| Commit message (Collapse) | Author | Age | Files | Lines |
|---|
| | |
|
| |\ |
|
| | | |
|
| |/ |
|
| |
|
|
| |
It is a harmless event when BGP tries to delete already deleted routes.
|
| |
|
|
|
|
|
|
|
|
| |
When zebra changes routes it uses the netlink cmd socket, but these
also show up on the netlink listen socket. Use more kernel level socket
filtering so that any changes that happend because of command socket
are ignored.
Uses the netlink port id (pid) to identify the changes from the
command socket.
|
| | |
|
| |
|
|
|
|
|
|
| |
When interface comes down, need two passes over the rib to correctly
process rib updates. If interface has both ipv6 and ipv4 routes and
comes down, need to do it as a two step process. Not sure exactly why,
but suspect it has to do with the the work queue and the flagging of
activity to the head of the list.
|
| |
|
|
|
|
| |
Need to check if next hop is interface and if so then only mark
route active is interface is active. This addresses some of the problems
with static routes when interface goes down.
|
| |\ |
|
| | |
| |
| |
| |
| |
| | |
Check if nexthop is active when doing RIB update.
This should fix issues with static routes staying active after
link going down.
|
| | |
| |
| |
| |
| | |
Watchlink is no longer needed, and if it was used it would interfere
with other route management. Therefore don't build it.
|
| |/
|
|
|
| |
Watchlink won't work with the new kernel link-detect changes.
Therefore don't build it.
|
| |
|
|
|
|
| |
When link becomes unavailable, unselect any routes that use that link
as the next hop. This keeps routing protocols from propogating routes
these routes and greating black holes.
|
| | |
|
| | |
|
| |
|
|
|
| |
The watchlink process causes more problems than it fixes right
now. See bugs: 3092, 3145 etc
|
| |
|
|
|
|
| |
Revert "Disable watchlink from the build"
This reverts commit 09d3fdd15c9bf316127f13a9322fdbea13a2628c.
|
| |
|
|
| |
No longer build/start/install watchlink
|
| | |
|
| |\ |
|
| | |
| |
| |
| |
| |
| |
| | |
The filter rules were incorrect (jump to wrong offset), so the kernel
would not accept them.
Fixes: 2570
|
| | | |
|
| |\| |
|
| | |
| |
| |
| |
| |
| | |
The IFF_RUNNING flag is indeterminate if interface is not IFF_UP.
Basically the carrier state is driver dependent and wrong when
the interface is admin down.
|
| |/ |
|
| |
|
|
|
| |
The vyatta-interfaces script needs to update the linkstatus file,
so locking is needed to prevent concurrent read/update problems.
|
| |
|
|
|
|
| |
Use proper include files so watchlink can be built on 64 bit
platforms. Not strictly necessary yet, since vyatta isn't doing
64 bit builds, but some development machines run 64 bit OS.
|
| |
|
|
|
|
| |
the address provided will exclude all
matching addresses across all ethernet addresses.
|
| |\ |
|
| | |
| |
| |
| |
| |
| | |
Use socket filter to drop unwanted messages on the netlink listen socket.
This prevents problems where the listener socket buffer gets overrruns
with echos of the new route update that occurs when link changes.
|
| |/
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
> comments are now allowed after the # symbol after each entry
> address ranges are now supported in the matching if the user
enters a netmask value, then every address within the range is
exlcluded
> ip value (without a mask) is treated as a /32 for matching
> matching is against the masked off address value, the mask
is specified by the exclusion entry (and not by the configured
address value)
> sig_usr1 will reload the configuration and is required after
each modification of the exclude file.
The format of the file is the following:
[interface] [address|network] #comments
Only one entry per line is allowed
vifs are supported in the interface line (i.e. eth0.1w)
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
[interface]
[interface] [ip/mask]
It should also accept a plain IP--but I haven't tested this yet.
Also added syslogging at the info level on interface up/down events.
This change provides support for HA to work with interface cable event. HA and VRRP will need to
populate the /var/linkstatus/exclude file with the VIPs under their management.
In addition to circumvent an additional problem in HA there needs to be an additional address on
the monitored interface, otherwise HA will admin down the interface when the cable is pulled
|
| | |
|
| | |
|
| |
|
|
|
|
| |
error msgs that occur under large routing tables when layer
2 event occurs.
|
| |\
| |
| |
| |
| |
| | |
Conflicts:
debian/changelog
|
| | | |
|
| |/ |
|
| | |
|
| |
|
|
| |
IFF_RUNNING|IFF_UP event.
|
| |
|
|
| |
- 'vtysh -c "sh run" | grep bgp' gives an accurate 32-bit AS number now
|
| |
|
|
| |
is lowered
|
| |\ |
|
| | |
| |
| |
| | |
- 32 bit asns show allright now
|
| |/
|
|
| |
Glendale is licensed under GPL version 2.
|
| |
|
|
| |
Re-enable handling of RTM_DELLINK.
|
| | |
|
| | |
|
| |
|
|
|
|
| |
will also need to modify
kernel proc settings
|
