Attempt to address lagg race in PR234864
AbandonedPublic
Actions

Authored by shurd on Jan 16 2019, 10:34 PM.

Details

Reviewers

Summary

A naive look at the if_lagg.c race suggests that extending
the RLOCK coverage over the entire transaction may resolve the problem.

Test Plan

Patch added to PR234846 for testing

Diff Detail

Repository

rS FreeBSD src repository - subversion

Lint

Lint Passed

Unit

No Test Coverage

Build Status

Buildable 21988
Build 21222: arc lint + arc unit

Event Timeline

shurd created this revision.Jan 16 2019, 10:34 PM

Herald added subscribers: ae, imp. · View Herald TranscriptJan 16 2019, 10:34 PM

Harbormaster completed remote builds in B21988: Diff 52901.Jan 16 2019, 10:34 PM

RLOCK() before setting ifp->if_type

Harbormaster completed remote builds in B21989: Diff 52902.Jan 16 2019, 10:36 PM

Can you explain your goal? Not sure what you are trying to solve with this rlock. Please, note that LAGG_RLOCK() is epoch_enter() and it seem used to make safe lagg_port reclamation using epoch_call().

So the basic goal is to have the test at sys/net/if_ethersubr.c:583 see a consistent state of the changes from the /* Change the interface type */ line to the lagg_proto_addport() line.

The naive thinking that led to this was "Oh, there's something with LOCK in it inside that block, maybe if I just expand the coverage of that, everything will be fine."

I'm taking a real look at the problem now.

shurd abandoned this revision.Jan 18 2019, 5:42 PM