On Fri, Feb 21, 2014 at 4:42 PM, Antti Kantee <[email protected]> wrote:
> On 21/02/14 12:13, Justin Cormack wrote:
>>> The options I can see are:
>>> 1) make signals completely mimic POSIX (and fix bugs forever)
>>> 2) leave things as they currently are (rump kernels make a reasonable
>>> effort to cater to signal-happy programs, though semantics are so-so)
>>> 3) completely rip out any host signal delivery (equally good/bad for all
>>> hosts, and does not give a false sense of "1")
>>>
>> Not sure, 2 or 3 seem valid. Is there a description of exactly what works
>> under 2? I haven't actually knowingly used signals under rump.
>
> First of all, there are several signal "models" that you can select for
> a rump kernel by calling rump_boot_setsigmodel(): panic, ignore, record
> and raise.  The first three are hopefully obvious (if not, see
> src/sys/rump/librump/rumpkern/signals.c).  I'm talking about the last
> one (raise), which is also the default model.
>
> A signal can be delivered either to a local client or a remote client.
> The signal is raised in the host container (if available).  On POSIX
> hosts it's simply raise(), and called either via rumpuser_kill() or the
> RUMPSP_RAISE PDU for local or remote clients, respectively.  This is
> what works and happen by default.  It's not very many lines of code
> (~10'ish).
>
> As I mentioned in the previous mail, the other part is tracking signal
> masks per rump kernel process (well, thread) context and interrupting
> syscalls when a signal is delivered (iff the syscall is interruptable).
>   This is what doesn't work and would be quite a few lines of code.
>
> I think I originally implemented the "raise" signal model so that
> SIGXFSZ from a file system could be tested.  Even with several years of
> hindsight, I'm not sure if implementing it was a mistake, pretty much
> confirming that deciding what "the right thing to do" is not easy here.

That seems like a fairly weak justification. I am becoming more
inclined to go for option 3 now... If there are no syscalls that
generate signals (itimer etc) and its all there just for SIGXFSZ which
is not really needed as a signal, then its just much simpler to remove
it all. None of these generated signals dont also have perfect;y good
return codes if the signal is not delivered.

Justin

------------------------------------------------------------------------------
Managing the Performance of Cloud-Based Applications
Take advantage of what the Cloud has to offer - Avoid Common Pitfalls.
Read the Whitepaper.
http://pubads.g.doubleclick.net/gampad/clk?id=121054471&iu=/4140/ostg.clktrk
_______________________________________________
rumpkernel-users mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/rumpkernel-users

Reply via email to