Re: [PATCH net v3] xfrm: iptfs: avoid canceling reorder window drop timer

[Roshan Kumar]

Lilly, thanks for working on this. I need the same timer correction as a
prerequisite for the queued skb device lifetime fix.

While testing the interaction reported by Steffen/Sashiko, I reproduced a
second problem with only skipping the cancellation: a timer armed for a
completed reassembly can retain its old deadline and expire a subsequent
reassembly too early. The runt created reassembly path also needs to arm the
timer.

I have a version which records separate absolute deadlines for reassembly and
the reorder window, arms their shared hrtimer for the earlier deadline, and
expires only state whose own deadline has passed. A regression starts one
reassembly, queues a later reorder entry near its deadline, completes the
first reassembly, and verifies that the reorder entry remains until its own
deadline.

That regression now passes on both the regular KASAN and reference tracker
kernel and an enhanced lockdep and RCU debug kernel, with the queued entry
retained for its full five second interval. The runt created reassembly
regression passes as well.

I am preparing that as patch 1 of a two patch net series, retaining your
reporting credit and links to this patch and the review. If you would prefer to
post the corrected timer patch as v4 yourself, please say so and I will base
the device lifetime patch on it instead.