https://bugzilla.kernel.org/show_bug.cgi?id=197861 Willem de Bruijn (willemb@xxxxxxxxxx) changed: What |Removed |Added ---------------------------------------------------------------------------- CC| |willemb@xxxxxxxxxx --- Comment #6 from Willem de Bruijn (willemb@xxxxxxxxxx) --- Does reverting only that patch resolve the issue? The logic in it is quite complex, but it is only needed for zerocopy with MSG_ZEROCOPY. And then only in edge cases. This is likely not used here. The code is blocking on vhost_net use of zerocopy. Which does not build skbuffs with zerocopy data in compound pages. The patch adds checks against shared and cloned skbs that were not present before. It is not safe to modify skb frags[] on on either, but perhaps this changed return path causes buffers to not be released, causing vhost_net_ubuf_put_and_wait to wait seemingly indefinitely. + if (skb_shared(skb) || skb_unclone(skb, gfp_mask)) + return -EINVAL; -- You are receiving this mail because: You are watching the assignee of the bug.
