Repeated pipe refeed should not end route refresh as the prune routine
may start pruning otherwise valid routes.
The same applies for BGP repeated route refresh.
When changing default table behavior, I missed that it enabled to
configure multiple master4 and master6 tables. Now BIRD recognizes it
and fails properly.
The import table feed wasn't resetting the table-specific route values
like REF_FILTERED and thus made the route look like filtered even though
it should have been re-evaluated as accepted.
If no channel is flushing, table prune doesn't walk over routes in nets
and also doesn't walk over importing channel lists. This helps to
alleviate the memory caching burdens a lot.
There were some confusion about validity and usage of pflags, which
caused incorrect usage after some flags from (now removed) protocol-
specific area were moved to pflags.
We state that pflags:
- Are secondary data used by protocol-specific hooks
- Can be changed on an existing route (in contrast to copy-on-write
for primary data)
- Are irrelevant for propagation (not propagated when changed)
- Are specific to a routing table (not propagated by pipe)
The patch did these fixes:
- Do not compare pflags in rte_same(), as they may keep cached values
like BGP_REF_STALE, causing spurious propagation.
- Initialize pflags to zero in rte_get_temp(), avoid initialization in
protocol code, fixing at least two forgotten initializations (krt
and one case in babel).
- Improve documentation about pflags
When filtered routes (enabled by 'import keep filtered' option) are
updated, they trigger announcements by rte_announce(). For regular
channels (e.g. type RA_OPTIMAL or RA_ANY) such announcement is just
ignored, but in case of RA_ACCEPTED (BGP peer with 'secondary' option)
it just reannounces the old (and still valid) best route.
The patch ensures that such no-change is ignored even for these channels.
Add BGP channel option 'next hop prefer global' that modifies BGP
recursive next hop resolution to use global next hop IPv6 address instead
of link-local next hop IPv6 address for immediate next hop of received
routes.
In some specific configurations, it was possible to send BIRD into an
infinite loop of recursive next hop resolution. This was caused by route
priority inversion.
To prevent priority inversions affecting other next hops, we simply
refuse to resolve any next hop if the best route for the matching prefix
is recursive or any other route with the same preference is recursive.
Next hop resolution doesn't change route priority, therefore it is
perfectly OK to resolve BGP next hops e.g. by an OSPF route, yet if the
same (or covering) prefix is also announced by iBGP, by retraction of
the OSPF route we would get a possible priority inversion.
By this, the requesting channels do the timers in their own loops,
avoiding unnecessary synchronization when the central timer went off.
This is of course less effective for now, yet it allows to easily
implement selective reloads in future.
Instead of synchronous notifications, we use the asynchronous export
framework to notify flowspec src route updates. This allows us to
invoke flowspec revalidation without locking collisions.
Instead of synchronous notifications, we use the asynchronous export
framework to notify also hostcache updates. This allows us to do the
hostcache update and the subsequent next hop update notification without
locking collisions.
We can't free the network structures before the export has been cleaned
up, therefore it makes more sense to request prune only after export
cleanup. This change also reduces prune calls on table shutdown.
These routines detect the export congestion (as defined by configurable
thresholds) and propagate the state to readers. There are no readers for
now, they will be added in following commits.
There were more conflicts that I'd like to see, most notably in route
export. If a bisect identifies this commit with something related, it
may be simply true that this commit introduces that bug. Let's hope it
doesn't happen.
The invalid routes were filtered out before they could ever get
exported, yet some of the routines need them available, e.g. for
display or import reload.
Now the invalid routes are properly exported and dropped in channel
export routines instead.
For BGP LLGR purposes, there was an API allowing a protocol to directly
modify their stale routes in table before flushing them. This API was
called by the table prune routine which violates the future locking
requirements.
Instead of this, BGP now requests a special route export and reimports
these routes into the table, allowing for asynchronous execution without
locking the table on export.
Until now, we were marking routes as REF_STALE and REF_DISCARD to
cleanup old routes after route refresh. This needed a synchronous route
table walk at both beginning and the end of route refresh routine,
marking the routes by the flags.
We avoid these walks by using a stale counter. Every route contains:
u8 stale_cycle;
Every import hook contains:
u8 stale_set;
u8 stale_valid;
u8 stale_pruned;
u8 stale_pruning;
In base_state, stale_set == stale_valid == stale_pruned == stale_pruning
and all routes' stale_cycle also have the same value.
The route refresh looks like follows:
+ ----------- + --------- + ----------- + ------------- + ------------ +
| | stale_set | stale_valid | stale_pruning | stale_pruned |
| Base | x | x | x | x |
| Begin | x+1 | x | x | x |
... now routes are being inserted with stale_cycle == (x+1)
| End | x+1 | x+1 | x | x |
... now table pruning routine is scheduled
| Prune begin | x+1 | x+1 | x+1 | x |
... now routes with stale_cycle not between stale_set and stale_valid
are deleted
| Prune end | x+1 | x+1 | x+1 | x+1 |
+ ----------- + --------- + ----------- + ------------- + ------------ +
The pruning routine is asynchronous and may have high latency in
high-load environments. Therefore, multiple route refresh requests may
happen before the pruning routine starts, leading to this situation:
| Prune begin | x+k | x+k | x -> x+k | x |
... or even
| Prune begin | x+k+1 | x+k | x -> x+k | x |
... if the prune event starts while another route refresh is running.
In such a case, the pruning routine still deletes routes not fitting
between stale_set and and stale_valid, effectively pruning the remnants
of all unpruned route refreshes from before:
| Prune end | x+k | x+k | x+k | x+k |
In extremely rare cases, there may happen too many route refreshes
before any route prune routine finishes. If the difference between
stale_valid and stale_pruned becomes more than 128 when requesting for
another route refresh, the routine walks the table synchronously and
resets all the stale values to a base state, while logging a warning.
Until now, if export table was enabled, Nest was storing exactly the
route before rt_notify() was called on it. This was quite sloppy and
spooky and it also wasn't reflecting the changes BGP does before
sending. And as BGP is storing the routes to be sent anyway, we are
simply keeping the already-sent routes in there to better rule out
unneeded reexports.
Some of the route attributes (IGP metric, preference) make no sense in
BGP, therefore these will be probably replaced by something sensible.
Also the nexthop shown in the short output is the BGP nexthop.
Passing protocol to preexport was in fact a historical relic from the
old times when channels weren't a thing. Refactoring that to match
current extensibility needs.
In the multithreaded environment, it is not supposed that anybody
traverses the routing table as the CLI show-route was doing. Now the
routing table traversal is gone and CLI won't hold the table locked
while computing filters.
Added an option for export filter to allow for prefiltering based on the
prefix. Routes outside the given prefix are completely ignored. Config
is simple:
export in <net> <filter>;
There were quite a lot of conflicts in flowspec validation code which
ultimately led to some code being a bit rewritten, not only adapted from
this or that branch, yet it is still in a limit of a merge.
Validation is called internally from route table at the same place where
nexthop resolution is done. Also accounting for rte->sender semantics
change (not a channel but the import hook instead).
Use timer (configurable as 'gc period') to schedule routing table
GC/pruning to ensure that prune is done on time but not too often.
Randomize GC timers to avoid concentration of GC events from different
tables in one loop cycle.
Fix a bug that caused minimum inter-GC interval be 5 us instead of 5 s.
Make default 'gc period' adaptive based on number of routing tables,
from 10 s for small setups to 600 s for large ones.
In marge multi-table RS setup, the patch improved time of flushing
a downed peer from 20-30 min to <2 min and removed 40s latencies.
For now, all route attributes are stored as eattrs in ea_list. This
should make route manipulation easier and it also allows for a layered
approach of route attributes where updates from filters will be stored
as an overlay over the previous version.
As there is either a nexthop or another destination specification
(or othing in case of ROAs and Flowspec), it may be merged together.
This code is somehow quirky and should be replaced in future by better
implementation of nexthop.
Also flowspec validation result has its own attribute now as it doesn't
have anything to do with route nexthop.
This doesn't do anything more than to put the whole structure inside
adata. The overall performance is certainly going downhill; we'll
optimize this later.
Anyway, this is one of the latest items inside rta and in several
commits we may drop rta completely and move to eattrs-only routes.
The prefix hash table in BGP used the same hash function as the rtable.
When a batch of routes are exported during feed/flush to the BGP, they
all have similar hash values, so they are all crowded in a few slots in
the BGP prefix table (which is much smaller - around the size of the
batch - and uses higher bits from hash values), making it much slower due
to excessive collisions. Use a different hash function to avoid this.
Also, increase the batch size to fill 4k BGP packets and increase minimum
BGP bucket and prefix hash sizes to avoid back and forth resizing during
flushes.
This leads to order of magnitude faster flushes (on my test data).
Changes in internal API:
* Every route attribute must be defined as struct ea_class somewhere.
* Registration of route attributes known at startup must be done by
ea_register_init() from protocol build functions.
* Every attribute has now its symbol registered in a global symbol table
defined as SYM_ATTRIBUTE
* All attribute ID's are dynamically allocated.
* Attribute value custom formatting hook is defined in the ea_class.
* Attribute names are the same for display and filters, always prefixed
by protocol name.
Also added some unit testing code for filters with route attributes.
The prune loop may may rebuild the prefix trie and therefore invalidate
walk state for asynchronous walks (used in 'show route in' cmd). Fix it
by adding locking that keeps the old trie in memory until current walks
are done.
In future this could be improved by rebuilding trie walk states (by
lookup for last found prefix) after the prefix trie rebuild.
When rtable is pruned and network fib nodes are removed, we also need to
prune prefix trie. Unfortunately, rebuilding prefix trie takes long time
(got about 400 ms for 1M networks), so must not be atomic, we have to
rebuild a new trie while current one is still active. That may require
some considerable amount of temporary memory, so we do that only if
we expect significant trie size reduction.
Implement flowspec validation procedure as described in RFC 8955 sec. 6
and RFC 9117. The Validation procedure enforces that only routers in the
forwarding path for a network can originate flowspec rules for that
network.
The patch adds new mechanism for tracking inter-table dependencies, which
is necessary as the flowspec validation depends on IP routes, and flowspec
rules must be revalidated when best IP routes change.
The validation procedure is disabled by default and requires that
relevant IP table uses trie, as it uses interval queries for subnets.
Attach a prefix trie to IP/VPN/ROA tables. Use it for net_route() and
net_roa_check(). This leads to 3-5x speedups for IPv4 and 5-10x
speedup for IPv6 of these calls.
TODO:
- Rebuild the trie during rt_prune_table()
- Better way to avoid trie_add_prefix() in net_get() for existing tables
- Make it configurable (?)
This commit prevents use-after-free of routes belonging to protocols
which have been already destroyed, delaying also all the protocols'
shutdown until all of their routes have been finally propagated through
all the pipes down to the appropriate exports.
The use-after-free was somehow hypothetic yet theoretically possible in
rare conditions, when one BGP protocol authors a lot of routes and the
user deletes that protocol by reconfiguring in the same time as next hop
update is requested, causing rte_better() to be called on a
not-yet-pruned network prefix while the owner protocol has been already
freed.
In parallel execution environments, this would happen an inter-thread
use-after-free, causing possible heisenbugs or other nasty problems.
This basically means that:
* there are some more levels of indirection and asynchronicity, mostly
in cleanup procedures, requiring correct lock ordering
* all the internal table operations (prune, next hop update) are done
without blocking the other parts of BIRD
* the protocols may get their own loops very soon
There is a simple universal IO loop, taking care of events, timers and
sockets. Primarily, one instance of a protocol should use exactly one IO
loop to do all its work, as is now done in BFD.
Contrary to previous versions, the loop is now launched and cleaned by
the nest/proto.c code, allowing for a protocol to just request its own
loop by setting the loop's lock order in config higher than the_bird.
It is not supported nor checked if any protocol changed the requested
lock order in reconfigure. No protocol should do it at all.
In some specific configurations, it was possible to send BIRD into an
infinite loop of recursive next hop resolution. This was caused by route
priority inversion.
To prevent priority inversions affecting other next hops, we simply
refuse to resolve any next hop if the best route for the matching prefix
is recursive or any other route with the same preference is recursive.
Next hop resolution doesn't change route priority, therefore it is
perfectly OK to resolve BGP next hops e.g. by an OSPF route, yet if the
same (or covering) prefix is also announced by iBGP, by retraction of
the OSPF route we would get a possible priority inversion.
* internal tables are now more standalone, having their own import and
export hooks
* route refresh/reload uses stale counter instead of stale flag,
allowing to drop walking the table at the beginning
* route modify (by BGP LLGR) is now done by a special refeed hook,
reimporting the modified routes directly without filters
Channels have now included rt_import_req and rt_export_req to hook into
the table instead of just one list node. This will (in future) allow for:
* channel import and export bound to different tables
* more efficient pipe code (dropping most of the channel code)
* conversion of 'show route' to a special kind of export
* temporary static routes from CLI
The import / export states are also updated to the new algorithms.
Routes are now allocated only when they are just to be inserted to the
table. Updating a route needs a locally allocated route structure.
Ownership of the attributes is also now not transfered from protocols to
tables and vice versa but just borrowed which should be easier to handle
in a multithreaded environment.
