[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [PATCH] eth-multiplxer: Implement ds_device_close()

From: Justus Winter
Subject: Re: [PATCH] eth-multiplxer: Implement ds_device_close()
Date: Thu, 21 Sep 2017 12:07:45 +0200

Hi :)

"Joan Lledó" <joanlluislledo@gmail.com> writes:

> Hello,
> One difference between pfinet and lwip translators is that pfinet doesn't 
> delete interfaces (or I haven't found how to do it) while lwip does. For 
> instance, if one starts pfinet with:
> settrans -fga /servers/socket/2 /hurd/pfinet -i /dev/eth1 -a 
> -m -g
> and then runs:
> fsysopts /servers/socket/2 --interface=/dev/eth2 --address= 
> --netmask= --gateway=
> the result is this:
> root@hurd:/home/jlledom# fsysopts /servers/socket/2
> /hurd/pfinet --interface=/dev/eth1 --address= 
> --netmask= --address6=fc00:123::5054:ff:feb6:1ab3/64 
> --address6=fe80::5254:b6:1ab3/10 --address6=fe80::5054:ff:feb6:1ab3/10 
> --gateway6=fe80::5054:ff:fef6:d496 --interface=/dev/eth2 
> --address= --netmask= --gateway= 
> --address6=fe80::5254:4b:ad11/10 --address6=fe80::5054:ff:fe4b:ad11/10
> fsysopts added a new interface. But doing the same with lwip leads to this:
> root@hurd:/home/jlledom# fsysopts /servers/socket/2
> /hurd/lwip --interface=/dev/eth2 --address= 
> --netmask= --gateway= 
> --address6=FE80::5054:FF:FE4B:AD11/64 
> --address6=FC00:124::5054:FF:FE4B:AD11/64
> The stack is reset and reconfigured, and only the given interface is added. I 
> implemented this by simply removing all interfaces anytime somebody calls 
> fsysopts, and adding the new ones. If some interface already exists, it's 
> deleted and created again.

Afaik using multiple interfaces is not something that many people have
done, so either behavior looks okay to me.

> This behaviour generated a problem when working with
> eth-multiplexer. Adding a new interface calls ds_device_open in
> eth-multiplexer and that returns a port name, whereas deleting the
> interface deallocates the port name in lwip, but nothing happens in
> eth-multiplexer since ds_device_close is unimplemented,

ds_device_close is not unimplemented, it merely does nothing.  But
indeed, nothing happens when the only sender deallocates its port
because of the way eth-multiplexer handles the device objects (it puts
them into a linked list with a hard reference, preventing the proper
deallocation of the port).  I'll try to fix that.

> when lwip
> tries to create the same interface again, eth-multiplexer returns the
> same port name,

That should be fine though.

> and trying to use it leads to a MIG_SERVER_DIED
> error.

Please confirm that eth-multiplexer does not die.

MIG_SERVER_DIED is generated by the mig-generated client stubs when the
replies message id does not matche the expected id, but
MACH_NOTIFY_SEND_ONCE.  This message is generated by the kernel if a
send-once right is destroyed.  This usually happens when the server
dies, but could in theory happen when the server explicitly destroys the
reply port, but tbh I couldn't see why or where that should happen.


reply via email to

[Prev in Thread] Current Thread [Next in Thread]