Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
MIME-Version: 1.0
From:   Jeffrey Merkey <jeffmerkey@gmail.com>
Date:   Wed, 8 May 2019 21:38:55 -0600
Message-ID: <CA+ekxPVXW-wrSKXoULjJiAJDLNWGRJPe+2u6_YckrW7k2w1LYg@mail.gmail.com>
Subject: Question about dev_queue_xmit and guaranteed packet transmission
To:     linux-kernel <linux-kernel@vger.kernel.org>
Content-Type: text/plain; charset="UTF-8"
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

I noticed that Linux has been updated on later versions to always
consume the skb in calls to dev_queue_xmit, and may even drop the skb
after returning a return code of NETDEV_TX_OK which seems kind of
broken.  I have an application which requires guaranteed skb
transmission and the following code seems to work well for me if I
call the
netdev->netdev_ops->ndo_start_xmit function directly instead of using
dev_queue_xmit.

Is there a method of calling dev_queue_xmit which guarantees packet
transmission?  The current implementation of dev_queue_xmit used to
return error codes without consuming the skb, which allowed programs
which call it to simply resend the failing skb by calling
dev_queue_xmit again without having to recreate and repopulate the
skb.  There also appears to be no reliable way to know for certain if
dev_queue_xmit even sent a packet since it can at times return
NETDEV_TX_OK then turn around and drop the packet later on.

I have it working calling the ndo_start_xmit driver function directly
which does seem to guarantee  transmission, but would like to know if
there is a way to get guaranteed TX of skb's with dev_queue_xmit.

Jeff