New version. Changelog: * Split nest API and bgp implementation * Simplify new nest code a bit * Fix crash on CLI lookups (forgot to fill in sym->def) * Remove protocols on dynamic templates renaming * Permit deleting dynamic instances via removal of disabled protos on reconfigure * Do not create new dynamic protocols on shutdown * Permit using table names (%T) in templates * Allow to inherit allow list from common templates (handy in link-local setups) 12.02.2015, 03:48, "Alexander V. Chernikov" <melifaro@ipfw.ru>:
I'm sorry, previous posting was in html, duplicating as plain-text
I'd like to share some (beta-quality) patches permitting to create BGP protocol instances dynamically, based on templates. Main purpose is to get feedback and make them good enough to be integrated into main tree.
Similar feature was implemented by major vendors yeas ago. Juniper: http://www.juniper.net/techpubs/en_US/junos11.4/topics/usage-guidelines/rout... Cisco IOS: http://www.cisco.com/c/en/us/td/docs/ios/12_2sr/12_2srb/feature/guide/tbgp_c...
Typically it is used in peering setups.
Patches still have some rough edges (see below) but should work^Wnot make bird crash every 5 minutes.
Example setup:
# Set up template with ALL info except and IP address # In this case we don't have any option to specify interface, so it is specified in # neighbor part template bgp bgg { debug all; local as 4444; neighbor fe80::1%em0 as 4443; import all; export all; };
# Set up dynamic template with ranges and neighbor mask # Accepted format (%) values: # 1..4 - octets in IPv4 address # 1..8 - hex groups in IPv6 address # I - full address # format values can be specified more than once in format string # e.g. "b_x_%4_%I_%4" dynamic template bgp bgp_int from bgg { dynamic template name "b_s%4-%8-%8"; dynamic allow fe80::f000/112, fe80::f00/116; };
# You need to have at least one "normal" bgp peer to get listen socket opened # That's all
# Example from working system: # using several { dynamic template name "b_Vrf1_s%8"; dynamic allow fe80::/64; } # templates
bird> show protocols name proto table state since info direct1 Direct master up 22:06:05 kernel1 Kernel master up 22:06:05 device1 Device master up 22:06:05 ospf3 OSPF master up 22:06:05 Running .. b_Vrf2_dc1_c1 BGP Vrf2 up 22:06:06 Established b_Vrf1_se48 BGP Vrf1 up 22:06:05 Established b_Vrf1_se10 BGP Vrf1 up 22:06:06 Established b_Vrf1_se45 BGP Vrf1 up 22:06:06 Established b_Vrf1_se60 BGP Vrf1 up 22:06:07 Established b_Vrf2_se91 BGP Vrf2 up 22:06:08 Established ..
Comments on individual patches: 0001 is basic LPM for the f_trie and is part of aggregator patch 0002 decouples get-proto-from-address function from incoming connection handling. This makes hacking here much simpler (for example, implementing IP hash lookup) 0003 just moves SYM_MAX_LEN definition to header 0004 moves sizeof(XXX_proto) info inside protocol structure permitting anyone to call proto_config_init() without knowing details 0005 tries to isolate/limit global new_config/cfg_mem variables usage to absolute minimum. Despite the fact that actual BGP patch does not require much more than config_start_edit() / config_stop_edit() implementation, I (after hours of debugging cores) decided to make things more transparent here. 0006 is the actual patch. Basically, it * adds SYM_TEMPLATE class * adds dyn_parent proto_config field to specify parent template * adds condition-check and copy-dynamic protocol callbacks * adds runtime config/protocol creation functions to nest * adds bgp-specific structure to struct config allocated on demand
Rough edges: * non-optimized case for IPv6 LL prefixes * non-optimized peer lookup for existing sessions * dynamic expire not implemented (next version) * no docs (next version) * some bugs, definitely