Maria Matejka [Sun, 2 Apr 2023 17:15:22 +0000 (19:15 +0200)]
Sockets: Unified API for main and other loops
Now sk_open() requires an explicit IO loop to open the socket in. Also
specific functions for socket RX pause / resume are added to allow for
BGP corking.
And last but not least, socket reloop is now synchronous to resolve
weird cases of the target loop stopping before actually picking up the
relooped socket. Now the caller must ensure that both loops are locked
while relooping, and this way all sockets always have their respective
loop.
Maria Matejka [Fri, 24 Feb 2023 08:13:35 +0000 (09:13 +0100)]
More efficient IO loop event execution to avoid long loops
If there are lots of loops in a single thread and only some of the loops
are actually active, the other loops are now kept aside and not checked
until they actually get some timers, events or active sockets.
This should help with extreme loads like 100k tables and protocols.
Also ping and loop pickup mechanism was allowing subtle race
conditions. Now properly handling collisions between loop ping and pickup.
Maria Matejka [Thu, 9 Mar 2023 15:34:17 +0000 (16:34 +0100)]
Fixed default table configuration
When changing default table behavior, I missed that it enabled to
configure multiple master4 and master6 tables. Now BIRD recognizes it
and fails properly.
Maria Matejka [Wed, 8 Mar 2023 20:38:18 +0000 (21:38 +0100)]
Fixed bad filter re-evaluation with import table if filtered->accepted
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.
Maria Matejka [Wed, 29 Mar 2023 16:55:46 +0000 (18:55 +0200)]
BGP: LLGR Staleness optimization dropped.
This brought unnecessary complexity into the decision procedures while the
performance aspects weren't worth it. It just saved one ea_list traversal
when many others are also done.
Ondrej Zajicek [Sun, 19 Feb 2023 02:59:10 +0000 (03:59 +0100)]
Conf: Fix too early free of old configuration
The change 371eb49043d225d2bab8149187b813a14b4b86d2 introduced early free
of old_config. Unfortunately, it did not properly check whether it is not
still in use (blocked by obstacle during reconfiguration). Fix that.
It also means that we still could have a short peak when three configs
are in use (when a new reconfig is requeste while the previous one is
still active).
Maria Matejka [Tue, 31 Jan 2023 12:07:46 +0000 (13:07 +0100)]
Interface updates are asynchronous
Instead of propagating interface updates as they are loaded from kernel,
they are enqueued and all the notifications are called from a
protocol-specific event. This change allows to break the locking loop
between protocols and interfaces.
Anyway, this change is based on v2 branch to keep the changes between v2
and v3 smaller.
Maria Matejka [Tue, 31 Jan 2023 12:11:55 +0000 (13:11 +0100)]
Moved interface list flush to device protocol cleanup hook.
The interface list must be flushed when device protocol is stopped. This
was done in a hardcoded specific hook inside generic protocol routines.
The cleanup hook was originally used for table reference counting late
cleanup, yet it can be also simply used for prettier interface list flush.
Maria Matejka [Tue, 24 Jan 2023 10:01:34 +0000 (11:01 +0100)]
Object locks use events
Instead of calling custom hooks from object locks, we use standard event
sending mechanism to inform protocols about object lock changes. This is
a backport from version 3 where these events are passed across threads.
This implementation of object locks doesn't use mutexes to lock the
whole data structure. In version 3, this data structure may get accessed
from multiple threads and must be protected by mutex.
Maria Matejka [Tue, 24 Jan 2023 10:01:34 +0000 (11:01 +0100)]
Object locks use events
Instead of calling custom hooks from object locks, we use standard event
sending mechanism to inform protocols about object lock changes. As
event sending is lockless, the unlocking protocol simply enqueues the
appropriate event to the given loop when the locking is done.
The onlink attribute implementation collides with the nexthop attribute
behavior in v3; keeping it aside until finding out how to reimplement it
correctly.
Ondrej Zajicek [Sun, 22 Jan 2023 22:42:08 +0000 (23:42 +0100)]
BFD: Improve incoming packet matching
For active sessions, ignore received packets with zero local id and
mismatched remote id. That forces a session timeout instead of an
immediate session restart. It makes BFD sessions more resilient to
packet spoofing.
Ondrej Zajicek [Sun, 22 Jan 2023 17:12:04 +0000 (18:12 +0100)]
VRF: Fix issues with reconfiguration
Protocols receive if_notify() announcements that are filtered according
to their VRF setting, but during reconfiguration, they access iface_list
directly and forgot to check VRF setting here, which leads to all
interfaces be addedd.
Fix this issue for Babel, OSPF, RAdv and RIP protocols.
Maria Matejka [Thu, 19 Jan 2023 09:49:47 +0000 (10:49 +0100)]
Decoupling loops from threads to allow fixed thread count
On large configurations, too many threads would spawn with one thread
per loop. Therefore, threads may now run multiple loops at once. The
thread count is configurable and may be changed during run. All threads
are spawned on startup.
This change helps with memory bloating. BIRD filters need large
temporary memory blocks to store their stack and also memory management
keeps its hot page storage per-thread.
Known bugs:
* Thread autobalancing is not yet implemented.
* Low latency loops are executed together with standard loops.
Maria Matejka [Thu, 19 Jan 2023 09:56:16 +0000 (10:56 +0100)]
Table prune routine doesn't walk over lists unless needed.
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.
Maria Matejka [Thu, 19 Jan 2023 09:53:09 +0000 (10:53 +0100)]
CLI closing fix when its action is run asynchronously.
Some CLI actions, notably "show route", are run by queuing an event
somewhere else. If the user closes the socket, in case such an action is
being executed, the CLI must free the socket immediately from the error
hook but the pool must remain until the asynchronous event finishes and
cleans everything up.
Maria Matejka [Wed, 18 Jan 2023 08:39:45 +0000 (09:39 +0100)]
Fix memory pre-allocation
When BIRD has no free memory mapped, it allocates several pages in
advance just to be sure that there is some memory available if needed.
This hysteresis tactics works quite well to reduce memory ping-ping with
kernel.
Yet it had a subtle bug: this pre-allocation didn't take a memory
coldlist into account, therefore requesting new pages from kernel even
in cases when there were other pages available. This led to slow memory
bloating.
To demonstrate this behavior fast enough to be seen well, you may:
* temporarily set the values in sysdep/unix/alloc.c as follows to
exacerbate the issue:
#define KEEP_PAGES_MAIN_MAX 4096
#define KEEP_PAGES_MAIN_MIN 1000
#define CLEANUP_PAGES_BULK 4096
* create a config file with several millions of static routes
* periodically disable all static protocols and then reload config
* log memory consumption
This should give you a steady growth rate of about 16kB per cycle. If
you don't set the values this high, the issue happens much more slowly,
yet after 14 days of running, you are going to see an OOM kill.
After this fix, pre-allocation uses the memory coldlist to get some hot
pages and the same test as described here gets you a perfectly stable
constant memory consumption (after some initial wobbling).
Thanks to NIX-CZ for reporting and helping to investigate this issue.
Thanks to Santiago for finding the cause in the code.
Ondrej Zajicek [Tue, 17 Jan 2023 16:13:50 +0000 (17:13 +0100)]
Alloc: Disable transparent huge pages
The usage pattern implemented in allocator seems to be incompatible with
transparent huge pages, as memory released using madvise(MADV_DONTNEED)
with regular page size and alignment does not seem to trigger demotion
of huge pages back to regular pages, even when significant number of
pages is released. Even if demotion is triggered when system memory
is low, it still breaks memory accounting.
Ondrej Zajicek [Fri, 13 Jan 2023 12:17:46 +0000 (13:17 +0100)]
BSD: Add support for kernel route metric
Add support for kernel route metric/priority, exported as krt_metric
attribute, like in Linux. This should also fix issues with overwriting
or removing system routes.
Ondrej Zajicek [Sat, 7 Jan 2023 19:18:44 +0000 (20:18 +0100)]
Filter: Change linearization of branches in switch instruction
Most branching instructions (FI_CONDITION, FI_AND, FI_OR) linearize its
branches in a recursive way, while FI_SWITCH branches are linearized
from parser even before the switch instruction is allocated.
Change linearization of FI_SWITCH branches to make it similar to other
branching instructions. This also fixes an issue with constant
switch evaluation, where linearized branch is mistaken for
non-linearized during switch construction.
Ondrej Zajicek [Tue, 3 Jan 2023 18:09:51 +0000 (19:09 +0100)]
BGP: Allow role specific keywords to be used as symbols
Some of these new BGP role keywords use generic names that collides with
user-defined symbols. Allow them to be redefined. Also remove duplicit
keyword definition for 'prefer'.
Ondrej Zajicek [Tue, 3 Jan 2023 16:01:54 +0000 (17:01 +0100)]
Nest: Fix leaking internal attributes in RIP and Babel
During backporting attribute changes from 3.0-branch, some internal
attributes (RIP iface and Babel seqno) leaked to 'show route all' output.
Allow protocols to hide specific attributes with GA_HIDDEN value.
Ondrej Zajicek [Sun, 1 Jan 2023 19:10:23 +0000 (20:10 +0100)]
Nest: Fix several issues with pflags
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).
The seqno request retransmission handling was tracking the destination
that a forwarded request was being sent to and always retransmitting to
that same destination. This is unnecessary because we only need to
retransmit requests we originate ourselves, not those we forward on
behalf of others; in fact retransmitting on behalf of others can lead to
exponential multiplication of requests, which would be bad.
So rework the seqno request tracking so that instead of storing the
destination of a request, we just track whether it was a request that we
forwarded on behalf of another node, or if it was a request we originated
ourselves. Forwarded requests are not retransmitted, they are only used
for duplicate suppression, and for triggering an update when satisfied.
If we end up originating a request that we previously forwarded, we
"upgrade" the old request and restart the retransmit counter.
One complication with this is that requests sent in response to unfeasible
updates (section 3.8.2.2 of the RFC) have to be sent as unicast to a
particular peer. However, we don't really need to retransmit those as
there's no starvation when sending such a request; so we just change
such requests to be one-off unicast requests that are not subject to
retransmission or duplicate suppression. This is the same behaviour as
babeld has for such requests.
Ondrej Zajicek [Sat, 10 Dec 2022 16:32:42 +0000 (17:32 +0100)]
CLI: Fix for long-lived sessions during high loads
When there is a continuos stream of CLI commands, cli_get_command()
always returns 1 (there is a new command). Anyway, the socket receive
buffer was reset only when there was no command at all, leading to a
strange behavior: after a while, the CLI receive buffer came to its end,
then read() was called with zero size buffer, it returned 0 which was
interpreted as EOF.
The patch fixes that by resetting the buffer position after each command
and moving remaining data at the beginning of buffer.
Thanks to Maria Matejka for examining the bug and for the original bugfix.
Ondrej Zajicek [Fri, 9 Dec 2022 21:43:27 +0000 (22:43 +0100)]
Doc: Document issue with import tables
The import table does not work reliably together with re-evaluation of
routes due to recursive next hops or flowspec validation. We will at
least document that here, as import tables are completely redesigned and
this issue is fixed in BIRD 3.x branch.