_generic.c

@@ -3161,6 +3161,15 @@ ppp_connect_channel(struct channel *pch, int

unit)
 		goto outl;

 	ppp_lock(ppp);
+	spin_lock_bh(&pch->downl);
+	if (!pch->chan) {
+		/* Don't connect unregistered channels */
+		ppp_unlock(ppp);
+		spin_unlock_bh(&pch->downl);

This is obviously wrong. It should have been
+		spin_unlock_bh(&pch->downl);
+		ppp_unlock(ppp);

Sorry, I shouldn't have hurried.
This is fixed in the official version.

quoted

+		ret = -ENOTCONN;
+		goto outl;
+	}
+	spin_unlock_bh(&pch->downl);
 	if (pch->file.hdrlen > ppp->file.hdrlen)
 		ppp->file.hdrlen = pch->file.hdrlen;
 	hdrlen = pch->file.hdrlen + 2;	/* for protocol bytes */

Ok, i will try to test that at night.
Thanks a lot! For me also problem solved anyway by removing unit-cache, just
i think it's nice to have bug fixed :)

I think this bug has been there forever, indeed it's good to have it fixed.
Thanks a lot for your help (and patience!).

FYI, if you see accel-ppp logs like
"ioctl(PPPIOCCONNECT): Transport endpoint is not connected", then that
means the patch prevented the scenario that was leading to the original
crash.

Out of curiosity, did unit-cache really bring performance improvements
on your workload?