[Nbd] deadlock in nbd?

[Peter Daum <gator_ml@...141...>]

I am trying to run nbd, without any luck: whatever I tried, I always
 after a while had to reboot the (client) machine because of processes
 stuck forever in uninteruptable sleep.  This happens on several
 different machines running different kernel versions (2.6.16.54,
 2.6.22.1, 2.6.18-5 Debian Etch) and with different nbd versions
 (2.9.7, 1:2.8.7-4 from Debian Etch).
 
 I tried to isolate it from any networking peculiarities (All machines
 use VLANs, most also interface bonding), but it seems that nbd
 generally doesn't work over loppback interfaces. I also loaded the
 module with "debugflags=0xffff" trying to get all possible debug
 information, but the logged data looks pretty garbled (example:
  <7quest c0f92e94: sending 4096 bytes data
  <7uest e2c0fc90: sending 4096 bytes data
   data
  <5)
  <7quest e3510c90: sending 4096 bytes data
  ...
 ) and when it is starting
 to hang, nothing is logged. On the server side,
 there generally aren't any errors in the syslog. So unfortunately,
 I can't give a very detailed description. Here a pretty typical example
 of what I am doing:
 # server : # nbd-server 666 /dev/some/lvm/volume
 # client : # nbd-client bs=4096 host 666 /dev/nbd0
 # client : # mke2fs -j /dev/nbd0
 
 Often, this will get stuck with the last message on the screen
 "Writing superblocks and filesystem accounting information".
 The mke2fs process is waiting on "sync_buffer", additionally 
 there are several pdflush threads stuck in "blk_congestion_wait".
 
 Any ideas, what is going on and how to fix it?
 
 Regards,
                        Peter Daum

---

Ihr erstes Fernweh? Wo gibt es den schönsten Strand.