Skip to contentSkip to conversation

problem with libipq and userprocess and 'clean' exit

DISCUSSION Tassilo Schütz 3 participants 5 messages Jun 13, 2002 – Jun 14, 2002 netfilter

lore
Share & tools
lore
#0 lore raw

Hello list,

I have a userprocess using libipq. It stores packets comming in through traget -j QUEUE, delay them for a while, and send them out via verdict NF_ACCEPT. That works fine. But when I quit the process normaly (use ipq_destroy_handle()) and not all packets have been send out again, I see sometimes in /proc/net/ip_queue, that the values are not set back. It is the same, when the userprocess starts, calls irq_read() and before a packet arrives, I quit the process normaly. In /proc/net/ip_queue you can see that Peer pid, Peer copy mode and Peer copy range are not set back.

I would like to know, if I have to do anything else, before quit my userprocess? Is there maybe a flush_queue() call or term_queue() call? I thought about, because of Queue flushing and Queue terminate in /proc/net/ip_queue.

Looking through the source of ip_queue.c I found some similar calls and a comment :
* Packets arrive here from netfilter for queuing to userspace. *
* All of them must be fed back via nf_reinject() or Alexey will kill Rusty. *

Can someone give me a guidance how to quit my process without the above effets :-) ?

Thanks in advance

Tassilo

#1 lore raw
Quoted 8 lines · On Thu, 13 Jun 2002, [ISO-8859-15] Tassilo Schütz wrote:

Hello list,

I would like to know, if I have to do anything else, before quit my userprocess? Is there
maybe a flush_queue() call or term_queue() call? I thought about, because of Queue flushing
and Queue terminate in /proc/net/ip_queue.

Nothing to worry about, there is currentl no way for the ip_queue module to know that the user process has exited (unless another process starts using the queue).

The kernel module also ensures that all packets are fed back into the stack, no matter what the userspace application does.

- James

Signature · 2 lines
James Morris
<[email protected]>
#2 lore raw
Quoted 6 lines · James Morris

Nothing to worry about, there is currentl no way for the ip_queue module to know that the user process has exited (unless another process starts using the queue).

The kernel module also ensures that all packets are fed back into the stack, no matter what the userspace application does.

So one way to get out of the situation is to remove the QUEUE target to stop new packets from being queued, and then start a dummy program that attaches to the queue? This should both stop new packets from being queued and release any pending packets, right?

Regards
Henrik

#3 lore raw
Quoted 9 lines · Henrik Nordstrom
Quoted 6 lines · James Morris

Nothing to worry about, there is currentl no way for the ip_queue module to know that the user process has exited (unless another process starts using the queue).

The kernel module also ensures that all packets are fed back into the stack, no matter what the userspace application does.

Actually, the module does detect that the process has exited if it tries to send a packet to userspace. At this point, it will flush the queue and not queue any more packets unless another process sends a queue mode message.

If the user process exits and no new packets arrive, any queued packets will stay queued until the module is removed or any associated device is downed. This is what is meant by the kernel module ensuring that all queued packets make it back to the stack.

Quoted 5 lines

So one way to get out of the situation is to remove the QUEUE target to stop new packets from being queued, and then start a dummy program that attaches to the queue? This should both stop new packets from being queued and release any pending packets, right?

Yes, although, it would be better to get a notifcation from Netlink when the process closes the socket -- the kernel code and userspace semantics could be then made much cleaner.

- James

Signature · 2 lines
James Morris
<[email protected]>
#4 lore raw
In reply to James Morris
Quoted 8 lines · James Morris

Nothing to worry about, there is currentl no way for the ip_queue module to know that the user process has exited (unless another process starts using the queue).

The kernel module also ensures that all packets are fed back into the stack, no matter what the userspace application does.

- James

Thank you for the information.

best regards

Tassilo