Thank you for the explanation. You're right – I see now that netdev.c pre-allocates enough batches based on the total segment count, so the batch will not overflow in this path.
Sorry for the noise, and thanks for the clarification. SeungJu On Wed, Jan 21, 2026 at 7:39 PM David Marchand <[email protected]> wrote: > Hello, > > On Wed, 21 Jan 2026 at 11:27, SeungJu Cheon <[email protected]> wrote: > > > > dp_packet_batch_add() may flush a full batch and free the packet. > > In dp_packet_gso__(), the original packet was added to the batch > > before all header and payload information was extracted, which could > > lead to accessing a freed packet. > > The generated packets are not sent as part of the segmentation / > putting in different batch objects. > > The code calling this function ensures there is enough space in the > batch where to add a reference to the first segment. > No "flush" or free should happen. > > So I don't understand the issue you want to fix here. > > > -- > David Marchand > > _______________________________________________ dev mailing list [email protected] https://mail.openvswitch.org/mailman/listinfo/ovs-dev
