Hey Daniel,<div><br></div><div>I am all for option 3, but also quite happy to deal with SIP-based dialog matching only. If we are going to use some form of alternative matching, then agreed, probably better to use the new uid as opposed to an id based on storage structures and their relative positions.</div>
<div><br></div><div>Cheers</div><div>Jason<br><br><div class="gmail_quote">On Wed, Apr 11, 2012 at 2:22 PM, Daniel-Constantin Mierla <span dir="ltr"><<a href="mailto:miconda@gmail.com">miconda@gmail.com</a>></span> wrote:<br>
<blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">Hello,<br>
<br>
the internal dialog id - the pair h_entry,h_id - seems exposed to some issues. They:<br>
- h_entry is the slot index (in the internal hash table) where the dialog structure will be stored<br>
- h_id is an incremented counter for each slot<br>
<br>
h_entry is computed as a hash function over Call-Id value, then made modulo nr. of slots.<br>
<br>
The problems can appear upon restart when changing the hash_size module parameter:<br>
<br>
<a href="http://kamailio.org/docs/modules/stable/modules_k/dialog.html#id2534423" target="_blank">http://kamailio.org/docs/<u></u>modules/stable/modules_k/<u></u>dialog.html#id2534423</a><br>
<br>
If it is decremented, then some dialogs might be left outside since their h_entry might not be valid anymore. Also, when hash_size is changed, for matching on SIP dialog attributes of ongoing calls, the hash function over callid made modulo over the new nr. of slots may get a different hash table slot index, resulting in dialog structure not being found.<br>
<br>
A very unlikely issue is caused by the h_id, which being incremented can overflow and start from 0 again. But will overlap if an dialog is still going on after 2^32 new dialogs on that hash table slot were initiated.<br>
<br>
I wanted to brought all in the discussion to make a decision. I see few options for the moment:<br>
1) document this issue, so a restart with change of hash_size parameter must have 0 active dialogs<br>
2) replace h_entry with the value of the hash function over call id, doing modulo operation each time to find the slot. The unlikely issue with overflow of h_id is still there<br>
3) switch over another kind of unique id (sriud.{c,h} added recently in lib/srutils/ can be an option). This will imply changing the database structure<br>
<br>
Besides these, might be better to drop using the Record-Route header parameter to carry dialog id and just do SIP based dialog matching. Completely trusting those tokens in RR parameter might add some security risks and if there is a fallback to SIP-based dialog matching, using the shortcut with RR does not impact anymore that much.<br>
<br>
Opinions and other suggestions?<br>
<br>
Cheers,<br>
Daniel<span class="HOEnZb"><font color="#888888"><br>
<br>
-- <br>
Daniel-Constantin Mierla<br>
Kamailio Advanced Training, April 23-26, 2012, Berlin, Germany<br>
<a href="http://www.asipto.com/index.php/kamailio-advanced-training/" target="_blank">http://www.asipto.com/index.<u></u>php/kamailio-advanced-<u></u>training/</a><br>
<br>
<br>
______________________________<u></u>_________________<br>
sr-dev mailing list<br>
<a href="mailto:sr-dev@lists.sip-router.org" target="_blank">sr-dev@lists.sip-router.org</a><br>
<a href="http://lists.sip-router.org/cgi-bin/mailman/listinfo/sr-dev" target="_blank">http://lists.sip-router.org/<u></u>cgi-bin/mailman/listinfo/sr-<u></u>dev</a><br>
</font></span></blockquote></div><br></div>