[Iscsitarget-devel] iscsi-target lockup with blockio and Solaris Volume Manager

[Chris S. <ck...@cs...>]

 I have a wierd lockup situation that I am hoping people can help
tell me how to diagnose and fix. Here is the situation:

 The target is Red Hat Enterprise 5 x86_64 running current SVN
iscsi-target (although I believe I've seen this with 0.4.15); the
exported disks are LVM volumes on top of an 11-disk software RAID-6.

 The initiator is 64-bit Solaris 10 x86 (currently S10U4, but we have
seen this on S10U3 as well), running Solaris Volume Manager (with a
ufs filesystem on top of it).

 If I configure the exported LUNs as 'Type=fileio', everything works OK
but the (streaming IO) performance is not very good to terrible.

 If I configure the exported LUNs as 'Type=blockio', the Solaris machine
does something that eventually wedges iSCSI processing on the target.
The system doesn't crash but it stops answering iSCSI stuff, and things
like '/etc/init.d/iscsi-target stop' hang.

 The Solaris machine is fine with blockio LUNs for ZFS and UFS
filesystems directly on the exported LUNs; only when Solaris Volume
Manager gets into the picture do things explode.

 I have SysRq-T output from the target system if there is something I
should look for (or I can send the whole thing to the mailing list or
specific people). I would be happy to run debugging code, poke things,
and so on, as these are test systems.

 Thanks in advance.

	- cks