Re: port leak when starting a translator

[Richard Braun]

On Thu, May 05, 2011 at 01:54:30PM +0200, Samuel Thibault wrote:
> It's the big leak I have on the buildds. 50k ports after like 10h
> building stuff.

Assuming this leak only happens when starting a new translator, this
implies building stuff starts a lot of them. Identifying which ones
may help narrowing the leak origin. Also, does portinfo print lots of
receive only (or maybe receive/send-once) rights (I assume the leak
leaves no send right in ext2fs) ?

> > It seems to be triggered by starting a new translator only.
> 
> So you are actually observing it too?

No, but that's what the code seems to do.

> the ports array is given to fshelp_start_translator_long, which just
> passes it to the file_exec() RPC. AIUI, that won't make a copy inside
> ext2fs itself.

Check the status and the number of user references of newpi/dirport
using mach_port_get_receive_status() and mach_port_get_refs() right
before the call to mach_port_deallocate on dirport. The number of urefs
should be 1, and status->mps_nsrequest should be true. You should also
find how no-sender notifications are received, since that's how such
receive rights get deallocated: once the child translator terminates,
all its rights on the port are destroyed, which generates a no-sender
notification, on which the receive port is destroyed. I hope the child
translator doesn't copy this right to other tasks, in which case we
would have to track where and how references are passed.

-- 
Richard Braun