Ticket: None
Reviewed By: CCR-4692
Testing Done: smoke + scale tests
If called with close_fds=True the subprocess module will try to close every fd
from 3 to MAXFD before executing the specified command. This is done in Python
not even with a C-implementation which truly affecting performances.
This patch aims to better handle the file descriptor used by ifupdown2. Either
by closing them after use or by setting the close-on-exec flag for the file
descriptor, which causes the file descriptor to be automatically
(and atomically) closed when any of the exec-family functions succeed.
With the actual patch all tests are passing, I can't think of any future issue
but if any a possible future modification might be to use the parameter
'preexec_fn', which allows us to set function which will be executed in the
child process before executing the command line. We can always manually close
any remaining open file descriptors with something like:
>>> os.listdir('/proc/self/fd/')
['0', '1', '2', ‘3’, etc..]
>>> for fd in os.listdir('/proc/self/fd/')
>>> if int(fd) > 2:
>>> os.close(fd)
This patch is also totally re-organising the use of subprocesses. By removing
all subprocess code redundancy.
Ticket: CM-7678
Reviewed By: Roopa Prabhu
Testing Done: yes, with vlan aware/unaware bridge on dell-s3000-02
This patch enables bpdu filter and bpdu guard enabled for vxland device under
the bridge if these two attributes are unconfigured
Signed-off-by: Nikhil <nikhil@cumulusnetworks.com>
Ticket: CM-8736
Reviewed By: Roopa
Testing Done: Smoke tests + the ones from the ticket
By default ifupdown2 will adjust logical devices MTU
based on the physical interface they are running on top of.
set this flag to 0 to disable this behaviour
adjust_logical_dev_mtu=1
Ticket: CM-10281
Reviewed By: Roopa Prabhu
Testing Done: Tested with the configuration mentioned in the bug id
Due to bug CM-10188, defualt route via gateway is not installed in time.
It needs 2 sec delay, and fix to CM-10188 shall rectify this issue.
Signed-off-by: Nikhil <nikhil@cumulusnetworks.com>
Ticket: CM-7840
Reviewed By: Roopa Prabhu
Testing Done: yes, by installing ifupdown .deb file onto dell-s3000-02
This patch adds a new argument '--with-defaults' to 'ifquery'
when 'ifquery --with-defaults' is executed, running states of all interface
attributes are compared against respective configured attributes from
/etc/network/interfaces file, if configured. Otherwise, compared against
default attributes from policy file
This patch also: (1) fixes ifquery check failure for bridge-* stp attributes.
(2) removes vrf-default-route and vrf-cgroup attributes from ifupdown2 policy
and just have the vrf-helper attribute
Signed-off-by: Nikhil <nikhil@cumulusnetworks.com>
Ticket: CM-10437
Reviewed By: CCR-4480
Testing Done: Created a bond and made sure it's lacp_rate could be set fast or
slow wioth or without specifying the mode.
The current code assumes that for a bond to be in 802.3ad mode the "bond-mode"
parameter has to be specified in the list of bond attributes being set. Only
then will ifupdown2 allow the lacp_rate and lacp_bypass attributes to be set.
But since we have changed the default mode to be 802.3ad, the mode attribute no
longer needs to be specified for a bond to be in 802.3ad mode.
So, instead of doing a check for the configuration attributes to determine the
mode, the linkinfo cache is checked. And the cache is refreshed, just in case
the mode attribute has been specified and just previously been set. The order in
which the attributes are set is fixed for bonds, and the mode is always set
prior to the lacp_rate and lacp_bypass attributes.
Also, while testing this I noticed another assumption about the default value of
an attribute: min_links. A warning is displayed when min_links is not specified
or set to 0, and the mode is set to 802.3ad. Well, now that the default for
min_links is 1 it doesn't have to be specified, even if the mode is set to
802.3ad. So ifupdown2 checks the current state of min_links if it is not
specified as one of the attibutes. If it is 0, either by being specified as an
attribute or not being specified but having that value, and the mode is set to
802.3ad, a warning is displayed.
And there was this other little problem where if a bond already existed in the
linkcache it wouldn't be refreshed, even if the refresh=True parameter was
specified in the _bond_linkinfo_fill function.
Ticket: cleanup
Reviewed By:
Testing Done: Tested ifupdown sanity
This gets rid of some ugly previous flag handling which was
passed through modules. This creates a global instance of
flags that all addon modules and helper modules can use.
Signed-off-by: Roopa Prabhu <roopa@cumulusnetworks.com>
Ticket: CM-10292, CM-10282
Reviewed By: dsa, nikhil, julien
Testing Done: Tested dhcp config on an vrf slave
- Add support for policy module_globals variable
"vrf-exec-cmd-prefix". It is read into per module
self.vrf_exec_cmd_prefix variable
- If self.vrf_exec_cmd_prefix is present and interface is a
vrf slave, use ifupdown2 will call such command in vrf
context using:
"%s %s %s" %(<vrf-exec-cmd-prefix>, <vrfname>, <cmd>)
- This also fixes calling of dhcp refresh when a dhcp slave is
removed from a vrf
Signed-off-by: Roopa Prabhu <roopa@cumulusnetworks.com>
Ticket: CM-10302
Reviewed By: julien
Testing Done: Tested ifdown of an interface and make sure the addresses
are deleted
cache_invalidate can be called when an interface gets deleted
or on any other deletion event (this is because we dont really
know, what other things have been implicitly deleted by the kernel).
on ifdown, the cache can get invalidated. But if we dont
reset the cache flag, the fill cache functions can still think
the cache is valid and can return None for any queries.
This was happening to address queries, which resulted in
leaving behind configured addresses. because ifdown
thought that there were no running addresses on the interface.
Ticket: CM-10175
Reviewed By: Roopa Prabhu
Testing Done: yes, by installing ifupdown .deb onto dell-s3000-02
This patch is a git revert of commit 934c4c49c0e77289e7d56349c44d14ca2c307621.
In addition to that, order of _addr_filter() function call arguments are
changed to match the function definition.
Signed-off-by: Nikhil <nikhil@cumulusnetworks.com>
Ticket:
Reviewed By: trivial
Testing Done: tested address gets with refresh flag + ran ssim test testifupdown2.py
- also fixes refresh flag handling in cache _addr_fill function
- this api can be used to get addresses from the cache by first
refreshing the cache. So the caller gets fresh running addresses.
(its not used right now but came up during the need to re-apply
addresses for ipv6 vrf slaves since their addresses may disappear)
This reverts commit 934c4c49c0e77289e7d56349c44d14ca2c307621.
Ticket: CM-10175
Reviewed By: Roopa Prabhu
Testing Done: yes, by installing ifupdown .deb file onto dell-s3000-02
default addr fix for vrf check is deleting lo addrs accidentally
info: rtnetlink: setting link lo up
info: executing ip addr del ::1/128 dev lo
info: executing ip addr del 127.0.0.1/8 dev lo
info: eth0: running ops ...
Signed-off-by: Nikhil <nikhil@cumulusnetworks.com>
Ticket: CM-10188, CM-10061
Reviewed By: dsa, nikhil, julien
Testing Done: Tested static routes with vrf names for tables
This patch does the following:
- if a single vrf device is present in the config,
builds the vrf map by reading vrf interfaces from the kernel (with
existing link cache. Builds a shadow vrf only attribute cache)
- reads existing table map and adjusts it if required
- main change is the iproute2 map file on disk is updated
immediately on vrf creation, so that static routes used along with the
vrf slaves can use the vrf name for the table. This also helps dhclient dns
hook script which may use mgmt table name directly.
- cleans up default routes on down
Signed-off-by: Roopa Prabhu <roopa@cumulusnetworks.com>
Ticket: CM-9929
Reviewed By:
Testing Done: Tested vrf dhcp slave enslavement on ifreload
algo:
- when a vrf slave enslavement changes, before performing the
new master enslavement,
- check if vrf slave has dhcp configured
- if yes, release dhclient running on the vrf slave
Ticket: CM-9677
Reviewed By: nikhil
Testing Done: Tested with bond config file in CM-9677
This patch replaces the following attributes:
bond-ad-sys-priority with bond-ad-actor-sys-prio
bond-ad-sys-mac-addr with bond-ad-actor-system
The new attributes correspond to the new sysfs files below:
/sys/class/net/sidelink/bonding/ad_actor_sys_prio
/sys/class/net/sidelink/bonding/ad_actor_system
Old values will be accepted with a deprecated warning:
warning: attribute bond-ad-sys-priority is deprecated. Use bond-ad-actor-sys-prio instead
warning: attribute bond-ad-sys-mac-addr is deprecated. Use bond-ad-actor-system instead
Ticket: CM-9594
Reviewed By: Roopa Prabhu
Testing Done: Yes, by installing ifupdown2 deb onto cel-e1031-01
This patch fixes the failure to execute the command
'ip link set dev eth0 nomaster' when 'eth0' is associated with any vrf
This patch also include a review comment update for CCR-4210
Ticket: CM-8729
Reviewed By: trivial
Testing Done: Tested with a config with vlan-raw-device
'ip -o -d link show' introduced a new attribute between
'vlan and id'. This makes the move to json or netlink
even more necessary.
The fixes were done for the following format:
61: vlan100@swp1: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc
noqueue state UP mode DEFAULT group default \ link/ether
00:e0:ec:27:4e:b7 brd ff:ff:ff:ff:ff:ff promiscuity 0 \ vlan protocol
802.1Q id 100 <REORDER_HDR> addrgenmode eui64
The vrf device may not be up when ifup is executed on the
slaves. This commit makes sure:
- vrf slaves dont try to enslave themselves when vrf device is
not present
- And vrf master enslaves any missing slaves during ifup of vrf master
- Also make vrf device the link master, this will make sure
the vrf device brings the vrf slave links up. This is needed to work
around the ipv6 address flush issue
Closes: CM-9493
Review: dsa, nikhil, wkok (via git send-mail)
Signed-off-by: Roopa Prabhu <roopa@cumulusnetworks.com>
Ticket: CM-9494
Reviewed By: CCR-4184
Testing Done:
Also removed legacy 'fallback' options.
Now lacp bypass is enabled and disabled on a bond with:
bond-lacp-bypass-allow [0|1]
Ticket: CM-9520
Reviewed By: CCR-4152
Testing Done: verified service node configuration
Use "remote" attribute in iproute2 command to provision
service node address for service node based replication. Changes also
include allowing only one service node per vxlan device, so its user's
responsiblity to select one service node per vxlan device if there
are multiple nodes to distribute the load.
Ticket:CM-9106
Reviewed By: Roopa Prabhu and Dave Olson
Testing Done: Yes, by installing ifupdown and cumulus-tools deb on to cel-e1031-01
This patch adds per vrf default route: "ip route add table <tbid> unreachable default"
It also accepts a policy file with vrf-default-route [yes/no]
This also includes ifquery fixes for running and check.
Ticket: CM-6806
Reviewed By: CCR-4126
Testing Done:
Currently, when doing ifup of a bridge, the bridge is created
and ports are added to bridge before vlan_filtering is set on
the bridge. This causes extra churn on switchd which has to
configure the hardware one way and then tear it down and
reconfigure it again in the new way. For mlx, it causes even
more problems.
This patch moves the vlan_filtering setting of bridge to before
member ports are being added to the bridge, and it uses the new
iproute2 command for setting the attribute instead of through
sysfs.
Ticket: CM-9097
Reviewed By: Trivial
Testing Done: Tested on Chris' config
In 3.0, the bridge vlan show command does not print
VLAN ranges unless you use the "-c" option.
This patch modifies the bridge vlan show call in
iproute2.py to use "-c".
Ticket: CM-9182
Reviewed By: Trivial
Testing Done: ifup'd bond when bonding module was not yet loaded.
The bond support in ifupdown2 would check to see if the bonding module is
loaded when creating a bond. If it was not it would load the driver and return.
The correct operation is to load the driver and then continue to create the
bond.
Ticket: CM-9078
Reviewed By: Trivial
Testing Done: bond ifdown/ifup on s6000
This needs to be replaced with protodown of clag bond slaves
pre-enslave (TBD; coming soon).
Ticket: None
Reviewed By: CCR-4058
Testing Done: ifup'd interface with both dhcp-wait: "no" and dhcp-wait: "yes"
and not specified at all.
A previous patch implemented the nowait option for DHCP. This patch changes the
name of the option to "dhcp-wait" and makes the default, if nothing is specified
in the policy files, to be "yes", which means dhclient will be called without
the "-nw" option, causing it to wait for up to a minute for a response from the
DHCP server before continuing.
The format of the JSON in the policy file for this option was also changed so
that it conforms to the other ifupdown2 policy options. This format is now:
{
"dhcp": {
"defaults": { "dhcp-wait": "no" }
}
}
Also, the documented argument values are "yes" and "no". Any other values, will
be interpreted as "yes".
A subsequent patch in cl-basefiles will be made to include this fragment in
/var/lib/ifupdown2/policy.d/dhcp.json so that Cumulus Linux will default to
not waiting for DHCP to complete.
Ticket: None
Reviewed By: CCR-4058
Testing Done: ifup'd interface with both nowait=0 and nowait=1 and not specified
at all.
The Mellanox platform, as well as some others probably, has two management
interfaces: eth0 and eth1. The customer may plug a cable into either one of
these interfaces, and very rarely both of them. If only one cable is plugged in
and we don't know which one, then /etc/network/interfaces must be configured
by default to automatically bring up both interfaces using DHCP. But when an
interface does not have link, it stalls the boot process for 60 seconds while
dhclient times out.
This patch changes the default dhclient behavior to not wait for DHCP to
complete, by using the "-nw" option when calling dhclient. This means that
dhclient will immediately return and DHCP will complete in the background.
A module attribute has been added for the DHCP addon called "nowait", which
defaults to 1. If this attribute is set to 0, then dhclient will revert to its
previous behavior and delay up to a minute while DHCP completes. This attribute
can be specified in a policy file, e.g. /etc/network/ifupdown2/policy.d/dhcp.json,
with contents such as:
{
"dhcp" :
{
"nowait" : 0
}
}
This patch adds initial support for vrf in ifupdown2.
Example interfaces file section:
auto swp1.100
iface swp1.100
vrf blue
auto blue
iface blue
vrf-table 10
iproute2 vrf map is generated under:
/etc/iproute2/rt_tables.d/ifupdown2.vrf_map
this patch also adds prelimnary support for 'vrf-table auto'.
But this needs more work.
Signed-off-by: Roopa Prabhu <roopa@cumulusnetworks.com>
Ticket: CM-7756
Reviewed By: roopa
Testing Done: built powerpc and tested ifupdown2 as well as new tests
Once mstpctl-portbpdufilter or mstpctl-bpduguard is enabled for
an interface, removing the configuration in /etc/network/interfaces
does not toggle the mstpctl state back to no.
The root cause of this problem is that "ifreload" does not check default settings for MSTP configuration
for bridge ports and use a default when that setting is removed from the configuration.
This patch adds a check in the existing loop on attribute values.
If there is no configured value, we check to see if a default exists.
self._modinfo['attrs'][attrname]['default'] exists.
If it exists and it is different then the currently running value
we reset the attribute to its default. This is what a customer would expect if
they removed a configured value.
(also added test in cl-tests to check this functionality)
(cherry picked from commit 63d0f7082d44fedabe002aebbb658751dc655a46)
Ticket:
Reviewed By: CCR-3804
Testing Done: Tested regex parsing failures
This is mostly a cosmetic fix. we were failing with weird/unclear errors
on unable to parse regex expressions correctly.
This patch mainly adds the interface name to the message and plus adds
an info message showing the actual regex being used in searches.
example config:
{noformat}
auto br-roopa
iface br-roopa
bridge-vlan-aware yes
bridge-ports regex '(\\Aswp3\\Z|\\Aswp4\\Z)'
bridge-pvid 20
{noformat}
before the patch:
warning: br-roopa: error getting dependent interfaces (unbalanced
parenthesis)
after the patch (not pretty but easier to debug)
info: br-roopa: evaluating port expr '['regex', "'(", 'Aswp3', 'Z|',
'Aswp4', "Z)'"]'
warning: br-roopa: error getting dependent interfaces (br-roopa: error
searching regex ''(' in swp38 (unbalanced parenthesis))
(cherry picked from commit bcca6f753a25494666d53f1f2f3c855ffa41d7f0)
