Jorge Boncompte [Fri, 4 Aug 2017 17:33:15 +0000 (19:33 +0200)]
bgpd: remove bogus change in attrhash_key_make()
Commit c8e7b895 ("bgpd: use Jenkins hash for BGP transit, cluster and
attr hashes") changed attrhash_key_make() to use Jenkins hash, commit c8f3fe30 ("bgpd: Remove AS Path limit/TTL functionality") introduced
a bogus change with a snippet of code that was deleted in the first
one.
Chirag Shah [Thu, 3 Aug 2017 23:34:17 +0000 (16:34 -0700)]
ospfd: Fix MI-OSPF configuraton clis
Multi-Instance OSPF configuration CLI would fail because
first client return error upon seeing qobj_index being 0.
With new marco generate new error code to return from each
instance (vtysh client) and if the command is intended for given
instance, its qobj_index would be nonzero and process the command
and push correct ospf context. Other instance would return the error.
On vtysh end, check all instance return an error log a message to a
file.
Testing Done:
Verfied various MI-OSPF configuration CLI with multi instances.
Signed-off-by: Chirag Shah <chirag@cumulusnetworks.com>
isisd: don't validate dict integrity for regular builds
This method is intended to be only used for debugging as per the author
and profiling shows we are spending a lot of cycles on it. Remove it for
regular builds by guarding it with a define.
Signed-off-by: Christian Franke <chris@opensourcerouting.org>
Donald Sharp [Thu, 3 Aug 2017 12:42:29 +0000 (08:42 -0400)]
lib: Add ETH_ALEN to prefix.h and deprecate ETHER_ADDR_LEN
From discussion we decided that we should use ETH_ALEN instead
of ETHER_ADDR_LEN. Add ETH_ALEN to prefix.h and make
ETHER_ADDR_LEN generate a warning when used( but still work ).
Signed-off-by: Donald Sharp <sharpd@cumulusnetworks.com>
Donald Sharp [Thu, 3 Aug 2017 11:43:47 +0000 (07:43 -0400)]
zebra: Fix crash when OOM happens.
The hash key function choosen for mac vni's would tend
to clump the key value to the same number. Use a better
hash key generator to spread the hash values out.
A bad hash key might lead to O(2^n) memory consumption
because the hash size is doubled, each time a backet
exceeds a predefined threshold. This quickly leads
to OOM. Fixing this issue by fixing the hash
key generation to actually spread the keys out.
Ticket: CM-17412 Signed-off-by: Donald Sharp <sharpd@cumulusnetworks.com>
zebra: do not treat kernel routes as implicit withdraws
Kernel does not send the best route after adding or deleting routes, if
we treat routes for an existing prefix as implicit withdraw the zebra RIB
goes out of sync with FIB and can announce wrong route to protocols.
host:~# vtysh -c 'show ip route'
S>* 0.0.0.0/0 [0/0] via 192.168.1.1, eth0
C>* 192.168.1.0/24 is directly connected, eth0
host:~# ip route add 192.0.2.0/24 via 192.168.1.101 metric 100
host:~# vtysh -c 'show ip route'
S>* 0.0.0.0/0 [0/0] via 192.168.1.1, eth0
K>* 192.0.2.0/24 via 192.168.1.101, eth0
C>* 192.168.1.0/24 is directly connected, eth0
host:~# ip route add 192.0.2.0/24 via 192.168.1.102 metric 50
host:~# vtysh -c 'show ip route'
S>* 0.0.0.0/0 [0/0] via 192.168.1.1, eth0
K>* 192.0.2.0/24 via 192.168.1.102, eth0
C>* 192.168.1.0/24 is directly connected, eth0
host:~# ip route del 192.0.2.0/24 via 192.168.1.102 metric 50
host:~# vtysh -c 'show ip route'
S>* 0.0.0.0/0 [0/0] via 192.168.1.1, eth0
C>* 192.168.1.0/24 is directly connected, eth0
host:~# ip route show 192.0.2.0/24
192.0.2.0/24 via 10.10.1.101 dev eth0 metric 100
RTA_PAYLOAD() return value depends on the platform bits.
make[5]: Nothing to be done for 'all-am'.
Making all in zebra
CC rt_netlink.o
../../zebra/rt_netlink.c: In function 'netlink_macfdb_change':
../../zebra/rt_netlink.c:1695:63: error: format '%ld' expects argument of type 'long int', but argument 7 has type 'unsigned int' [-Werror=format=]
"%s family %s IF %s(%u) brIF %u - LLADDR is not MAC, len %ld",
^
../../zebra/rt_netlink.c: In function 'netlink_ipneigh_change':
../../zebra/rt_netlink.c:2024:57: error: format '%ld' expects argument of type 'long int', but argument 6 has type 'unsigned int' [-Werror=format=]
"%s family %s IF %s(%u) - LLADDR is not MAC, len %ld",
^
Donald Sharp [Wed, 2 Aug 2017 17:16:51 +0000 (13:16 -0400)]
pimd: Squash some more memory leaks
When a interface is flapped we were leaking some memory
associated with link lists. Especially in the jp_agg
code. If your network was/is stable and you are not
running at any scale you probably would not see this
as impactful at all.
Ticket: CM-16392 Signed-off-by: Donald Sharp <sharpd@cumulusnetworks.com>
Donald Sharp [Wed, 2 Aug 2017 16:10:50 +0000 (12:10 -0400)]
pimd: Clean back pointer upon deletion
The channel_oil has a back pointer(up) to the upstream data structure.
If we are planning on keeping the channel oil (due to ref count issues)
longer than keeping the upstream, when we delete the upstream we were
not clearing the back pointer to up. This would result in a situation
where if that memory has started to be used again it will cause a
crash and other fun things.
Ticket: CM-17092 Signed-off-by: Donald Sharp <sharpd@cumulusnetworks.com>
Donald Sharp [Mon, 31 Jul 2017 13:58:42 +0000 (09:58 -0400)]
pimd: Remove almost duplicate pim->ifchannel_list
The pim->ifchannel_list can be duplicated by iterating
over each vrf and then over each pim_ifp->pim_ifchannel_list
Since list handling is taking allot of time at scale
convert over to using this value.
Additionally clean up pim_cmd.c to have helper functions
to handle the actual encoding of data for output.
Signed-off-by: Donald Sharp <sharpd@cumulusnetworks.com>
Renato Westphal [Wed, 2 Aug 2017 01:12:24 +0000 (22:12 -0300)]
lib: fix build from git repository on OpenBSD
The OpenBSD's cp(1) command doesn't support the -v option. This will
change in the next releases (starting from v6.2) but this patch fixes
the problem for v6.1 and older releases.
Donald Sharp [Wed, 2 Aug 2017 00:36:33 +0000 (20:36 -0400)]
lib: Remove sockopt_cork functionality
Linux -> TCP_CORK is enabled by default
*BSD -> The equivalent is TCP_NOPUSH
As such sockopt_cork is effectively useless
especially since the two places it is used
do it right before TCP writes then disable
it right after the writes are over.
Signed-off-by: Donald Sharp <sharpd@cumulusnetworks.com>
Daniel Walton [Tue, 1 Aug 2017 18:31:56 +0000 (18:31 +0000)]
bgpd: multipath change for VRF route is not updated in zebra
Signed-off-by: Daniel Walton <dwalton@cumulusnetworks.com>
If you are doing multipath in a VRF and bounce one of the multipaths for
a prefix, bgp is not updating the zebra entry for that prefix with the
new multipaths. We start with:
cel-redxp-10# show bgp vrf RED ipv4 unicast 6.0.0.16/32
BGP routing table entry for 6.0.0.16/32
Paths: (4 available, best #4, table RED)
Advertised to non peer-group peers:
spine-1(swp1) spine-2(swp2) spine-3(swp3) spine-4(swp4)
104 65104 65002
fe80::202:ff:fe00:2d from spine-4(swp4) (6.0.0.12)
(fe80::202:ff:fe00:2d) (used)
Origin incomplete, localpref 100, valid, external, multipath, bestpath-from-AS 104
AddPath ID: RX 0, TX 21
Last update: Tue Aug 1 18:28:33 2017
101 65104 65002
fe80::202:ff:fe00:21 from spine-1(swp1) (6.0.0.9)
(fe80::202:ff:fe00:21) (used)
Origin incomplete, localpref 100, valid, external, multipath, bestpath-from-AS 101, best
AddPath ID: RX 0, TX 8
Last update: Tue Aug 1 18:28:33 2017
cel-redxp-10#
cel-redxp-10# show ip route vrf RED 6.0.0.16/32
Routing entry for 6.0.0.16/32
Known via "bgp", distance 20, metric 0, vrf RED, best
Last update 00:00:25 ago
* fe80::202:ff:fe00:21, via swp1
* fe80::202:ff:fe00:25, via swp2
* fe80::202:ff:fe00:29, via swp3
* fe80::202:ff:fe00:2d, via swp4
cel-redxp-10#
And then on spine-1 we bounce all peers
spine-1# clear ip bgp *
spine-1#
On the leaf (cel-redxp-10) we remove the route from spine-1
cel-redxp-10# show ip route vrf RED 6.0.0.16/32
Routing entry for 6.0.0.16/32
Known via "bgp", distance 20, metric 0, vrf RED, best
Last update 00:00:01 ago
* fe80::202:ff:fe00:25, via swp2
* fe80::202:ff:fe00:29, via swp3
* fe80::202:ff:fe00:2d, via swp4
cel-redxp-10#
So far so good. The problem is when the session to spine-1 comes back up
bgp will mark the flag from spine-1 as `multipath` but does not update
zebra. We end up in a state where BGP has 4 paths flags as multipath but
only 3 paths are in the RIB.
Renato Westphal [Tue, 1 Aug 2017 00:37:46 +0000 (21:37 -0300)]
bgpd: don't make any assumptions about the size of an enum
The size of an enum is compiler dependent and thus we shouldn't use
enums inside structures that represent fields of a packet.
Problem detected by the 'test_capability' unit test.
The problem was not apparent before because the 'iana_safi_t' enum didn't
exist and 'safi_t' was a typedef to uint8_t. Now we have two different
enums, 'iana_afi_t' and 'iana_safi_t', and both need to be encoded in
different ways on the wire (2 bytes vs 1 byte).
Renato Westphal [Tue, 1 Aug 2017 00:09:01 +0000 (21:09 -0300)]
lib: remove SAFI_RESERVED_4 and SAFI_RESERVED_5
SAFI values have been a major source of confusion over the last few
years. That's because each SAFI needs to be represented in two different
ways:
* IANA's value used to send/receive packets over the network;
* Internal value used for array indexing.
In the second case, defining reserved values makes no sense because we
don't want to index SAFIs that simply don't exist. The sole purpose of
the internal SAFI values is to remove the gaps we have among the IANA
values, which would represent wasted memory in C arrays. With that said,
remove these reserved SAFIs to avoid further confusion in the future.
Chirag Shah [Mon, 31 Jul 2017 20:12:33 +0000 (13:12 -0700)]
ospf6d: adjust changes per PR 870 review
called list_delete instead of list_free
Moved MTYPE_STATIC in ospfd/zebra.c
Revert changes in ospf6_zebra.c where malloc is called for
multiple nexthops.
Signed-off-by: Chirag Shah <chirag@cumulusnetworks.com>
Daniel Walton [Mon, 31 Jul 2017 21:22:23 +0000 (21:22 +0000)]
bgpd: peer hash expands until we are out of memory
Signed-off-by: Daniel Walton <dwalton@cumulusnetworks.com>
swpX peers all start out with the same sockunion so initially they all
go into the same hash bucket. Once IPv6 ND has worked its magic they
will have different sockunions and will go in different buckets...life
is good.
Until then though, we are in a phase where all swpX peers have the same
socknunion. Once we have HASH_THRESHOLD (10) swpX peers and call
hash_get for a new swpX peer the hash code calls hash_expand(). This
happens because there are more than HASH_THRESHOLD entries in a single
bucket so the logic is "expand the hash to spread things out"...in our
case expanding doesn't spread out the swpX peers because all of their
sockunions are the same.
I looked at having peer_hash_make and peer_hash_same consider the ifname
of the swpX peer but that is a large change that we don't want to make
at the moment. So the fix is to put a cap on how large we are
willing to let the hash table get. By default there is no limit but if
max_size is set we will not allow the hash to expand above that.
projects / mirror_frr.git / log