Initial IPv6 Support #533
Open
Initial IPv6 Support #533
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Removes one of the multiple duplicate types that implement an IP Prefix. Signed-off-by: Trey Aspelund <[email protected]>
Moves to ddm/oxnet types to represent an IP Prefix instead of custom types. Step 2 in standardizing around oxnet Prefix types. Signed-off-by: Trey Aspelund <[email protected]>
Remove unused BgpAttributes4, BgpAttributes6, OriginChangeSet, and Status types. There were a couple other unused types that seemed more forward-looking, so I left those in place (i.e. fn to_buf(), Policy4Key) Signed-off-by: Trey Aspelund <[email protected]>
Updates Db to incorporate new methods that support IPv6. The RIB is now split into three separate types: - Rib: Holds Prefix (protocol-agnostic enum) - Rib4: Holds Prefix4 (IPv4 specific struct) - Rib6: Holds Prefix6 (IPv6 specific struct) Updates bestpaths() to be completely prefix-agnostic (now accepts paths as an argument in lieu of a prefix and Rib). Callers of prefix add/remove methods are now responsible for managing locks for the appropriate RIB structures, allowing each protocol's RIB to be locked independently and giving the caller the option to reduce how many times the lock needs to be acquired/released. Signed-off-by: Trey Aspelund <[email protected]>
Initial implementation of IPv6 Static Routing: - New mgd APIs - New mgadm commands (mirroring v4 static routing commands) Signed-off-by: Trey Aspelund <[email protected]>
Signed-off-by: Trey Aspelund <[email protected]>
During creation of Prefix4/Prefix6 structs, an Ipv{4,6}Addr was blindly
allowed w/o any validation or zeroing of the host bits. This allowed
for the RIB to hold multiple entries for routes w/ the same underlying
subnet (e.g. 1.1.1.1/24 and 1.1.1.2/24 would be considered unique keys,
but the underlying subnet is still 1.1.1.0/24), possibly even with
differing next-hop values.
This commit addresses that issue by adding a step to zero the host bits
during the creation of new Prefix4/Prefix6 structs. Figuring out a
migration strategy for the on-disk sled::Db remains a concern, but
this prevents new problematic entries from being created.
Fixes: #529
Signed-off-by: Trey Aspelund <[email protected]>
Second half of the fix for #529. First we moved to a constructor that enforced all host bits be zeroed when creating a Prefix* type. Now, we check whether the routes in the on-disk db are valid (host bits unset) and remove them if they're invalid. This also moves most of the logic into methods of the Prefix* types, with the exception of the mg-admin-client mirrors of them, as progenitor can't know about methods/traits from the openapi spec so that logic has to live somewhere. Fixes: #529 Signed-off-by: Trey Aspelund <[email protected]>
Signed-off-by: Trey Aspelund <[email protected]>
Signed-off-by: Trey Aspelund <[email protected]>
Some helpful parsing and retry macros were sitting in bgp and other crates couldn't use them... so I moved them to mg-common. Signed-off-by: Trey Aspelund <[email protected]>
Adds APIs, db methods, mgadm commands and unit tests for IPv6 BGP Origin Signed-off-by: Trey Aspelund <[email protected]>
Signed-off-by: Trey Aspelund <[email protected]>
Signed-off-by: Trey Aspelund <[email protected]>
Signed-off-by: Trey Aspelund <[email protected]>
Moves RIB queries from /bgp to /rib (this will likely require a corresponding update in Nexus), as rib_in and rib_loc are not specific to BGP. Moves RIB logic into its own set of files (mgd/src/rib_admin.rs, mgadm/src/rib.rs). Refactors print_rib() and move it into rib.rs. Adds support for server-side filtering of RIB queries by AddressFamily and Protocol (BGP/Static). Removes a bunch of duplicate types and impls by using progenitor's "replace" directive in generate_api!() macro to automatically map progenitor's auto-derived types back to their original rdb types (this is something we should do further refactoring on, as there is quite a lot of type/impl duplication going on). Fixes: #525 Signed-off-by: Trey Aspelund <[email protected]>
Signed-off-by: Trey Aspelund <[email protected]>
| @@ -83,23 +42,28 @@ fn test_basic_peering() { | |||
| // Ensure that r1's peer session to r2 has gone back to connect. | |||
| r2.shutdown(); | |||
| d2.shutdown(); | |||
| wait_for_eq!(r1_session.state(), FsmStateKind::Connect); | |||
There was a problem hiding this comment.
Why are we expecting state machine behavior changes? It looks like tests are now failing here.
There was a problem hiding this comment.
The main reason I modified this is because I believed the FSM state observed is timing-dependent, not consistent.
It seemed to me like the goal of this test case is to confirm that "shutdown" on one side brings the session down.
If so, then looking for "!established" is a less timing-dependent way to confirm that things have been completely torn down.
Otherwise, the state on the active side could be "connect", "opensent" (since the listener isn't torn down on the shutdown side, the transport can still come up) or "idle".
I was hoping to avoid a scenario where the test fails because the active peer is in "idle" or "opensent" and we were only looking for "connect".
I don't have super strong opinions about this though, so if reverting that change means the test passes consistently again, then I'm fine with that.
There was a problem hiding this comment.
I'm not sure at what point this test became unreliable. I'll go back through and figure out which of my changes borked things. I don't suspect the issue is with the logic change of "state != established", but I could be wrong
| }; | ||
|
|
||
| for p in prefixes { | ||
| update.nlri.push(p.clone()); |
There was a problem hiding this comment.
Shouldn't v6 be in mp_reach_nlri/mp_unreach_nlri?
There was a problem hiding this comment.
Eventually, yes. MP_{REACH,UNREACH}_NLRI hasn't been implemented yet
Signed-off-by: Trey Aspelund <[email protected]>
Signed-off-by: Trey Aspelund <[email protected]>
Signed-off-by: Trey Aspelund <[email protected]>
The idle hold timer was not being handled properly before. This led to situations where a session that was configured as passive to transition from the Idle FSM state into Connect, when passive sessions should only ever go into Active. This adds explicit handling of the timer in the FSM when transitions to Idle state are occurring or when the IdleHoldTimerExpires FSM Event is received. Signed-off-by: Trey Aspelund <[email protected]>
The Dispatcher loop was calling Listener::bind() after any event,
regardless of whether that was a normal/good event, e.g. accept()'ing a
connection, or a bad/error event, e.g. an accept() or send() failure.
This moves the accept() and send() logic into an inner loop and sets up
flow control such that a normal event will not trigger a re-binding of
the listener socket. The output below are from two successful test runs
with additional debugs added, one with the flow control changes and one
without.
Without loop updates (169 bindings occurs):
```
treyaspelund@Tallon-IV 01:25:46 PM | ~/git/maghemite trey/ipv6 ○
‣ grep 'binding' bgp/r1.basic_peering.log | wc
169 676 26343
```
With loop updates (only 1 bind occurs):
```
treyaspelund@Tallon-IV 01:24:24 PM | ~/git/maghemite trey/ipv6 ✓
‣ grep 'binding' bgp/r1.basic_peering.log
{"msg":"bgp dispatcher binding 1.0.0.1:179","v":0,"name":"slog-rs","level":20,"time":"2025-08-22T13:23:51.058233-04:00","hostname":"Tallon-IV","pid":22176}
```
Signed-off-by: Trey Aspelund <[email protected]>
Signed-off-by: Trey Aspelund <[email protected]>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This PR adds initial support for IPv6 static routing in maghemite, with the very beginnings of BGP support (really just originated IPv6 prefixes).
This addresses a few bugs found along the way (e.g. #529), adds unit tests for old/new functionality, and does a good bit of code cleanup, e.g.