| Age | Commit message (Collapse) | Author |
|---|
|
Fixes: 421cf856ef86db250a86be01437d0a668b463dcc ("bgpd: Cleanup multipath figuring out in bgp")
Signed-off-by: Donatas Abraitis <donatas@opensourcerouting.org>
|
|
Signed-off-by: Donald Sharp <sharpd@nvidia.com>
|
|
This function is no doing any work. Let's remove.
Signed-off-by: Donald Sharp <sharpd@nvidia.com>
|
|
Currently bgp multipath has these properties:
a) mp_info may or may not be on a single path, based
upon path perturbations in the past.
b) mp_info->count started counting at 0( meaning 1 ). As that the
bestpath path_info was never included in the count
c) The first mp_info in the list held the multipath data associated
with the multipath. As such if you were at any other node that data
was not filled in.
d) As such the mp_info's that are not first on the list basically
were just pointers to the corresponding bgp_path_info that was in
the multipath.
e) On bestpath calculation, a linklist(struct linklist *) of bgp_path_info's was
created.
f) This linklist was passed in to a comparison function that took the
old mpinfo list and compared it item by item to the linklist and
doing magic to figure out how to create a new mp_info list.
g) the old mp_info and the link list had to be memory managed and
freed up.
h) BGP_PATH_MULTIPATH is only set on non bestpath nodes in the
multipath.
This is really complicated. Let's change the algorithm to this:
a) When running bestpath, mark a bgp_path_info node that could be in the ecmp path as
BGP_PATH_MULTIPATH_NEW.
b) When running multipath, just walk the list of bgp_path_info's and if
it has BGP_PATH_MULTIPATH_NEW on it, decide if it is in BGP_MULTIPATH.
If we run out of space to put in the ecmp, clear the flag on the rest.
c) Clean up the counting of sometimes adding 1 to the mpath count.
d) Only allocate a mpath_info node for the bestpath. Clean it up
when done with it.
e) remove the unneeded list management associated with the linklist and
the mp_list.
This greatly simplifies multipath computation for bgp and reduces memory
load for large scale deployments.
2 full feeds in work_queue_run prior:
0 56367.471 1123 50193 493695 50362 493791 0 0 0 TE work_queue_run
BGP multipath info : 1941844 48 110780992 1941844 110780992
2 full feeds in work_queue_run after change:
1 52924.931 1296 40837 465968 41025 487390 0 0 1 TE work_queue_run
BGP multipath info : 970860 32 38836880 970866 38837120
Aproximately 4 seconds of saved cpu time for convergence and ~75 mb
smaller run time.
Signed-off-by: Donald Sharp <sharpd@nvidia.com>
|
|
The mpath data structure has data that is only relevant
for the first mpath in the list. It is not being used
anywhere else. Let's document that a bit more.
Signed-off-by: Donald Sharp <sharpd@nvidia.com>
|
|
Signed-off-by: Donald Sharp <sharpd@nvidia.com>
|
|
Under a setup where two BGP prefixes are available from multiple sources,
if one of the two prefixes is recursive over the other BGP prefix, then
it will not be considered as multipath. The below output shows the two
prefixes 192.0.2.24/32 and 192.0.2.21/32. The 192.0.2.[5,6,8] are the
known IP addresses visible from the IGP.
> # show bgp ipv4 192.0.2.24/32
> *>i 192.0.2.24/32 192.0.2.21 0 100 0 i
> * i 192.0.2.21 0 100 0 i
> * i 192.0.2.21 0 100 0 i
> # show bgp ipv4 192.0.2.21/32
> *>i 192.0.2.21/32 192.0.2.5 0 100 0 i
> *=i 192.0.2.6 0 100 0 i
> *=i 192.0.2.8 0 100 0 i
The bgp best selection algorithm refuses to consider the paths to
'192.0.2.24/32' as multipath, whereas the BGP paths which use the
BGP peer as nexthop are considered multipath.
> ... has the same nexthop as the bestpath, skip it ...
Previously, this condition has been added to prevent ZEBRA from
installing routes with same nexthop:
> Here you can see the two paths with nexthop 210.2.2.2
> superm-redxp-05# show ip route 2.23.24.192/28
> Routing entry for 2.23.24.192/28
> Known via "bgp", distance 20, metric 0, best
> Last update 00:32:12 ago
> * 210.2.2.2, via swp3
> * 210.2.0.2, via swp1
> * 210.2.1.2, via swp2
> * 210.2.2.2, via swp3
> [..]
But today, ZEBRA knows how to handle it. When receiving incoming routes,
nexthop groups are used. At creation, duplicated nexthops are
identified, and will not be installed. The below output illustrate the
duplicate paths to 172.16.0.200 received by an other peer.
> r1# show ip route 172.18.1.100 nexthop-group
> Routing entry for 172.18.1.100/32
> Known via "bgp", distance 200, metric 0, best
> Last update 00:03:03 ago
> Nexthop Group ID: 75757580
> 172.16.0.200 (recursive), weight 1
> * 172.31.0.3, via r1-eth1, label 16055, weight 1
> * 172.31.2.4, via r1-eth2, label 16055, weight 1
> * 172.31.0.3, via r1-eth1, label 16006, weight 1
> * 172.31.2.4, via r1-eth2, label 16006, weight 1
> * 172.31.8.7, via r1-eth4, label 16008, weight 1
> 172.16.0.200 (duplicate nexthop removed) (recursive), weight 1
> 172.31.0.3, via r1-eth1 (duplicate nexthop removed), label 16055, weight 1
> 172.31.2.4, via r1-eth2 (duplicate nexthop removed), label 16055, weight 1
> 172.31.0.3, via r1-eth1 (duplicate nexthop removed), label 16006, weight 1
> 172.31.2.4, via r1-eth2 (duplicate nexthop removed), label 16006, weight 1
> 172.31.8.7, via r1-eth4 (duplicate nexthop removed), label 16008, weight 1
Fix this by proposing to let ZEBRA handle this duplicate decision.
Fixes: 7dc9d4e4e360 ("bgp may add multiple path entries with the same nexthop")
Signed-off-by: Philippe Guibert <philippe.guibert@6wind.com>
|
|
Signed-off-by: Donatas Abraitis <donatas@opensourcerouting.org>
|
|
This is needed to implement and use larger bandwidths rather than limiting only
to theoretical 34Gbps max bandwidth.
Signed-off-by: Donatas Abraitis <donatas@opensourcerouting.org>
|
|
Move mp_nexthop_prefer_global boolean attribute to nh_flags. It does
not currently save memory because of the packing.
Signed-off-by: Louis Scalbert <louis.scalbert@6wind.com>
|
|
Let's use the natural data structure in bgp for the prefix display
instead of a bunch of places where we call a translator function.
The %pBD does this and actually ensures data is correct.
Also fix a few spots in bgp_zebra.c where the cast to a NULL
pointer causes the catcher functionality to not work and fix
the resulting crash that resulted.
Signed-off-by: Donald Sharp <sharpd@nvidia.com>
|
|
Even if some of the attributes in bgp_path_info_extra are
not used, their memory is still allocated every time. It
cause a waste of memory.
This commit code deletes all unnecessary attributes and
changes the optional attributes to pointer storage. Memory
will only be allocated when they are actually used. After
optimization, extra info related memory is reduced by about
half(~400B -> ~200B).
Signed-off-by: Valerian_He <1826906282@qq.com>
|
|
Done with a combination of regex'ing and banging my head against a wall.
Signed-off-by: David Lamparter <equinox@opensourcerouting.org>
|
|
This reverts commit 16aa1809e7c8caad37e8edd4e5aaac4f344bc7d3, reversing
changes made to f616e716089b16d9a678846282a6ac5c55e31a56.
|
|
Previous commits have introduced a new 8 bits nh_flag in the attr
struct that has increased the memory footprint.
Move the mp_nexthop_prefer_global boolean in the attr structure that
takes 8 bits to the new nh_flag in order to go back to the previous
memory utilization.
Signed-off-by: Louis Scalbert <louis.scalbert@6wind.com>
|
|
This seems just a mistake, drop `u` prefix.
Signed-off-by: Donatas Abraitis <donatas@opensourcerouting.org>
|
|
Memory allocations that fail crash the program. Checking
for NULL is not going to do anything.
Signed-off-by: Donald Sharp <sharpd@nvidia.com>
|
|
The maxpaths same_clusterlen value was a uint16_t
with a single bit being used. No other values are
being stored. Let's remove the bitfield and simplify
to a bool.
Signed-off-by: Donald Sharp <sharpd@nvidia.com>
|
|
Signed-off-by: Donatas Abraitis <donatas@opensourcerouting.org>
|
|
Signed-off-by: Donatas Abraitis <donatas@opensourcerouting.org>
|
|
Signed-off-by: Donatas Abraitis <donatas@opensourcerouting.org>
|
|
Signed-off-by: Donatas Abraitis <donatas.abraitis@gmail.com>
|
|
Signed-off-by: Donatas Abraitis <donatas.abraitis@gmail.com>
|
|
This is an extra work before moving attr->ecommunity to attra_extra struct.
Signed-off-by: Donatas Abraitis <donatas.abraitis@gmail.com>
|
|
Description:
Incorrect behavior during best path selection for the imported routes.
Imported routes are always treated as eBGP routes.
Change is intended for fixing the issues related to
bgp best path selection for leaked routes:
- FRR does ecmp for the imported routes,
even without any ecmp related config.
If the same prefix is imported from two different VRFs,
then we configure the route with ecmp even without
any ecmp related config.
- Locally imported routes are preferred over imported
eBGP routes.
If there is a local route and eBGP learned route
for the same prefix, if we import both the routes,
imported local route is selected as best path.
- Same route is imported from multiple tenant VRFs,
both imported routes point to the same VRF in nexthop.
- When the same route with same nexthop in two different VRFs
is imported from those two VRFs, route is not installed as ecmp,
even though we had ecmp config.
- During best path selection, while comparing the paths for imported routes,
we should correctly refer to the original route i.e. the ultimate path.
- When the same route is imported from multiple VRF,
use the correct VRF while installing in the FIB.
- When same route is imported from two different tenant VRFs,
while comparing bgp path info as part of bgp best path selection,
we should ideally also compare corresponding VRFs.
See-also: https://github.com/FRRouting/frr/files/7169555/FRR.and.Cisco.VRF-Lite.Behaviour.pdf
Co-authored-by: Santosh P K <sapk@vmware.com>
Co-authored-by: Kantesh Mundaragi <kmundaragi@vmware.com>
Signed-off-by: Iqra Siddiqui <imujeebsiddi@vmware.com>
|
|
When debugging issues for routes in multiple vrf's. It would
be extremely useful if the debug output had which vrf we
are acting on.
Signed-off-by: Donald Sharp <sharpd@nvidia.com>
|
|
This is to avoid breaking changes between existing deployments of
extended community for bandwidth encoding. By default FRR uses uint32
to encode bandwidth, which is not as the draft requires (IEEE floating-point).
This switch enables the required encoding per-peer.
Signed-off-by: Donatas Abraitis <donatas.abraitis@gmail.com>
|
|
Signed-off-by: Donatas Abraitis <donatas.abraitis@gmail.com>
|
|
Signed-off-by: Donald Sharp <sharpd@nvidia.com>
|
|
Should be just a couple non-development, non-test occurrences of this
function left now.
Signed-off-by: Quentin Young <qlyoung@qlyoung.net>
|
|
The multipath arrays are always being allocated, irrelevant
if we actually have multipath information for a prefix.
This is because the link bandwidth code was always adding the
data structure. We should not be allocated multipath information
unless we actually have multipath information
Signed-off-by: Donald Sharp <sharpd@nvidia.com>
|
|
Remove mid-string line breaks, cf. workflow doc:
.. [#tool_style_conflicts] For example, lines over 80 characters are allowed
for text strings to make it possible to search the code for them: please
see `Linux kernel style (breaking long lines and strings)
<https://www.kernel.org/doc/html/v4.10/process/coding-style.html#breaking-long-lines-and-strings>`_
and `Issue #1794 <https://github.com/FRRouting/frr/issues/1794>`_.
Scripted commit, idempotent to running:
```
python3 tools/stringmangle.py --unwrap `git ls-files | egrep '\.[ch]$'`
```
Signed-off-by: David Lamparter <equinox@diac24.net>
|
|
This is the bulk part extracted from "bgpd: Convert from `struct
bgp_node` to `struct bgp_dest`". It should not result in any functional
change.
Signed-off-by: Donald Sharp <sharpd@cumulusnetworks.com>
Signed-off-by: David Lamparter <equinox@opensourcerouting.org>
|
|
Support configurable options to control how link bandwidth is handled
by the receiver. The default behavior is to automatically honor the
link bandwidths received and use it to perform a weighted ECMP BUT only
if all paths in the multipath have associated link bandwidth; if one or
more paths do not have link bandwidth, normal ECMP is performed among
the multipaths. This behavior is as recommended by
https://tools.ietf.org/html/draft-ietf-idr-link-bandwidth.
The additional options available are to (a) completely ignore any link
bandwidth (i.e., weighted ECMP is effectively disabled), (b) skip paths
in the multipath which do not have link bandwidth and perform weighted
ECMP among the other paths (if at least some paths have the bandwidth)
or (c) use a default weight (value chosen is 1) for the paths which
do not have link bandwidth.
The command syntax is
bgp bestpath bandwidth <ignore|skip-missing|default-weight-for-missing>
Signed-off-by: Vivek Venkatraman <vivek@cumulusnetworks.com>
|
|
Implement the code to handle the other route-map options to generate
the link bandwidth, namely, to use the cumulative bandwidth or to
base this on the number of multipaths. In the latter case, a reference
bandwidth is internally chosen - the implementation uses a value of
1 Gbps.
These additional options mean that the prefix may need to be advertised
if there is a link bandwidth change, which is a new criteria. Define a
new path (change) flag to support this and implement the advertisement.
Signed-off-by: Vivek Venkatraman <vivek@cumulusnetworks.com>
Reviewed-by: Donald Sharp <sharpd@cumulusnetworks.com>
Reviewed-by: Don Slice <dslice@cumulusnetworks.com>
|
|
Perform weighted ECMP if the multipaths have link bandwidth. This involves
assigning weights to each of the next hops associated with the prefix based
on the link bandwidth of the corresponding path as a factor of the total
(cumulative) link bandwidth for the prefix. The weight values used are
between 1 and 100. Weights are assigned only if all paths in the multipath
have link bandwidth, otherwise any bandwidths are ignored and regular
ECMP is performed. This is as recommended in
https://tools.ietf.org/html/draft-ietf-idr-link-bandwidth
A subsequent commit will implement additional (user-configurable) behaviors.
Signed-off-by: Vivek Venkatraman <vivek@cumulusnetworks.com>
Reviewed-by: Donald Sharp <sharpd@cumulusnetworks.com>
Reviewed-by: Don Slice <dslice@cumulusnetworks.com>
|
|
During multipath update, track the cumulative link bandwidth
as well as update flags appropriately.
Signed-off-by: Vivek Venkatraman <vivek@cumulusnetworks.com>
Reviewed-by: Donald Sharp <sharpd@cumulusnetworks.com>
Reviewed-by: Don Slice <dslice@cumulusnetworks.com>
|
|
Introduce fields in the multipath structure for link bandwidth handling.
In the process, the mp_count field is changed to a uint16 as that is the
value set anyway.
Signed-off-by: Vivek Venkatraman <vivek@cumulusnetworks.com>
|
|
Add new function `bgp_node_get_prefix()` and modify
the bgp code base to use it.
This is prep work for the struct bgp_dest rework.
Signed-off-by: Donald Sharp <sharpd@cumulusnetworks.com>
|
|
Defer the grabbing of the prefix for as long as is possible.
This is a long term rework of how we access the `struct bgp_node`
to only use accessor functions.
Signed-off-by: Donald Sharp <sharpd@cumulusnetworks.com>
|
|
Most of the code uses macros, thus let's keep the code unified.
Signed-off-by: Donatas Abraitis <donatas.abraitis@gmail.com>
|
|
Signed-off-by: Donatas Abraitis <donatas.abraitis@gmail.com>
|
|
Signed-off-by: Quentin Young <qlyoung@cumulusnetworks.com>
|
|
yeah
Signed-off-by: Quentin Young <qlyoung@cumulusnetworks.com>
|
|
Since we don't set a value from the return of bgp_path_info_mpath_next
it is impossible for this function to do anything as such the if statement
is dead code as well.
Signed-off-by: Donald Sharp <sharpd@cumulusnetworks.com>
|
|
Clean up several SA warnings.
Signed-off-by: Mark Stapp <mjs@voltanet.io>
|
|
Signed-off-by: Quentin Young <qlyoung@cumulusnetworks.com>
|
|
community_free, lcommunity_free and ecommunity_free are similar type of functions. Most of the places, these three are called together. The signature of community_free is different from other two functions. Modified the community_free API signature to align with other two functions to avoid any confusion. There is no functionality impact with this and this is just to avoid any confusion.
Testing: manual testing and show commands
Signed-off-by: Sri Mohana Singamsetty msingamsetty@vmware.com
|
|
ri -> pi
bi -> bpi
info -> path
info -> rmap_path ( for routemap applications )
Signed-off-by: Donald Sharp <sharpd@cumulusnetworks.com>
|
|
Convert the binfo variable to path.
Signed-off-by: Donald Sharp <sharpd@cumulusnetworks.com>
|
