Allow self-produced xxxNETCONF netlink messages through the BPF
filter we use. Just like address-configuration actions, we'll
process NETCONF changes in one path, whether the changes were
generated by zebra or by something else in the host OS.
Signed-off-by: Mark Stapp <mstapp@nvidia.com>
a) We'll need to pass the info up via some dataplane control method
(This way bsd and linux can both be zebra agnostic of each other)
b) We'll need to modify `struct interface *` to track this data
and when it changes to notify upper level protocols about it.
c) Work is needed to dump the entire mpls state at the start
so we can gather interface state. This should be done
after interface data gathering from the kernel.
Signed-off-by: Donald Sharp <sharpd@nvidia.com>
Signed-off-by: Mark Stapp <mstapp@nvidia.com>
Description:
===========
Change is intended for fixing the NHT resolution logic.
While recursively resolving nexthop, keep looking for a valid/useable route in the rib,
by not stopping at the first/most-specific route in the rib.
Consider the following set of events taking place on R1:
R1(config)# ip route 2.2.2.0/24 ens192
R1# sharp watch nexthop 2.2.2.32 connected
R1# show ip nht
2.2.2.32(Connected)
resolved via static
is directly connected, ens192
Client list: sharp(fd 33)
-2.2.2.32 NHT is resolved over the above valid static route.
R1# sharp install routes 2.2.2.32 nexthop 2.2.2.32 1
R1# 2.2.2.32(Connected)
resolved via static
is directly connected, ens192
Client list: sharp(fd 33)
-.32/32 comes which is going to resolve through itself, but since this is an invalid route,
it will be marked as inactive and will not affect the NHT.
R1# sharp install routes 2.2.2.31 nexthop 2.2.2.32 1
R1# 2.2.2.32(Connected)
unresolved(Connected)
Client list: sharp(fd 50)
-Now a .31/32 comes which will resolve over .32 route, but as per the current logic,
this will trigger the NHT check, in turn making the NHT unresolved.
-With fix, NHT should stay in resolved state as long as the valid static or connected route stays installed
Fix:
====
-While resolving nexthops, walk up the tree from the most-specific match,
walk up the tree without any ZEBRA_NHT_CONNECTED check.
Co-authored-by: Vishal Dhingra <vdhingra@vmware.com>
Co-authored-by: Kantesh Mundaragi <kmundaragi@vmware.com>
Signed-off-by: Iqra Siddiqui <imujeebsiddi@vmware.com>
Two minor changes:
1) Change `zebra_evpn_mac_gw_macip_add()` 's return type to `void`.
2) Since `zebra_evpn_mac_gw_macip_add()` has already `assert` the returned
`mac`, the check of its return value makes no sense. And keep setting
`mac->flags` inside `zebra_evpn_mac_gw_macip_add()` is more reasonable. So
just move the setting `mac->flags` inside `zebra_evpn_mac_gw_macip_add()`.
Signed-off-by: anlan_cs <vic.lan@pica8.com>
The recently-added hashtable of nlsock objects needs to be
thread-safe: it's accessed from the main and dplane pthreads.
Add a mutex for it, use wrapper apis when accessing it. Add
a per-OS init/terminate api so we can do init that's not
per-vrf or per-namespace.
Signed-off-by: Mark Stapp <mstapp@nvidia.com>
NetDEF CI has been whining about multiline string style.
Make the strings single-line and call it a day.
Signed-off-by: Trey Aspelund <taspelund@nvidia.com>
Trying to call multiple ioctl calls on ifreq will result in
overwriting ifreq with garbage data. On if_get_flags call,
try to keep the flags field safe from another possible ioctl
call before applying the flags field.
Modified code as per Code Review, done by Donald Sharp.
Signed-off-by: Bijan <bijanebrahimi@riseup.net>
Currently when the kernel sends netlink messages to FRR
the buffers to receive this data is of fixed length.
The kernel, with certain configurations, will send
netlink messages that are larger than this fixed length.
This leads to situations where, on startup, zebra gets
really confused about the state of the kernel. Effectively
the current algorithm is this:
read up to buffer in size
while (data to parse)
get netlink message header, look at size
parse if you can
The problem is that there is a 32k buffer we read.
We get the first message that is say 1k in size,
subtract that 1k to 31k left to parse. We then
get the next header and notice that the length
of the message is 33k. Which is obviously larger
than what we read in. FRR has no recover mechanism
nor is there a way to know, a priori, what the maximum
size the kernel will send us.
Modify FRR to look at the kernel message and see if the
buffer is large enough, if not, make it large enough to
read in the message.
This code has to be per netlink socket because of the usage
of pthreads. So add to `struct nlsock` the buffer and current
buffer length. Growing it as necessary.
Fixes: #10404
Signed-off-by: Donald Sharp <sharpd@nvidia.com>
Store the fd that corresponds to the appropriate `struct nlsock` and pass
that around in the dplane context instead of the pointer to the nlsock.
Modify the kernel_netlink.c code to store in a hash the `struct nlsock`
with the socket fd as the key.
Why do this? The dataplane context is used to pass around the `struct nlsock`
but the zebra code has a bug where the received buffer for kernel netlink
messages from the kernel is not big enough. So we need to dynamically
grow the receive buffer per socket, instead of having a non-dynamic buffer
that we read into. By passing around the fd we can look up the `struct nlsock`
that will soon have the associated buffer and not have to worry about `const`
issues that will arise.
Signed-off-by: Donald Sharp <sharpd@nvidia.com>
Store and use the sequence number instead of using what is in
the `struct nlsock`. Future commits are going away from storing
the `struct nlsock` and the copy of the nlsock was guaranteeing
unique sequence numbers per message. So let's store the
sequence number to use instead.
Signed-off-by: Donald Sharp <sharpd@nvidia.com>
Using memcmp is wrong because struct ipaddr may contain unitialized
padding bytes that should not be compared.
Signed-off-by: Igor Ryzhov <iryzhov@nfware.com>
When using wait for install there exists situations where
zebra will issue several route change operations to the kernel
but end up in a state where we shouldn't be at the end
due to extra data being received. Example:
a) zebra receives from bgp a route change, installs sends the
route to the kernel.
b) zebra receives a route deletion from bgp, removes the
struct route entry and then sends to the kernel a deletion.
c) zebra receives an asynchronous notification that (a) succeeded
but we treat this as a new route.
This is the ships in the night problem. In this case if we receive
notification from the kernel about a route that we know nothing
about and we are not in startup and we are doing asic offload
then we can ignore this update.
Ticket: #2563300
Signed-off-by: Donald Sharp <sharpd@nvidia.com>
The ctx->zd_is_update is being set in various
spots based upon the same value that we are
passing into dplane_ctx_ns_init. Let's just
consolidate all this into the dplane_ctx_ns_init
so that the zd_is_udpate value is set at the
same time that we increment the sequence numbers
to use.
As a note for future me's reading this. The sequence
number choosen for the seq number passed to the
kernel is that each context gets a copy of the
appropriate nlsock to use. Since it's a copy
at a point in time, we know we have a unique sequence
number value.
Signed-off-by: Donald Sharp <sharpd@nvidia.com>
When nl_batch_read_resp gets a full on failure -1 or an implicit
ack 0 from the kernel for a batch of code. Let's immediately
mark all of those in the batch pass/fail as needed. Instead
of having them marked else where.
Signed-off-by: Donald Sharp <sharpd@nvidia.com>
I'm seeing this crash in various forms:
Program terminated with signal SIGSEGV, Segmentation fault.
50 ../sysdeps/unix/sysv/linux/raise.c: No such file or directory.
[Current thread is 1 (Thread 0x7f418efbc7c0 (LWP 3580253))]
(gdb) bt
(gdb) f 4
267 (*func)(hb, arg);
(gdb) p hb
$1 = (struct hash_bucket *) 0x558cdaafb250
(gdb) p *hb
$2 = {len = 0, next = 0x0, key = 0, data = 0x0}
(gdb)
I've also seen a crash where data is 0x03.
My suspicion is that hash_iterate is calling zebra_nhg_sweep_entry which
does delete the particular entry we are looking at as well as possibly other
entries when the ref count for those entries gets set to 0 as well.
Then we have this loop in hash_iterate.c:
for (i = 0; i < hash->size; i++)
for (hb = hash->index[i]; hb; hb = hbnext) {
/* get pointer to next hash bucket here, in case (*func)
* decides to delete hb by calling hash_release
*/
hbnext = hb->next;
(*func)(hb, arg);
}
Suppose in the previous loop hbnext is set to hb->next and we call
zebra_nhg_sweep_entry. This deletes the previous entry and also
happens to cause the hbnext entry to be deleted as well, because of nhg
refcounts. At this point in time the memory pointed to by hbnext is
not owned by the pthread anymore and we can end up on a state where
it's overwritten by another pthread in zebra with data for other incoming events.
What to do? Let's change the sweep function to a hash_walk and have
it stop iterating and to start over if there is a possible double
delete operation.
Signed-off-by: Donald Sharp <sharpd@nvidia.com>
Add to `show zebra` whether or not RA is compiled into FRR
and whether or not BGP is using RFC 5549 at the moment.
Signed-off-by: Donald Sharp <sharpd@nvidia.com>
The dataplane pthread only processes a limited set of incoming
netlink notifications: only register for that set of events,
reducing duplicate incoming netlink messages.
Signed-off-by: Mark Stapp <mstapp@nvidia.com>
Current code treats all metaqueues as lists of route_node structures.
However, some queues contain other structures that need to be cleaned up
differently. Casting the elements of those queues to struct route_node
and dereferencing them leads to a crash. The crash may be seen when
executing bgp_multi_vrf_topo2.
Fix the code by using the proper list element types.
Signed-off-by: Igor Ryzhov <iryzhov@nfware.com>
The name 'opaque' is a little general - call the route_entry
struct 're_opaque' to make it more specific.
Signed-off-by: Mark Stapp <mstapp@nvidia.com>
RA packets are pretty chatty and when there is a warning from
a missconfiguration on the network, the log file gets filed
up with warnings. Modify the code in rtadv.c to only spit
out the warning in these cases at most every 6 hours.
Signed-off-by: Donald Sharp <sharpd@nvidia.com>
EVPN route add should be queued to preserve the config order.
In particular, against deletion in rib_delete().
Signed-off-by: Xiao Liang <shaw.leon@gmail.com>
When a operator has a FRR based route installed into the
FIB and a better route comes in from the system. There
is code in the data plane to schedule the batching
and continue processing. But in this case we are done
so we can just return
Signed-off-by: Donald Sharp <sharpd@nvidia.com>
vrf_disable is always called first before
vrf_delete. The rnh_table and rnh_table_multicast tables
are already deleted as part of vrf_disable. No need
to do it again.
Signed-off-by: Donald Sharp <sharpd@nvidia.com>
VRF name should not be printed in the config since 574445ec. The update
was done for NB config output but I missed it for regular vty output.
Signed-off-by: Igor Ryzhov <iryzhov@nfware.com>
Add a thread_ignore_late_timer(struct thread *thread) function
that allows thread.c to ignore when timers are late to the party.
Signed-off-by: Donald Sharp <sharpd@nvidia.com>
Pass in the route_node that is under consideration
into route_notify_internal to allow calling functions
to reduce stack size as well as looking up data.
Signed-off-by: Donald Sharp <sharpd@nvidia.com>
The dest_pfx was pretty much only ever used for
debug output and FRR already knows the rn. So
use that instead.
Signed-off-by: Donald Sharp <sharpd@nvidia.com>
the dest_p and src_p values were only ever used for
debugs and %pFX, when we already have the rn.
There is no need to do this lookup
Signed-off-by: Donald Sharp <sharpd@nvidia.com>
FRR must give variable names instead of not defining
them in the .h file. This just cleans up this
problem for redistribute.h
Signed-off-by: Donald Sharp <sharpd@nvidia.com>
The function zsend_redistribute_route uses the prefix and
source prefix. Just pass in the route_node instead.
Signed-off-by: Donald Sharp <sharpd@nvidia.com>
FRR is passing around a bunch of data that is encapsulated
within the route node. Let's just pass that around instead.
Signed-off-by: Donald Sharp <sharpd@nvidia.com>
Signed-off-by: Donald Sharp <sharpd@nvidia.com>
FRR is passing around a bunch of data that is encapsulated
within the route node. Let's just pass that around instead.
Signed-off-by: Donald Sharp <sharpd@nvidia.com>
If you have this setup:
router ospf 3
redistribute sharp
!
and then install:
sharp install route 4.5.6.7 nexthop 192.168.100.1 1
sharp install route 4.5.6.8 nexthop 192.168.100.1 1 instance 3
sharp install route 4.5.6.9 nexthop 192.168.100.1 1 instance 4
The .8 and .9 routes are auto redistributed into ospf instance 3:
eva# show ip ospf data
OSPF Instance: 3
OSPF Router with ID (192.168.122.1)
AS External Link States
Link ID ADV Router Age Seq# CkSum Route
4.5.6.7 192.168.122.1 13 0x80000001 0x477c E2 4.5.6.7/32 [0x0]
4.5.6.8 192.168.122.1 5 0x80000001 0x3d85 E2 4.5.6.8/32 [0x0]
4.5.6.9 192.168.122.1 5 0x80000001 0x338e E2 4.5.6.9/32 [0x0]
This cannot be correct behavior. When redistributing in the absense
of an instance number the default instance of 0 should be used and should
be the only route redistributed. Here is the correct behavior:
eva# show ip route
Codes: K - kernel route, C - connected, S - static, R - RIP,
O - OSPF, I - IS-IS, B - BGP, E - EIGRP, N - NHRP,
T - Table, v - VNC, V - VNC-Direct, A - Babel, D - SHARP,
F - PBR, f - OpenFabric,
> - selected route, * - FIB route, q - queued, r - rejected, b - backup
t - trapped, o - offload failure
K>* 0.0.0.0/0 [0/100] via 192.168.119.1, enp39s0, 00:00:28
D>* 4.5.6.7/32 [150/0] via 192.168.100.1, virbr1, weight 1, 00:00:02
D[3]>* 4.5.6.8/32 [150/0] via 192.168.100.1, virbr1, weight 1, 00:00:02
D[4]>* 4.5.6.9/32 [150/0] via 192.168.100.1, virbr1, weight 1, 00:00:02
C>* 192.168.100.0/24 is directly connected, virbr1, 00:00:28
C>* 192.168.110.0/24 is directly connected, virbr2, 00:00:28
C>* 192.168.119.0/24 is directly connected, enp39s0, 00:00:28
C>* 192.168.122.0/24 is directly connected, virbr0, 00:00:28
eva# show ip ospf data
OSPF Instance: 3
OSPF Router with ID (192.168.122.1)
AS External Link States
Link ID ADV Router Age Seq# CkSum Route
4.5.6.7 192.168.122.1 6 0x80000001 0x477c E2 4.5.6.7/32 [0x0]
Signed-off-by: Donald Sharp <sharpd@nvidia.com>
FRR allows redistribution to a client with a specific
instance in mind. The code was not allowing you to figure
out what instance was being looked at. So let's clarify this
in the debugs.
Signed-off-by: Donald Sharp <sharpd@nvidia.com>
Update ospfd and ospf6d to send opaque route attributes to
zebra. Those attributes are stored in the RIB and can be viewed
using the "show ip[v6] route" commands (other than that, they are
completely ignored by zebra).
Example:
```
debian# show ip route 192.168.1.0/24
Routing entry for 192.168.1.0/24
Known via "ospf", distance 110, metric 20, best
Last update 01:57:08 ago
* 10.0.1.2, via eth-rt2, weight 1
OSPF path type : External-2
OSPF tag : 0
debian#
debian# show ip route 192.168.1.0/24 json
{
"192.168.1.0\/24":[
{
"prefix":"192.168.1.0\/24",
"prefixLen":24,
"protocol":"ospf",
"vrfId":0,
"vrfName":"default",
"selected":true,
[snip]
"ospfPathType":"External-2",
"ospfTag":"0"
}
]
}
```
Signed-off-by: Renato Westphal <renato@opensourcerouting.org>
Topology:
IXIA-----(ens192)FRR(ens224)------iXIA
Configuration:
1. Create 8 sub-interfaces on ens192 under Default VRF and configure 8
EBGP session between FRR and IXIA.
2. Create 1000 sub-interfaces on ens224 under Default VRF and configure
1000 EBGP session between FRR and IXIA.
3. 2M prefixes distributed from Left side Ixia each with 8 ECMP path.
4. So in total, there are 2M prefixes * 8 ECMP = 16M prefixes entries
in RIB and FIB.
Issue:
Shut ens192 and ens224, this is taking 1hr 15 mins to clean up the routes.
Root Cause:
In the case of route deletion, if the particular route node is having
nht count = 0, we are going to the parent and doing nht evaluation,
which is not needed.
Fix:
If the deleted the route node is having nht count > 0, then do a nht
evaluation on the parent node.
Shut ens192 and ens224, it is taking 1 min to clean up the routes
with the fix.
Signed-off-by: Sarita Patra <saritap@vmware.com>
Used for graceful-restart mostly.
Especially for bgp_show_neighbor_graceful_restart_capability_per_afi_safi()
Signed-off-by: Donatas Abraitis <donatas.abraitis@gmail.com>
This function is sure to return correct value by "assert", so the
checking its return value should be removed.
Signed-off-by: anlan_cs <anlan_cs@tom.com>
Currently, it is possible to rename the default VRF either by passing
`-o` option to zebra or by creating a file in `/var/run/netns` and
binding it to `/proc/self/ns/net`.
In both cases, only zebra knows about the rename and other daemons learn
about it only after they connect to zebra. This is a problem, because
daemons may read their config before they connect to zebra. To handle
this rename after the config is read, we have some special code in every
single daemon, which is not very bad but not desirable in my opinion.
But things are getting worse when we need to handle this in northbound
layer as we have to manually rewrite the config nodes. This approach is
already hacky, but still works as every daemon handles its own NB
structures. But it is completely incompatible with the central
management daemon architecture we are aiming for, as mgmtd doesn't even
have a connection with zebra to learn from it. And it shouldn't have it,
because operational state changes should never affect configuration.
To solve the problem and simplify the code, I propose to expand the `-o`
option to all daemons. By using the startup option, we let daemons know
about the rename before they read their configs so we don't need any
special code to deal with it. There's an easy way to pass the option to
all daemons by using `frr_global_options` variable.
Unfortunately, the second way of renaming by creating a file in
`/var/run/netns` is incompatible with the new mgmtd architecture.
Theoretically, we could force daemons to read their configs only after
they connect to zebra, but it means adding even more code to handle a
very specific use-case. And anyway this won't work for mgmtd as it
doesn't have a connection with zebra. So I had to remove this option.
Signed-off-by: Igor Ryzhov <iryzhov@nfware.com>
Add optional NHG ID output to `show ip route` dumps. We have
this in json output already as nexthopGroupID but nice
to have the option in a normal dump as well. Not including in main
output for now to avoid breaking screen scrapers.
Signed-off-by: Stephen Worley <sworley@nvidia.com>
On startup we create a thread timer event to do a rib sweep
of the system. On shutdown we never stopped this timer and
as such we have a situation where a thread event could be run
on shutdown after the data for it has been freed. Here is the
crash I am seeing:
(gdb) bt
(gdb)
Save the thread data in zebra_router and stop the thread so we don't
accidently do work on shutdown we don't mean to. In this case
it happened in our topotests with some severe system load.
Essentially we happened to kill the zebra daemon just as the
graceful_restart timer popped here.
Signed-off-by: Donald Sharp <sharpd@nvidia.com>
In some cases, zebra may install a nexthop-group id that is
different from the id of the nhe struct attached to a
route-entry. This happens for a singleton recursive nexthop,
for example, where a route is installed with the resolving
nexthop's id.
The installed value is the most useful value - that corresponds
to information in the kernel on linux/netlink platforms that
support nhgs. Display both values if they differ in ascii
output, and include both values in the json form.
Signed-off-by: Mark Stapp <mstapp@nvidia.com>
Since f60a1188 we store a pointer to the VRF in the interface structure.
There's no need anymore to store a separate vrf_id field.
Signed-off-by: Igor Ryzhov <iryzhov@nfware.com>
During zebra shutdown, we clear out the LSP workqueue. The LSPs
will be uninstalled and freed during the shutdown process, so
just ignore any LSPs that happen to be on the workqueue.
Signed-off-by: Mark Stapp <mstapp@nvidia.com>
At some scale we eventually run out of room displaying v4/v6 route
totals for `show zebra client summ`:
janelle# show zebra client summ
Name Connect Time Last Read Last Write IPv4 Routes IPv6 Routes
--------------------------------------------------------------------------------
bgp 04w0d18h 00:00:19 00:01:2411729127/4052681 2037786/903094
This total over ran the space in just a little over a week of uptime.
Expand to have a bit more room.
Signed-off-by: Donald Sharp <sharpd@nvidia.com>
The dplane_ctx_get_pbr_ipset_entry function only
failed when the caller did not pass in a valid
usable pointer. Change the code to assert on
a pointer not being passed in and remove the
bool return
Signed-off-by: Donald Sharp <sharpd@nvidia.com>
The only time this function ever failed is when
the developer does not pass in a usable pointer
to place the data in. Change it to an assert
to signify to the end developer that is what
we want and then remove all the if checks
for failure
Signed-off-by: Donald Sharp <sharpd@nvidia.com>
The function call dplane_ctx_get_pbr_ipset only
returns false when the calling function fails to
pass in a valid ipset pointer. This should
be an assertion issue since it's a programming
issue as opposed to an actual run time issue.
Change the function call parameter to not return
a bool on success/fail for a compile time decision.
Signed-off-by: Donald Sharp <sharpd@nvidia.com>
We should always treat the VRF interface as a loopback. Currently, this
is not the case, because in some old pre-VRF code we use if_is_loopback
instead of if_is_loopback_or_vrf. To avoid any future problems, the
proposal is to rename if_is_loopback_or_vrf to if_is_loopback and use it
everywhere. if_is_loopback is renamed to if_is_loopback_exact in case
it's ever needed, but currently it's not used anywhere.
Signed-off-by: Igor Ryzhov <iryzhov@nfware.com>
During shutdown, when table_manager_disable is called for the default
VRF, its vrf_id is already set to VRF_UNKNOWN, so the expression is true
and the table manager memory is not freed. Change the expression to
compare the VRF name instead of the id. The check in table_manager_enable
is changed for consistency.
Signed-off-by: Igor Ryzhov <iryzhov@nfware.com>
Free the LSP workqueue later during shutdown, so that zebra
has enough time to clean up and uninstall any LSPs.
Signed-off-by: Mark Stapp <mstapp@nvidia.com>
42d4b30e introduced per-VRF table manager.
Table manager is allocated when the VRF is created, but it is freed when
the VRF is disabled. When this VRF is re-enabled, zebra ends up with
table manager being NULL pointer and it crashes on any dereference.
Table manager should be freed when the VRF is deleted, not when it's
disabled.
Signed-off-by: Igor Ryzhov <iryzhov@nfware.com>
We don't receive interface down/delete notifications from kernel when a
netns is deleted. Therefore we have to manually replicate the necessary
actions, otherwise interfaces are kept in the system with stale pointers
to the deleted netns.
Signed-off-by: Igor Ryzhov <iryzhov@nfware.com>
Problem:
L2-VNI SVI down followed by L2-VNI's vxlan device
deletion leads to stale entry into L3VNI's
L2-VNI list.
Solution:
When L2-VNI associated SVI is down, default vrf
is the new tenant vrf.
Remove L2-VNI from L3VNI's l2vni list as
L3VNI/VRF is no longer valid in absence of associated
SVI.
When SVI is up re-add L2-VNI into associated VRF's
L3VNI.
The above remove/add from the L3VNI's L2VNI list is
already done when vxlan or L2-VNI is flaped, just need
to handle when SVI is flapped.
Ticket:#2817127
Reviewed By:
Testing Done:
After deleting SVI following by L2-VNI deletion,
L3VNI's L2-VNI list delets the L2-VNI. (no stale entry).
After adding back SVI/L2-VNI, L3VNI list adds back the
L2-VNI and it is associated right tenant VRF.
Signed-off-by: Chirag Shah <chirag@nvidia.com>
In the rtadv_timer(), it always uses the zvrf's socket to send RA
packets. In the vrf-lite mode, it's righ since it uses the default
vrf to send the RA packets. But in the netns mode, it uses socket
in each netns. So the issue only happens in the netns mode because
the zvrf's socket may not be in the same netns as the interface's
netns. In order to compatible with both vrf-lite and netns mode,
the fix uses the if_lookup_by_index() to check whether interfaces
can use the zvrf's socket.
Signed-off-by: LEI BAO <bali.baolei@cn.ibm.com>
Before 42d4b30e, table_manager_enable was called only once and the hook
was also registered once. After the change, the hook is registered per
each VRF that is created in the system. This is wrong.
Signed-off-by: Igor Ryzhov <iryzhov@nfware.com>
Currently the NEXTHOP_TYPE_IPV4 and NEXTHOP_TYPE_IPV6 are
not sending up the resolved ifindex for the route. This
is causing upper level protocols that have something like
this:
route-map FOO permit 10
match interface swp13
!
router ospf
redistribute static
!
ip route 4.5.6.7/32 10.10.10.10
where 10.10.10.10 resolves to interface swp13. The route-map
will never match in this case.
Since FRR has the resolved nexthop interface, FRR might as
well send it up to be selected on by the upper level protocol
as needed.
Signed-off-by: Donald Sharp <sharpd@nvidia.com>
It appears that without that change, there were no notifications
sent to bgp daemon, after flowspec operations have been sent to
zebra.
Signed-off-by: Philippe Guibert <philippe.guibert@6wind.com>
It is needed for the ipset entry to know for which address family
this ipset entry applies to. Actually, the family is in the original
ipset structure and was not passed as attribute in the dataplane
ipset_info structure. Add it.
Signed-off-by: Philippe Guibert <philippe.guibert@6wind.com>
When injecting an ipset entry into the zebra dataplane context, the
ipset name is stored in a separate structure. This will permit the
flowspec plugin to be able to know which ipset has to be appended with
relevant ipset entry.
The problem was that the zebra dataplane objects related to ipset entries
is made up of an union between the ipset structure and the ipset info
structure. This was implying that the two structures were on the same
memory zone, and when extracting the data stored, the data were incomplete.
Fix this by replacing the union structure by a defined struct.
Signed-off-by: Philippe Guibert <philippe.guibert@6wind.com>
When the netns is deleted, we should always clear the vrf->ns_ctxt
pointer. Currently, it is not cleared when there are interfaces in the
netns at the time of deletion.
If the netns is re-created, zebra crashes because it tries to use the
stale pointer.
Signed-off-by: Igor Ryzhov <iryzhov@nfware.com>
if_lookup_by_index_all_vrf doesn't work correctly with netns VRF backend
as the same index may be used in multiple netns simultaneously.
In both case where it's used, we know the VRF in which we need to lookup
for the interface.
Signed-off-by: Igor Ryzhov <iryzhov@nfware.com>
The kernel can return to us nested attributes for BRIDGE RTM_NEWNEIGH
attributes. Just ensure that we can parse and read them.
Signed-off-by: Donald Sharp <sharpd@nvidia.com>
With the addition of resillient hashing for nexthops, the
parsing of nexthops requires telling the decoder functions
that there may be nested attributes. This was found by
code inspection of iproute2/ipnexthop.c when trying to
understand resillient hashing as well as statistics
gathering for nexthops that are / will be in upstream
kernels in the near future.
Signed-off-by: Donald Sharp <sharpd@nvidia.com>
Add actual recent nexthop.h file from kernel
and fix up resulting fallout because FRR's
original nexthop.h did not match upstream
linux kernel.
Signed-off-by: Donald Sharp <sharpd@nvidia.com>
when gre information could not be retrieved because GRE interface has
been deleted, a GRE_UPDATE message may be sent to NHRP. In that case,
the gre values are reset. There was a missing tunnel destination value,
which has been omitted.
Signed-off-by: Philippe Guibert <philippe.guibert@6wind.com>
There is a bit of an impedance mismatch in the sequence of events here.
Depending on the dplane behavior, the `ROUTE_ENTRY_SELECTED` bit will be
inconsistent for rib_process_result().
With an asynchronous dataplane:
0. rib_process() is called
1. rib_install_kernel() is called, dplane action is queued
2. rib_install_kernel() returns
3. rib_process() sets the SELECTED bit appropriately, returns
4. dplane is done, triggers rib_process_result()
5. SELECTED bit is seen in "after" state
(5a. NHT code looks at the SELECTED bit, works correctly.)
With a synchronous dataplane:
0. rib_process() is called
1. rib_install_kernel() is called, dplane action is executed
2. dplane (should) trigger rib_process_result()
3. SELECTED bit is seen in "before" state
(3a. NHT code looks at the SELECTED bit, fails.)
4. rib_install_kernel() returns
5. rib_process() sets the SELECTED bit appropriately, too late.
Essentially, poking the dataplane is a sequencing point where control is
handed over to the dplane. Control may or may not return immediately.
Doing /anything/ after triggering the dataplane is a recipe for odd race
conditions.
(FWIW, I'm not sure rib_process_result() is called correctly in the
synchronous case, but that's a separate problem.)
Unfortunately, this change might have some unforeseen side effects. I
haven't dug through the code to see if anything breaks. There
/shouldn't/ be anything looking at the SELECTED bit here, but who knows.
Signed-off-by: David Lamparter <equinox@opensourcerouting.org>
Do not return pointer to the newly created thread from various thread_add
functions. This should prevent developers from storing a thread pointer
into some variable without letting the lib know that the pointer is
stored. When the lib doesn't know that the pointer is stored, it doesn't
prevent rescheduling and it can lead to hard to find bugs. If someone
wants to store the pointer, they should pass a double pointer as the last
argument.
Signed-off-by: Igor Ryzhov <iryzhov@nfware.com>
rib_update() was mallocing memory then attempting to schedule
and if the schedule failed( it was already going to be run )
FRR would then free the memory. Fix this memory usage pattern
Signed-off-by: Donald Sharp <sharpd@nvidia.com>
It allows FRR to read the interface config even when the necessary VRFs
are not yet created and interfaces are in "wrong" VRFs. Currently, such
config is rejected.
For VRF-lite backend, we don't care at all about the VRF of the inactive
interface. When the interface is created in the OS and becomes active,
we always use its actual VRF instead of the configured one. So there's
no need to reject the config.
For netns backend, we may have multiple interfaces with the same name in
different VRFs. So we care about the VRF of inactive interfaces. And we
must allow to preconfigure the interface in a VRF even before it is
moved to the corresponding netns. From now on, we allow to create
multiple configs for the same interface name in different VRFs and
the necessary config is applied once the OS interface is moved to the
corresponding netns.
Signed-off-by: Igor Ryzhov <iryzhov@nfware.com>
When something is used only from zebra and part of its description is
"should be called from zebra only" then it belongs to zebra, not lib.
Signed-off-by: Igor Ryzhov <iryzhov@nfware.com>
When an ES is deleted and re-added bgpd can start sending MAC-IP sync updates
before the dataplane and zebra have setup the VLAN membership for the ES. Such
MAC entries are not installed in the dataplane till the ES-EVI is created.
Ticket: #2668488
Signed-off-by: Anuradha Karuppiah <anuradhak@nvidia.com>
In the window immediately after an ES deletion bgpd can send MAC-IP updates
using that ES. Zebra needs to ignore these updates to prevent creation
of stale entries.
Ticket: #2668488
Signed-off-by: Anuradha Karuppiah <anuradhak@nvidia.com>
This addresses deletion of ES interfaces that are were not completely
configured.
Ticket: #2668488
Signed-off-by: Anuradha Karuppiah <anuradhak@nvidia.com>
When PTM sends a "cbl status" message it specifies the interface name
but not the VRF name. It is fine for VRF-lite, but doesn't work for
netns because it's possible to have multiple interfaces with the same
name. Be more restrictive in this case and return an error instead of
randomly using of the interface with the specified name.
Signed-off-by: Igor Ryzhov <iryzhov@nfware.com>
With netns VRF backend, we may have multiple interfaces with the same
name. Currently, the function output is not deterministic in this case,
it returns the first interface that it finds in the list. Be more
explicit and tell the user that we need the VRF name.
Signed-off-by: Igor Ryzhov <iryzhov@nfware.com>
```
exit1-debian-9# show ip route 172.16.16.1/32
Routing entry for 172.16.16.1/32
Known via "bgp", distance 20, metric 0, best
Last update 00:00:28 ago
* 192.168.0.2, via eth1, weight 1
AS-Path : 65003
Communities : first 65001:2 65001:3
Large-Communities: 65001:1:1 65001:1:2 65001:1:3
Selection reason : First path received
```
Signed-off-by: Donatas Abraitis <donatas.abraitis@gmail.com>
Currently, the ll_type is set only in `netlink_interface` which is
executed only during startup. If the interface is created when the FRR
is already running, the type is not stored.
Fixes#1164.
Signed-off-by: Igor Ryzhov <iryzhov@nfware.com>
When a client sends to zebra that GR mode is being turned
on. The client also passes down the time zebra should hold
onto the routes. Display this time with the output
of the `show zebra client` command as well.
Signed-off-by: Donald Sharp <sharpd@nvidia.com>
When issuing the `show zebra client` command data about
Graceful Restart state is being printed 2 times.
Signed-off-by: Donald Sharp <sharpd@nvidia.com>
In startup, zebra would dump interface information from Kernel in 3
steps w/o lock: step1, get interface information; step2, get interface
ipv4 address; step3, get interface ipv6 address.
If any interface gets added after step1, but before step2/3, zebra
would get extra interface addresses in step2/3 that has not been added
into zebra in step1. Returning error in the referenced interface lookup
would cause the startup interface retrieval to be incomplete.
Signed-off-by: Yuan Yuan <yyuanam@amazon.com>
FRR should only ever use the appropriate THREAD_ON/THREAD_OFF
semantics. This is espacially true for the functions we
end up calling the thread for.
Signed-off-by: Donatas Abraitis <donatas.abraitis@gmail.com>
There's a helper function to check whether the interface is loopback or
VRF - if_is_loopback_or_vrf. Let's use it whenever we need to check that.
There's no functional change in this commit.
Signed-off-by: Igor Ryzhov <iryzhov@nfware.com>
Pass down the safi for when we need address
resolution. At this point in time we are
hard coding the safi to SAFI_UNICAST.
Future commits will take advantage of this.
Signed-off-by: Donald Sharp <sharpd@nvidia.com>
PIM is going to need to be able to send down the address it is
trying to resolve in the multicast rib. We need a way to signal
this to the end developer. Start the conversion by adding the
ability to have a safi. But only allow SAFI_UNICAST at the moment.
Signed-off-by: Donald Sharp <sharpd@nvidia.com>
The entirety of the import checking no longer needs to be
in zebra as that no-one is calling it. Remove the code.
Signed-off-by: Donald Sharp <sharpd@nvidia.com>
These are no longer really needed. The client just needs
to call nexthop resolution instead.
So let's remove the zapi types.
Signed-off-by: Donald Sharp <sharpd@nvidia.com>
There were two identical blocks of code run at init time that
requested info about AF_BRIDGE - don't see any reason to do that
twice, so remove one block.
Signed-off-by: Mark Stapp <mstapp@nvidia.com>
Because vrf backend may be based on namespaces, each vrf can
use in the [16-(2^32-1)] range table identifier for daemons that
request it. Extend the table manager to be hosted by vrf.
That possibility is disabled in the case the vrf backend is vrflite.
In that case, all vrf context use the same table manager instance.
Add a configuration command to be able to configure the wished
range of tables to use. This is a solution that permits to give
chunks to bgp daemon when it works with bgp flowspec entries and
wants to use specific iptables that do not override vrf tables.
Signed-off-by: Philippe Guibert <philippe.guibert@6wind.com>
When using bgp evpn rt5 setup, after BGP configuration has been
loaded, if the user attempts to detach and reattach the bridged
vxlan interface from the bridge, then BGP loses its BGP EVPN
contexts, and a refresh of BGP configuration is necessary to
maintain consistency between linux configuration and BGP EVPN
contexts (RIB). The following command can lead to inconsistency:
ip netns exec cust1 ip link set dev vxlan1000 nomaster
ip netns exec cust1 ip link set dev vxlan1000 master br1000
consecutive to the, BGP l2vpn evpn RIB is empty, and the way to
solve this until now is to reconfigure EVPN like this:
vrf cust1
no vni 1000
vni 1000
exit-vrf
Actually, the link information is correctly handled. In fact,
at the time of link event, the lower link status of the bridge
interface was not yet up, thus preventing from establishing
BGP EVPN contexts. In fact, when a bridge interface does not
have any slave interface, the link status of the bridge interface
is down. That change of status comes a bit after, and is not
detected by slave interfaces, as this event is not intercepted.
This commit intercepts the bridge link up event, and triggers
a check on slaved vxlan interfaces.
Signed-off-by: Philippe Guibert <philippe.guibert@6wind.com>
when running bgp evpn rt5 setup, the Rmac sent in BGP updates
stands for the MAC address of the bridge interface. After
having loaded frr configuration, the Rmac address is not refreshed.
This issue can be easily reproduced by executing some commands:
ip netns exec cust1 ip link set dev br1000 address 2e🆎45:aa:bb:cc
Actually, the BGP EVPN contexts are kept unchanged.
That commit proposes to fix this by intercepting the mac address
change, and refreshing the vxlan interfaces attached to te bridge
interface that changed its MAC address.
Signed-off-by: Philippe Guibert <philippe.guibert@6wind.com>
We should not be using `case default` with an enumerated type
This prevents the developer of new cases from knowing where
they need to fix by just compiling.
Signed-off-by: Donald Sharp <sharpd@nvidia.com>
Move the handler for incoming interface address events
to a neutral source file - it's not netlink-specific and
shouldn't have been in a netlink file.
Signed-off-by: Mark Stapp <mjs.ietf@gmail.com>
Read incoming interface address change notifications in the
dplane pthread; enqueue the events to the main pthread
for processing. This is netlink-only for now - the bsd
kernel socket path remains unchanged.
Signed-off-by: Mark Stapp <mjs.ietf@gmail.com>
Add new apis for dplane interface address handling, based on
the existing api. The existing api is basically split in two:
the first part processes an incoming netlink message in the
dplane pthread, creating a dplane context with info about
the event. The second part runs in the main pthread and uses
the context data to update an interface or connected object.
Signed-off-by: Mark Stapp <mjs.ietf@gmail.com>
Add a new netlink socket for events coming in from the host OS
to the dataplane system for processing. Rename the existing
outbound dplane socket.
Signed-off-by: Mark Stapp <mjs.ietf@gmail.com>
Description: Currently IPv4 routes with IPv6 link local next hops are
not properly installed in FPM.
Reason is the netlink decoding truncates the ipv6 LL address to 4 byte
ipv4 address.
Ex : fe80:: is directly converted to ipv4 and it results in 254.128.0.0
as next hop for below routes
show ip route
Codes: K - kernel route, C - connected, S - static, R - RIP,
O - OSPF, I - IS-IS, B - BGP, E - EIGRP, N - NHRP,
T - Table, v - VNC, V - VNC-Direct, A - Babel, D - SHARP,
F - PBR, f - OpenFabric,
> - selected route, * - FIB route, q - queued, r - rejected, b - backup
B>* 2.1.0.0/16 [200/0] via fe80::268a:7ff:fed0:d40, Ethernet0, weight 1,
02:22:26
B>* 5.1.0.0/16 [200/0] via fe80::268a:7ff:fed0:d40, Ethernet0, weight 1,
02:22:26
B>* 10.1.0.2/32 [200/0] via fe80::268a:7ff:fed0:d40, Ethernet0, weight
1, 02:22:26
Hence this fix converts the ipv6-LL address to ipv4-LL (169.254.0.1)
address before sending it to FPM. This is inline with how these types of
routes are currently programmed into kernel.
Signed-off-by: Nikhil Kelapure <nikhil.kelapure@broadcom.com>
Current implementation doesn't copy nexthop_srv6. This causes unexpected
behavior when receiving SID information and nexthop isn't onlink.t
Signed-off-by: Ryoga Saito <contact@proelbtn.com>
Problem:
When IP1:M1 (local) moved to IP1:M2 (remote-VTEP) bgpd continues to
advertise IP1:M1.
Fix:
Local path del is sent to bgp if the neigh was {local-active||peer-active}.
So path del needs to be called before the sync flags (including peer-active)
are cleared.
Ticket: #2706744
Signed-off-by: Anuradha Karuppiah <anuradhak@nvidia.com>
When we hand set the router-id, but we have choosen a router-id
that is already the `winner` there is no point in updating anyone
with this data.
Signed-off-by: Donald Sharp <sharpd@nvidia.com>
At startup there exists a time frame where we might not know
a particular vrf's router id. When zebra gets a request for
it let's not just blindly send whatever we have. Let's be
a bit smart and only respond with one if we have one.
The upper level protocol can wait for it to have one.
Signed-off-by: Donald Sharp <sharpd@nvidia.com>
vrf_name_to_id() returned VRF_DEFAULT when the vrf name was
unknown, hiding errors. Per community recommendation, vrf_name_to_id()
is now removed and the few callers now use vrf_lookup_by_name()
directly.
Signed-off-by: G. Paul Ziemba <paulz@labn.net>
When running bgp evpn rt5 setup with vrf namespace backend, once the
BGP configuration loaded, some refresh like the config change of a
vxlan interface is not taken into account. As consequence, the BGP
l2vpn evpn entries are empty. This can happen by recreating vxlan
interface like follows:
ip netns exec cust1 ip li del vxlan1000
ip link add vxlan1000 type vxlan id 1000 dev loopback0 local 10.209.36.1 learning
ip link set dev vxlan1000 mtu 9000
ip link set dev vxlan1000 netns cust1
ip netns exec cust1 bash
ip link set dev vxlan1000 up
ip link set dev vxlan1000 master br1000
Actually, changing learning attribute requires recreation, and this
change needs to manually reload the frr configuration.
The update mechanism in zebra about vxlan interface updates is
already put in place, but it does not work well with namespace
based vrf backend. The function zl3vni_from_svi() is then
modified to parse all the interfaces of each namespace.
Signed-off-by: Philippe Guibert <philippe.guibert@6wind.com>
Description:
Change is intended for fixing the following issues related to vrf route leaking:
Routes with special nexthops i.e. blackhole/sink routes when imported,
are not programmed into the FIB and corresponding nexthop is set as 'inactive',
nexthop interface as 'unknown'.
While importing/leaking routes between VRFs, in case of special nexthop(ipv4/ipv6)
once bgp announces route(s) to zebra, nexthop type is incorrectly set as
NEXTHOP_TYPE_IPV6_IFINDEX/NEXTHOP_TYPE_IFINDEX
i.e. directly connected even though we are not able to resolve through an interface.
This leads to nexthop_active_check marking nexthop !NEXTHOP_FLAG_ACTIVE.
Unable to find the active nexthop(s), route is not programmed into the FIB.
Whenever BGP leaks routes, set the correct nexthop type, so that route gets resolved
and correctly programmed into the FIB, in the imported vrf.
Co-authored-by: Kantesh Mundaragi <kmundaragi@vmware.com>
Signed-off-by: Iqra Siddiqui <imujeebsiddi@vmware.com>
Insist on the fact that zclient neighbor state flags are
mapped over netlink state flags. List all the defines
currently known on kernel, and create a netlink API to
convert netlink values to zclient values. The function is
simplified as it is a 1-1 match.
Signed-off-by: Philippe Guibert <philippe.guibert@6wind.com>
As NHRP expects some notification of neighboring entries on GRE
interface, when a new interface notification is encountered, the
exact neighbor state flag is found. Previously, the flag passed
to the upper layer was forced to NDM_STATE which is REACHABLE,
as can be seen on below trace:
2021/08/25 10:58:39 NHRP: [QQ0NK-1H449] Netlink: new-neigh 102.1.1.1 dev gre1 lladdr 10.125.0.2 nud 0x2 cache used 1 type 5
When passing the real value, NHRP received an other value like STALE.
2021/08/25 11:28:44 NHRP: [QQ0NK-1H449] Netlink: new-neigh 102.1.1.1 dev gre1 lladdr 10.125.0.2 nud 0x4 cache used 0 type 5
This flag is important for NHRP, as it permits to monitor the link
layer of NHRP entries.
Fixes: d603c0774e ("nhrp, zebra, lib: enforce usage of zapi_neigh_ip structure")
Signed-off-by: Philippe Guibert <philippe.guibert@6wind.com>
"[no] netns NAME" commands are part of the lib, but they are actually
zebra-only:
- they are using vrf_netns_handler_create and its description clearly
says that it "should be called from zebra only"
- vtysh sends these commands only to zebra
- only zebra outputs the netns related config
- zebra notifies other daemons about netns attachment
Signed-off-by: Igor Ryzhov <iryzhov@nfware.com>
There is a possibility that the same line can be matched as a command in
some node and its parent node. In this case, when reading the config,
this line is always executed as a command of the child node.
For example, with the following config:
```
router ospf
network 193.168.0.0/16 area 0
!
mpls ldp
discovery hello interval 111
!
```
Line `mpls ldp` is processed as command `mpls ldp-sync` inside the
`router ospf` node. This leads to a complete loss of `mpls ldp` node
configuration.
To eliminate this issue and all possible similar issues, let's print an
explicit "exit" at the end of every node config.
This commit also changes indentation for a couple of existing exit
commands so that all existing commands are on the same level as their
corresponding node-entering commands.
Fixes#9206.
Signed-off-by: Igor Ryzhov <iryzhov@nfware.com>
For some reason commit #ef524230a6baa decided
to remove enums and switch to uint16_t. Which
is not the right thing to do. Put it back
Signed-off-by: Donald Sharp <sharpd@nvidia.com>
Maybe with empty nexthop to call zebra_mpls_transit_lsp():
"no mpls lsp (16-1048575)".
So just remove this "gate_str" check. If without "gate" in command, "gtype" is
set to NEXTHOP_TYPE_BLACKHOLE for subsequent processing.
Signed-off-by: anlan_cs <anlan_cs@tom.com>
When NHRP registers to zebra to receive link layer events related to
gre interfaces, then it is interested in receiving also RTM_GETNEIGH
messages.
Fixes ("b3b751046495") nhrpd: link layer registration to notifications
Signed-off-by: Philippe Guibert <philippe.guibert@6wind.com>
In zebra_evpn_proc_remote_nh if we do not pass in a long
enough stream, the stream reads will fail. Ensure that
we have enough data.
Signed-off-by: Donald Sharp <sharpd@nvidia.com>
Handle TYPE_IFINDEX nexthops more consistently in a few places;
be more specific about a few integer return values that were
being treated as booleans.
Signed-off-by: Mark Stapp <mjs.ietf@gmail.com>
When calling rib_add_multipath_nhe ensure that we have
well aligned return codes that mean something so that
interersted parties can properly handle the situation.
Signed-off-by: Donald Sharp <sharpd@nvidia.com>
When receiving a route via zapi, if the route is rejected
there exists a code path where we would not free the corresponding
re created.
Signed-off-by: Donald Sharp <sharpd@nvidia.com>
The command `debug zebra kernel msgdump is netlink specific.
There is no point at all to allow this to be configed on non
netlink platforms.
Signed-off-by: Donald Sharp <sharpd@nvidia.com>
There were a bunch of places where we converted the
route node to a prefix string via srcdest_rnode2str when
we should have been using %pRN in zebra_rib.c. Just
convert over the ones we should to use it.
Signed-off-by: Donald Sharp <sharpd@nvidia.com>
When we are calling rib_process and the route_node
in question has no dest, there is no work to do here
at all. As such we should just return before
attempting to do any other work. This is just a tiny bit
of simplification being done.
Signed-off-by: Donald Sharp <sharpd@nvidia.com>
There exists a call path where the nhlfe_alloc can return NULL
for blackhole nexthops. In this case we were still trying
to save the nhlfe pointer causing a crash when we attempted
to add it to a self-contained list.
Signed-off-by: Donald Sharp <sharpd@nvidia.com>
