I'm not sure it's the same problem. Though it sounds interesting, what's
in your torrc? Try the enabling:
CloseHSClientCircuitsImmediatelyOnTimeout 1
CloseHSServiceRendCircuitsImmediatelyOnTimeout 1
UserspaceIOCPBuffers 1
AvoidDiskWrites 1
I'm not sure what happened to my setup quite yet I've been trying to
figure it out while staying current with snapshots. So following the
logs the virtual address get's allocated and automapped for the .onion
and tries to pass it off to rewrite_and_attach_if_allowed and that's
where it gets stuck. Thought it was a permission thing or a side effect
of w^x being turned on but I've since eliminated those possibilities. It
seems the function evdns_server_callback() is unable to pass the automap
to rewrite_and_attach_if_allowed and just continues in a loop trying
until timeout or max retries is hit. Not really sure how to proceed at
this very moment but I'll keep pondering away.
On 2016-05-31 16:52, Juuso Lapinlampi wrote:
This may be related: >
em(4) interface hangs randomly, receive buffer full (Intel i210)
https://marc.info/?l=openbsd-misc&m=145696725605233&w=2
I've been having those Tor hangs for months. The same issue, clock
skips
ahead some minutes and breaks Tor connections but not clearnet. They
happen quite frequently, at least three times a week to say. Heck, I've
seen this issue for more than a year since OpenBSD 5.6 or 5.7. The
issue
became worse late OpenBSD 5.8.
It seems like the issue gets worse when Tor's HardwareAccel option is
enabled (HardwareAccel 1). Disabling it has helped a lot, but hasn't
resolved the issue. I've also never been able to get more than ~1 MB/s
throughput for some reason with or without HardwareAccel on OpenBSD. No
idea why, probably the constantly breaking circuits.
I spent 2.5 months debugging the issue and it's still persisting. I've
posted a full email transcript with more details.[1]
It seems to be a little random which Tor connections get dropped. There
are IRC users connecting to an .onion address, and for years only some
of them have dropped connections and some kept going.
This server also uses OpenNTPD for syncing the clock from NTP.org
Sweden
pool. I haven't tried disabling it yet to see what would happen.
[1]:
https://partyvan.eu/transparency/emails/2016-02-16-portlane-nic-downtime.txt
