Porting the block-iscsi hotplug script

[Gustau Pérez]

El 22/03/16 a les 12:08, Roger Pau Monné ha escrit:
> Hello, I've been on vacations last week and I'm trying to catch up
> with my email backlog, sorry for the delay in the reply. OTOH, George
> Dunlap from Citrix has posted the series that should enable hotplug
> scripts to work with HVM domains. I need to rebase my FreeBSD hotplug
> script support on top of it, so you can try it. George's series can be
> found at:
> http://lists.xenproject.org/archives/html/xen-devel/2016-03/msg02739.html
> I will try to rebase my work on top of this and provide you with a git
> branch so you can easily test it. Regarding your questions about the
> locking mechanism, I'm not that familiar with it, but I'm not sure why
> a lock array is need. AFAICT only one lock should be needed, that
> prevents multiple hotplug scripts from travessing the xenstore disk
> information at the same time. Please bring this question to xen-devel. 

   Hi Roger,

   great someone is working on HVM hotplug script support (even though I
would have loved to deal with that by myself). So to recap, the plan
would something like:

  * wait for you to rebase the George's hotplug script support for HVM
    and create an specific branch for me
  * test the hotplug scripts over that work to check if the scripts work
    for both HVM and PV domains
  * in the meantime, if the script looks good, perhaps it could be added
    to the xen-tools port while it's been accepted upstream. I'd do git
    patches, anything special I'd need to know when doing the patches
    with git (formating, etc?) Do the message need to state any kind of
    license?

  About the locking mechanism, I'll ask at xen-devel. The Linux script
says the locking mechanism works not among processes but among fds,
which seemed bit strange to me. Instead I decided to lock between
processes, implementing reentrant mutexes (the release function checks
if the lock is being help multiple times, if released lock is being held
one time, the file is removed).

   I'll ask next week in xen-devel. Anything else I can do?

   Best,

   Gustau

>> [1] http://pastebin.com/FhaQeuG1
>> [2] http://pastebin.com/ZHWKRg5f
> LGTM :). You will have to formally submit this as patches when the hotplug 
> work is committed to upstream Xen.
>
> Roger.