Re: Spurious TCP retransmissions on ack vs kfree_skb reordering

On Fri, 26 Feb 2021 17:35:11 +0100 Eric Dumazet wrote:

On Fri, Feb 26, 2021 at 5:09 PM Jakub Kicinski [off-list ref] wrote:

quoted

On Fri, 26 Feb 2021 11:41:22 +0100 Eric Dumazet wrote:  

quoted

Yes, this packetdrill test confirms TCP INFO stuff seems correct .  

Looks like it's TcpExtTCPSpuriousRtxHostQueues - the TFO fails as it
might, but at the time the syn is still not kfree_skb()d because of
the IRQ coalescing settings, so __tcp_retransmit_skb() returns -EBUSY
and we have to wait for a timeout.

Credit to Neil Spring @FB for figuring it out.  

Yes, this makes sense.

Presumably tcp_send_syn_data() could allocate a regular (non fclone)
skb, to avoid this.

But if skb_still_in_host_queue() returns true, __tcp_retransmit_skb()
should return -EBUSY
and your tracepoint should not be called ?

Right, looking at the stack trace the call I was tracing comes later
from the RTO timer.

In anycase, the bytes_acked should not be 742 as mentioned in your
email, if only the SYN was acked ?

You're right, it's 1. I did a braino in the trace print yesterday.