Re: kern/49072: in-kernel iscsi initiator not working

[Greg Oster <oster%cs.usask.ca@localhost>]

The following reply was made to PR kern/49072; it has been noted by GNATS.

From: Greg Oster <oster%cs.usask.ca@localhost>
To: gnats-bugs%NetBSD.org@localhost
Cc: 
Subject: Re: kern/49072: in-kernel iscsi initiator not working
Date: Tue, 5 Aug 2014 08:57:39 -0600

 On Mon,  4 Aug 2014 23:30:00 +0000 (UTC)
 zafer%aydogan.de@localhost wrote:
 
 > >Number:         49072
 > >Category:       kern
 > >Synopsis:       in-kernel iscsi initiator not working
 > >Confidential:   no
 > >Severity:       serious
 > >Priority:       high
 > >Responsible:    kern-bug-people
 > >State:          open
 > >Class:          sw-bug
 > >Submitter-Id:   net
 > >Arrival-Date:   Mon Aug 04 23:30:00 +0000 2014
 > >Originator:     Zafer Aydogan
 > >Release:        6.99.16 onward
 > >Organization:
 > TNF
 > >Environment:
 > 6.99.49 amd64
 > >Description:
 > The in-kernel iscsi initiator looks like to be broken since 6.99.16.
 > Symptoms are scsi SENSE KEY errors with "Overlapped Commands
 > Attempted" when you write data to disk.
 > If you continue to write data to the iscsi disk, the write process
 > will wedge in tstile. In this state a reboot will not succeed. You
 > need to power-cycle. Sometimes writing to the disk will panic the
 > system, especially overwriting the same file frequently.
 > 
 > I could trace the error back to two commits from mlelstv@ on
 > 2012-12-29:
 > http://freshbsd.org/commit/netbsd/0407d4dcf2917f3861618fc1cbc54e6bd5b583b3
 > and
 > http://freshbsd.org/commit/netbsd/77a0b440f42ba059ed279f043194a4b589203054
 > If you revert both or use a version prior to that date, iscsi works
 > as expected without any issues.
 > 
 > Screenshots:
 > http://www.aydogan.de/wtfs.png
 > http://www.aydogan.de/iscsi_panic.png
 > 
 > >How-To-Repeat:
 > mount an iscsi share and
 > newfs -O1 on a partition
 > 
 > newfs will exit with a wtfs write error
 > 
 > Attention: newfs -O2 will show no errors.
 > 
 > 
 > >Fix:
 > A workaround is to revert to the version before 2012-12-29 08:00:00
 
 If you can, two things to check:
 
 1) In your 'working' version, if you fix the ISCSI_THROTTLING_ENABLED
 #define in src/sys/dev/iscsi/iscsi_globals.h to make it look like this:
 
  #define ISCSI_THROTTLING_ENABLED      1
 
 do things continue to work?  (i.e. turn on the throttling code in the
 'working' code..)
 
 2) If you try -current using the following:
 
  #define ISCSI_THROTTLING_ENABLED      0
 
 in src/sys/dev/iscsi/iscsi_globals.h do things work? (i.e. disable the
 throttling code)
 
 It looks like the throttling bits weren't enabled (because of the typo
 in the #define ) originally.  They are now enabled, but with slightly
 different code.  The above checks might help narrow things down...
 
 Later...
 
 Greg Oster