Encryption options for every P2 on a given P1 are written to each P2 individually inside ipsec.conf with multiple P2 entries + split conn entries
When multiple Phase 2 entries exist, the encryption options for every P2 inside a given P1 are added to every split connection entries (IKEv1 or IKEv2 with split connections active)
For example, with three P2 entries all set for AES256-GCM-128, ipsec.conf has this in each conn entry:
esp = aes256gcm128,aes256gcm128,aes256gcm128!
The same identical option is written on each connection.
If different options are used for P2s within the same tunnel, all of them are written to each P2. So each split conn entry contains all possible P2 algorithms for any P2 on the P1, not just the selected algorithm for that specific P2 entry.
#6 Updated by Cullen Trey 12 months ago
we had problems with one of our IPSecs. Sometimes it can connect, sometimes not. After a reboot of pfsense, pfsense can mostly connect.
When I read about this bug, I disabled all P2 entries except one. Now it always works...
Could it really be, that the entry with the same encryption multiple times (esp = aes256gcm128,aes256gcm128,aes256gcm128!) is making problems with the remote site? In our setup, only we establish the ipsec connection.
We be greate to see this fixed, as it can not do anything good.
thanks for your work!
#11 Updated by PJ Goodwin 7 months ago
Ran into this bug as well, though it appears to break things if you have too many phase 2 entries. After a certain number (currently appears to be around 85 or 90) phase 2 networks, things start becoming flakey. Sometimes the phase 2 entries at the bottom will work, and other times they don't.
If I modify the /var/etc/ipsec.conf file to make the esp config entries just have the single entry for the last few phase 2 sections, all the networks work / are stable.
If you need any additional info on our setup, let me know.
If someone can point me in the direction of what file creates the ipsec.conf file, I can take a look at it and see if I can figure out a fix. I'm going to try to get time to search for the file myself, but not sure when I'll be able to take a look.
#12 Updated by PJ Goodwin 6 months ago
Looked into this and the attached patch appears to fix the issue in 2.4.2. The comparable change also corrected a 2.3.1 version.
It simply resets the $ealgoESPsp2arr variable for each phase2 entry.
I didn't look into it but it's possible that $ealgoAHsp2arr may need to be reset at the appropriate point as well. I don't have a setup that uses that so I didn't look into it.