sliver-openvswitch.git
12 years agoovs-xapi-sync: Make pychecker-able.
Ben Pfaff [Wed, 21 Sep 2011 21:56:55 +0000 (14:56 -0700)]
ovs-xapi-sync: Make pychecker-able.

pychecker imports the code that it checks, which means that code at top
level runs, so "ovs-xapi-sync" failed to import unless the user had write
access to /var/log/openvswitch.

12 years agoovs-bugtool: Fix --unlimited option.
Daniel Roman [Wed, 21 Sep 2011 21:44:15 +0000 (14:44 -0700)]
ovs-bugtool: Fix --unlimited option.

Bug #5443.

12 years agoRemove netdev_find_dev_by_in4
Simon Horman [Thu, 22 Sep 2011 12:24:14 +0000 (21:24 +0900)]
Remove netdev_find_dev_by_in4

netdev_find_dev_by_in4() appears to no longer be used and thus
can be removed. This also allows netdev_enumerate(), the
enumerate member of struct netdev_class and netdev_linux_enumerate()
to be removed.

I noticed this as netdev_linux_enumerate() makes use of if_nameindex()
and if_freenameindex() which are not available when compiling using
the Android NDK r6b (Android API level 13).

12 years agoDo not include net/ethernet.h and linux/if_tunnel.h
Simon Horman [Thu, 22 Sep 2011 12:24:13 +0000 (21:24 +0900)]
Do not include net/ethernet.h and linux/if_tunnel.h

net/ethernet.h and linux/if_tunnel.h do not appear to be needed
on lib/netdev-linux.c.

I noticed this while trying to build on the Android NDK r6b (Android API
level 13) as these headers are not present there.

12 years agoAdd fallback definition of SIG_ATOMIC_MAX
Simon Horman [Thu, 22 Sep 2011 12:24:12 +0000 (21:24 +0900)]
Add fallback definition of SIG_ATOMIC_MAX

Android appears to lack SIG_ATOMIC_MAX which is only
used in fatal-signal.c.

Observed when compiling using the Android NDK r6b (Android API level 13).

Patch based on a suggestion by Ben Pfaff

12 years agoInclude <sys/time.h> in socket-util.h
Simon Horman [Thu, 22 Sep 2011 12:24:11 +0000 (21:24 +0900)]
Include <sys/time.h> in socket-util.h

sys/time.h appears to be required for a definition of timespec on Android.

Observed when compiling using the Android NDK r6b (Android API level 13).

12 years agoInclude <sys/socket.h> in socket-util.h
Simon Horman [Thu, 22 Sep 2011 12:24:10 +0000 (21:24 +0900)]
Include <sys/socket.h> in socket-util.h

This appears to be necessary to get a definition of socket_t
when compiling using the Android NDK r6b (Android API level 13).

12 years agonetlink-socket: Fix typo in comment.
Ben Pfaff [Thu, 22 Sep 2011 03:49:32 +0000 (20:49 -0700)]
netlink-socket: Fix typo in comment.

12 years agodatapath: IFF_BRIDGE_PORT is backported by Centos 5.6.
Jesse Gross [Wed, 21 Sep 2011 19:41:52 +0000 (12:41 -0700)]
datapath: IFF_BRIDGE_PORT is backported by Centos 5.6.

Some versions of Centos 5.6 backport the flag IFF_BRIDGE_PORT
without the associated rx_handler changes, so this changes to
use a version check since we really don't care about the actual
symbol.

Reported-by: Srinivasan Ramasubramanian <vrsrini@gmail.com>
Signed-off-by: Jesse Gross <jesse@nicira.com>
12 years agodatapath: Cleanup actions.c:do_output().
Jesse Gross [Tue, 20 Sep 2011 23:44:46 +0000 (16:44 -0700)]
datapath: Cleanup actions.c:do_output().

The code for outputting a packet can be simplified a little and
also modernized.  There is no functional change.

Signed-off-by: Jesse Gross <jesse@nicira.com>
Acked-by: Ben Pfaff <blp@nicira.com>
12 years agodatapath: Send to userspace errors shouldn't halt processing.
Jesse Gross [Tue, 20 Sep 2011 23:29:38 +0000 (16:29 -0700)]
datapath: Send to userspace errors shouldn't halt processing.

If we encounter an error when sending a packet to userspace due to
an explicit action we stop processing further actions.  This makes
sense for things like push vlan, where to continue means outputting
an incorrect packet.  However, sending to userspace is more akin
to outputting to a port, which does not halt further processing.
For consistency, ignore errors in this case as well.

Signed-off-by: Jesse Gross <jesse@nicira.com>
Acked-by: Ben Pfaff <blp@nicira.com>
12 years agodatapath: Correctly validate vport attributes on old kernels.
Jesse Gross [Tue, 20 Sep 2011 17:31:29 +0000 (10:31 -0700)]
datapath: Correctly validate vport attributes on old kernels.

The vport policy for OVS_VPORT_ATTR_PORT_NO and OVS_VPORT_ATTR_TYPE
are present only in the section for newer kernels.  This means that
on older kernels the length of these attributes are never checked
anywhere but we go ahead and read from them anyways.

Signed-off-by: Jesse Gross <jesse@nicira.com>
Acked-by: Ben Pfaff <blp@nicira.com>
12 years agodatapath: Remove check for shared skbs.
Jesse Gross [Tue, 20 Sep 2011 21:08:57 +0000 (14:08 -0700)]
datapath: Remove check for shared skbs.

We never allow shared skbs to be present inside of the OVS datapath
but the presence of a check in the core makes this less clear.  Since
the check is very old and no longer relevant, drop it.

Signed-off-by: Jesse Gross <jesse@nicira.com>
Acked-by: Ben Pfaff <blp@nicira.com>
12 years agoovs-vsctl: Fix typo in documentation.
Ben Pfaff [Mon, 19 Sep 2011 20:29:05 +0000 (13:29 -0700)]
ovs-vsctl: Fix typo in documentation.

I don't know what was really supposed to go here.

Documentation #7371.
Reported-by: Reid Price <reid@nicira.com>
12 years agodebian: Correct path to ovs-controller in init script.
Ben Pfaff [Tue, 20 Sep 2011 16:38:33 +0000 (09:38 -0700)]
debian: Correct path to ovs-controller in init script.

Reported-by: George Shuklin <amarao@desunote.ru>
Bug-report: http://bugs.debian.org/642206

12 years agodatapath: Fully initialize datapath before local port.
Jesse Gross [Thu, 15 Sep 2011 23:41:36 +0000 (16:41 -0700)]
datapath: Fully initialize datapath before local port.

It's possible to start receiving packets on a datapath as soon as
the internal device is created.  It's therefore important that the
datapath be fully initialized before this, which it currently isn't.
In particular, the fact that dp->stats_percpu is not yet set is
potentially fatal.  In addition, if allocation of the Netlink response
failed it would leak the percpu memory.  This fixes both problems.

Found by code inspection, in practice the datapath is probably always
done initializing before someone can send a packet on it.

Signed-off-by: Jesse Gross <jesse@nicira.com>
Acked-by: Ben Pfaff <blp@nicira.com>
12 years agodatapath: Correctly set error code in queue_userspace_packets().
Jesse Gross [Mon, 19 Sep 2011 23:23:25 +0000 (16:23 -0700)]
datapath: Correctly set error code in queue_userspace_packets().

In a few places in queue_userspace_packets() when we encounter an
error, we don't actually set the 'err' variable.  Although we
free the packets we don't correctly account for these packets as
being lost.

Signed-off-by: Jesse Gross <jesse@nicira.com>
Acked-by: Ben Pfaff <blp@nicira.com>
12 years agodatapath-protocol: vport_stats types are unsigned.
Jesse Gross [Mon, 19 Sep 2011 23:11:27 +0000 (16:11 -0700)]
datapath-protocol: vport_stats types are unsigned.

The 'u' in uint64_t apparently got clipped off of the tx_dropped
member of struct vport_stats in between review and push, incorrectly
making this a signed type.

CC: Pravin Shelar <pshelar@nicira.com>
Signed-off-by: Jesse Gross <jesse@nicira.com>
Acked-by: Pravin Shelar <pshelar@nicira.com>
12 years agosflow: Use ODP-port number to lookup ifindex.
Pravin Shelar [Mon, 19 Sep 2011 21:55:31 +0000 (14:55 -0700)]
sflow: Use ODP-port number to lookup ifindex.

   dpif_sflow_odp_port_to_ifindex() expects odp_port number as sFlow
maintains ODP-port to ifindex mapping.

12 years agoovs-ofctl: Clarify in_port in manpage.
Ben Pfaff [Mon, 19 Sep 2011 20:22:16 +0000 (13:22 -0700)]
ovs-ofctl: Clarify in_port in manpage.

Suggestion #7370.
Suggested-by: Reid Price <reid@nicira.com>
12 years agonicira-ext: Clarify comment.
Ben Pfaff [Mon, 19 Sep 2011 20:04:34 +0000 (13:04 -0700)]
nicira-ext: Clarify comment.

I noticed a couple of typos and inaccuracies here while reviewing Jean's
changes to it for OXM at https://www.opennetworking.org/bugs/browse/EXT-1

12 years agodaemon.py: Silence return warning.
Ethan Jackson [Fri, 16 Sep 2011 23:46:18 +0000 (16:46 -0700)]
daemon.py: Silence return warning.

Pychecker complains about __read_pidfile() having too may returns.
I personally think the function is fine, but it's easy enough to
reduce them.

python/ovs/daemon.py:395: Function (__read_pidfile) has too many
returns (12)

12 years agodaemon.py: Don't shadow built-in 'file' variable.
Ethan Jackson [Fri, 16 Sep 2011 23:38:39 +0000 (16:38 -0700)]
daemon.py: Don't shadow built-in 'file' variable.

Pychecker considers it bad style.

12 years agodaemon.py: Whitespace cleanup.
Ethan Jackson [Fri, 16 Sep 2011 23:03:31 +0000 (16:03 -0700)]
daemon.py: Whitespace cleanup.

The python style guide requires two newlines between top level
definitions.  This patch also removes some trailing whitespace.

12 years agotests: Cleanup test-daemon.py style.
Ethan Jackson [Fri, 16 Sep 2011 22:48:33 +0000 (15:48 -0700)]
tests: Cleanup test-daemon.py style.

By convention, unused arguments should be named "_" and top level
definitions should be separated by two spaces.

12 years agocfm: Update cfm_remote_mpids documentation.
Ethan Jackson [Fri, 16 Sep 2011 18:31:55 +0000 (11:31 -0700)]
cfm: Update cfm_remote_mpids documentation.

12 years agodpif-linux: Handle nl_lookup_genl_mcgroup() failures.
Ethan Jackson [Tue, 13 Sep 2011 01:57:50 +0000 (18:57 -0700)]
dpif-linux: Handle nl_lookup_genl_mcgroup() failures.

The nl_lookup_genl_mcgroup() function can fail on older kernels
which do not support the required netlink interface.  Before this
patch, dpif-linux would refuse to create a datapath when this
happened.  With this patch, it attempts to use a workaround.  If
the workaround fails it simply disables the affected features
without completely disabling the dpif.

12 years agodpif-linux: Open dpif despite notifier failures.
Ethan Jackson [Wed, 14 Sep 2011 18:26:21 +0000 (11:26 -0700)]
dpif-linux: Open dpif despite notifier failures.

Before this patch, if dpif-linux failed to register a notifier it
would give up opening the datapath entirely.  This seems draconian
as a dpif can still perform the majority of its intended
functionality without vport notifications.

12 years agodatapath: Hardcode vport multicast group ID on older kernels.
Ethan Jackson [Mon, 12 Sep 2011 21:09:34 +0000 (14:09 -0700)]
datapath: Hardcode vport multicast group ID on older kernels.

Older kernels do not advertise the multicast groups of families
when requested by userspace.  As a workaround, this patch hardcodes
the multicast group ID of the ovs_vport family on these kernels.
Userspace will be able to fall back to this hardcoded value if the
standard mechanism is unavailable.

Signed-off-by: Ethan Jackson <ethan@nicira.com>
Acked-by: Jesse Gross <jesse@nicira.com>
12 years agonotifiers: Create and destroy nln_notifiers.
Ethan Jackson [Thu, 15 Sep 2011 18:21:23 +0000 (11:21 -0700)]
notifiers: Create and destroy nln_notifiers.

This patch changes the interface of netlink-notifier and
rtnetlink-link.  Now nln_notifiers are allocated and destroyed by
the module instead of passed in by callers.  This allows the
definition of nln_notifier to be hidden, and generally cleans up
the code.

12 years agonotifiers: Rename run and wait functions.
Ethan Jackson [Thu, 15 Sep 2011 18:23:08 +0000 (11:23 -0700)]
notifiers: Rename run and wait functions.

It makes more sense to call nln_notifier_run() and
nln_notifier_wait() simply nln_run() and nln_wait() since they
don't operate on notifiers but the entire nln object.  This patch
changes the nln and the rtnetlink-link modules to the new
convention.

12 years agoofproto-dpif: Fix behavior when a subset of VLANs is trunked.
Ben Pfaff [Wed, 24 Aug 2011 22:27:14 +0000 (15:27 -0700)]
ofproto-dpif: Fix behavior when a subset of VLANs is trunked.

Reported-by: Philippe Jung <phil.jung@free.fr>
12 years agodatapath: Always use generic stats for devices (vports)
Pravin Shelar [Fri, 16 Sep 2011 02:36:17 +0000 (19:36 -0700)]
datapath:  Always use generic stats for devices (vports)

    Currently ovs is using device stats for Linux devices and count them
itself in other situations. This leads to overlap with hardware stats,
inconsistencies, etc. It's much better to just always count the packets
flowing through the switch and let userspace do any merging that it wants.

Following patch removes vport->get_stats() interface. vport-stat is changed
to use new `struct ovs_vport_stat` rather than rtnl_link_stats64.
Definitions of rtnl_link_stats64 is removed from OVS.  dipf_port->stat is also
removed as aggregate stats are only available at netdev layer.

Signed-off-by: Pravin B Shelar <pshelar@nicira.com>
Acked-by: Jesse Gross <jesse@nicira.com>
12 years agoSet MTU in userspace rather than kernel.
Justin Pettit [Tue, 13 Sep 2011 05:13:30 +0000 (22:13 -0700)]
Set MTU in userspace rather than kernel.

Currently the kernel automatically sets the MTU of any internal
interfaces to the minimum of all attached interfaces because the Linux
bridge does this.  Userspace can do this with more knowledge and
flexibility.

Feature #7323

Signed-off-by: Justin Pettit <jpettit@nicira.com>
Acked-by: Jesse Gross <jesse@nicira.com>
12 years agoAUTHORS: Add Tyler Coumbes <coumbes@gmail.com>.
Ben Pfaff [Thu, 15 Sep 2011 22:58:59 +0000 (15:58 -0700)]
AUTHORS: Add Tyler Coumbes <coumbes@gmail.com>.

12 years agoovs-brcompatd: Delete ports when netdevs on fake bridges disappear.
Ben Pfaff [Thu, 15 Sep 2011 22:55:45 +0000 (15:55 -0700)]
ovs-brcompatd: Delete ports when netdevs on fake bridges disappear.

Until now, when a network device disappeared, netdev_changed_cb() passed
the name of the bridge that contained the network device to ovs-vsctl as
part of the "del-port" command.  However, when the network device was
actually a "fake bridge", it would pass the name of the real bridge, which
ovs-vsctl rejected as wrong (expecting the name of the fake bridge) and
not remove the port.

This fixes the problem by dropping the bridge name, which is simpler than
trying to get the name of the fake bridge in this case.

Reported-by: Tyler Coumbes <coumbes@gmail.com>
Tested-by: Tyler Coumbes <coumbes@gmail.com>
12 years agocfm: Delete spurious blank line.
Ben Pfaff [Thu, 15 Sep 2011 22:40:39 +0000 (15:40 -0700)]
cfm: Delete spurious blank line.

12 years agoovs-vsctl: Improve usage message.
Ben Pfaff [Thu, 15 Sep 2011 20:30:18 +0000 (13:30 -0700)]
ovs-vsctl: Improve usage message.

Bug #7332.
Reported-by: Gordon Good <ggood@nicira.com>
12 years agoofproto-dpif: Revalidate on cfm and lacp addition.
Ethan Jackson [Thu, 15 Sep 2011 22:27:22 +0000 (15:27 -0700)]
ofproto-dpif: Revalidate on cfm and lacp addition.

The introduction of cfm or lacp objects to ofproto, requires the
removal of all flows which originate from the newly "special"
in_port.

12 years agobugtool: Remove "ovs-bugtool" upon "make clean", to fix "make distcheck".
Ben Pfaff [Thu, 25 Aug 2011 17:20:28 +0000 (10:20 -0700)]
bugtool: Remove "ovs-bugtool" upon "make clean", to fix "make distcheck".

12 years agobugtool: Uninstall plugins on "make uninstall".
Ben Pfaff [Thu, 25 Aug 2011 17:20:01 +0000 (10:20 -0700)]
bugtool: Uninstall plugins on "make uninstall".

This works toward making "make distcheck" succeed.

12 years agoMark "uninstall-local" targets phony.
Ben Pfaff [Thu, 25 Aug 2011 17:19:23 +0000 (10:19 -0700)]
Mark "uninstall-local" targets phony.

12 years agorhel: Fix "make distcheck" failure due to regenerating spec files.
Ben Pfaff [Thu, 25 Aug 2011 17:18:47 +0000 (10:18 -0700)]
rhel: Fix "make distcheck" failure due to regenerating spec files.

We want to regenerate the RPM spec files whenever the version number
changes, hence the dependency on config.status.  But that means that we
try to modify the spec files even when the version number doesn't change,
which cause "make distcheck" to fail because it write-protects the source
directory.  So this commit instead just "touch"es the spec files when
they don't really change, which still works OK with a write-protected
source directory.

12 years agoofproto: Remove dummy 'in_port' variable.
Ben Pfaff [Thu, 8 Sep 2011 18:21:45 +0000 (11:21 -0700)]
ofproto: Remove dummy 'in_port' variable.

Seems like a very small cleanup.

12 years agoofp-util: Remove obsolete build assertion.
Ben Pfaff [Wed, 7 Sep 2011 17:12:13 +0000 (10:12 -0700)]
ofp-util: Remove obsolete build assertion.

Commit d1e9b9bf3 "nicira-ext: Renumber NXT_FLOW_MOD_TABLE_ID" eliminated
the need for the NXT_SET_FLOW_FORMAT and NXT_FLOW_MOD_TABLE_ID commands to
have different sizes, so asserting that they are different isn't useful
anymore (although it is still correct and always will be).

12 years agoodp-util: Drop pointless "if" statement.
Ben Pfaff [Tue, 6 Sep 2011 23:48:39 +0000 (16:48 -0700)]
odp-util: Drop pointless "if" statement.

12 years agonetdev: Allow get_mtu and set_mtu provider functions to be null.
Ben Pfaff [Thu, 15 Sep 2011 17:41:15 +0000 (10:41 -0700)]
netdev: Allow get_mtu and set_mtu provider functions to be null.

Most netdev provider functions are allowed to be null if the implementation
does not support this feature.  This commit adds this feature for get_mtu
and set_mtu, and changes netdev-vport to take advantage of it.

Also, changes netdev_get_mtu() to report an MTU of 0 on error, instead of
leaving the MTU indeterminate.

12 years agodatapath: Fix tunnel lookup
Pravin Shelar [Thu, 15 Sep 2011 00:39:43 +0000 (17:39 -0700)]
datapath: Fix tunnel lookup

        Attached patch fixes tunnel lookup to do correct port comparison.
This bug is introduced by commit 3544358aa5960b148bc31435a0062e9392530ec2

Signed-off-by: Pravin B Shelar <pshelar@nicira.com>
Acked-by: Jesse Gross <jesse@nicira.com>
12 years agodatapath: Set vport in skb when executed from userspace.
Jesse Gross [Mon, 1 Aug 2011 07:35:20 +0000 (00:35 -0700)]
datapath: Set vport in skb when executed from userspace.

Currently, the OVS_CB(skb)->vport member is never initialized for
packets coming from userspace.  This means that they can never be
sampled by sFlow and generally violates our principle that userspace
packets should be made to look the same as others.

Signed-off-by: Jesse Gross <jesse@nicira.com>
Acked-by: Pravin Shelar <pshelar@nicira.com>
12 years agoman: pic failed to run during manpage-check
Ethan Jackson [Tue, 13 Sep 2011 20:30:30 +0000 (13:30 -0700)]
man: pic failed to run during manpage-check

This patch fixes the following warnings on my system:

vswitchd/ovs-vswitchd.conf.db.5:62: warning: macro `PS' not defined
vswitchd/ovs-vswitchd.conf.db.5:138: warning: macro `PE' not defined

12 years agolib: TYPE_IS_SIGNED macro generates compiler warnings.
Ethan Jackson [Tue, 13 Sep 2011 20:27:38 +0000 (13:27 -0700)]
lib: TYPE_IS_SIGNED macro generates compiler warnings.

The TYPE_IS_SIGNED macro does a less than zero comparision with an
unsigned type which can cause compiler warnings like the following:

lib/tag.c:100:9: error: comparison of unsigned expression < 0 is
always false [-Werror=type-limits]

12 years agosocket-util: inet_parse_passive() had incorrect argument type.
Ethan Jackson [Tue, 13 Sep 2011 20:28:49 +0000 (13:28 -0700)]
socket-util:  inet_parse_passive() had incorrect argument type.

This patch fixes the following compiler warning:

lib/socket-util.c:621:5: error: comparison is always false due to
limited range of data type [-Werror=type-limits]

12 years agolib: Suppress comparison warnings in ovsdb libraries.
Ethan Jackson [Tue, 13 Sep 2011 20:15:48 +0000 (13:15 -0700)]
lib: Suppress comparison warnings in ovsdb libraries.

This patch fixes compiler warnings like the following:

./lib/ovsdb-types.h:171:5: error: comparison of unsigned expression
>= 0 is always true [-Werror=type-limits]

12 years agoofproto-dpif: Optimize flow revalidation for MAC learning.
Ben Pfaff [Mon, 12 Sep 2011 23:48:07 +0000 (16:48 -0700)]
ofproto-dpif: Optimize flow revalidation for MAC learning.

Without this commit, every NXAST_LEARN action that adds a flow causes every
facet to be revalidated.  With this commit, as long as the "Usage Advice"
in the large comment on struct nx_action_learn in nicira-ext.h is followed,
this no longer happens.

12 years agoofproto-dpif: Introduce an enum for the number of tables.
Ben Pfaff [Mon, 12 Sep 2011 23:40:03 +0000 (16:40 -0700)]
ofproto-dpif: Introduce an enum for the number of tables.

It seems reasonable that someone might want to reduce this number, so
make it an enum to simplify that.

If someone does reduce the number then rule_dpif_lookup() needs to validate
the table_id, so add code to do that too.

12 years agoflow: New function flow_wildcards_is_catchall().
Ben Pfaff [Mon, 12 Sep 2011 23:38:52 +0000 (16:38 -0700)]
flow: New function flow_wildcards_is_catchall().

This will be used in an upcoming commit.

12 years agoclassifier: Move zero_wildcards() to flow.c as public flow_zero_wildcards().
Ben Pfaff [Fri, 19 Aug 2011 16:39:16 +0000 (09:39 -0700)]
classifier: Move zero_wildcards() to flow.c as public flow_zero_wildcards().

This function will soon be used elsewhere.  As it doesn't inherently have
anything to with the classifier, move it to flow.c.

12 years agoImplement new "learn" action.
Ben Pfaff [Mon, 12 Sep 2011 23:19:57 +0000 (16:19 -0700)]
Implement new "learn" action.

There are a few loose ends here.  First, learning actions cause too much
flow revalidation.  Upcoming commits will fix that problem.  The following
additional issues have not yet been addressed:

    * Resource limits: nothing yet limits the maximum number of flows that
      can be learned.  It is possible to exhaust all system memory.

    * Age reporting: there is no way to find out how soon a learned table
      entry is due to be evicted.

To try this action out, here's a recipe for a very simple-minded MAC
learning switch.  It uses a 10-second MAC expiration time to make it easier
to see what's going on:

ovs-vsctl del-controller br0
ovs-ofctl del-flows br0
ovs-ofctl add-flow br0 "table=0 actions=learn(table=1, hard_timeout=10, \
NXM_OF_VLAN_TCI[0..11], NXM_OF_ETH_DST[]=NXM_OF_ETH_SRC[], \
output:NXM_OF_IN_PORT[]), resubmit(,1)"
ovs-ofctl add-flow br0 "table=1 priority=0 actions=flood"

You can then dump the MAC learning table with:

ovs-ofctl dump-flows br0 table=1

12 years agoofproto: Reinterpret meaning of OpenFlow hard timeouts with OFPFC_MODIFY.
Ben Pfaff [Fri, 19 Aug 2011 17:33:09 +0000 (10:33 -0700)]
ofproto: Reinterpret meaning of OpenFlow hard timeouts with OFPFC_MODIFY.

I finally found a good use for hard timeouts in OpenFlow, but they require
a slight reinterpretation of the meaning of hard timeouts.  Until now, a
hard timeout meant that a flow would be removed the specified number of
seconds after a flow was created.  Intervening modifications with
OFPFC_MODIFY(_STRICT) had no effect on the hard timeout; the flow would
still be deleted the specified number of seconds after its original
creation.

This commit changes the effect of OFPFC_MODIFY(_STRICT).  Now, modifying
a flow resets its hard timeout counter.  A flow will time out the specified
number of seconds after creation or after the last time it is modified,
whichever comes later.

12 years agotest-openflowd: Allow specifying port type on --ports option.
Ben Pfaff [Thu, 18 Aug 2011 18:20:12 +0000 (11:20 -0700)]
test-openflowd: Allow specifying port type on --ports option.

This allows a command like "test-openflowd --enable-dummy dummy@br0
--ports=dummy@eth0,dummy@eth1,dummy@eth2" to create a dummy datapath with
a number of dummy ports.  This is more useful for testing than a dummy
datapath with just an internal port, since output to "flood" and "normal"
has less pathological results.

12 years agodpif-netdev: Also allow "dummy" netdevs in a dpif-netdev.
Ben Pfaff [Thu, 18 Aug 2011 18:17:29 +0000 (11:17 -0700)]
dpif-netdev: Also allow "dummy" netdevs in a dpif-netdev.

I've always intended this to work, but either I never tested it or the
support rotted.

This will soon be used in some tests that I will add.

12 years agoofproto-dpif: Add -generate option to ofproto/trace command.
Ben Pfaff [Thu, 8 Sep 2011 21:32:13 +0000 (14:32 -0700)]
ofproto-dpif: Add -generate option to ofproto/trace command.

12 years agometa-flow: New library for working with fields by id.
Ben Pfaff [Mon, 12 Sep 2011 19:11:50 +0000 (12:11 -0700)]
meta-flow: New library for working with fields by id.

OVS already has a fairly good set of functions for working with fields that
are known at compile time, but support for working with fields that are
known only at runtime is fairly limited (and fairly unneeded).  However,
with NXM identifiers becoming more and more widely used throughout Nicira
extensions, it's becoming corresponding more and more common to need to
refer to fields at runtime.  This new library represents a first attempt
at a systematic approach for doing so.

12 years agoofproto: Mark some parameters 'const'.
Ben Pfaff [Wed, 17 Aug 2011 22:48:50 +0000 (15:48 -0700)]
ofproto: Mark some parameters 'const'.

12 years agoofproto: Add 'ofproto' parameter to most flow_mod functions.
Ben Pfaff [Wed, 10 Aug 2011 23:16:16 +0000 (16:16 -0700)]
ofproto: Add 'ofproto' parameter to most flow_mod functions.

12 years agoofproto: Avoid using list_size() to compute length of 'pending' list.
Ben Pfaff [Wed, 10 Aug 2011 21:48:33 +0000 (14:48 -0700)]
ofproto: Avoid using list_size() to compute length of 'pending' list.

Currently this only gets checked for incoming OpenFlow OFPT_FLOW_MOD
messages, so it's hard to imagine it being any kind of bottleneck, but the
NXAST_LEARN action that is soon to be added will be able to create flows
more quickly than we normally expect from a controller.  (On the other
hand, ofproto-dpif, outside of a special testing mode, always completes
operations immediately, so 'pending' will always have length 0.  But this
change still feels right to me for some reason.)

12 years agoofp-util: New function for parsing key-value pairs.
Ben Pfaff [Wed, 17 Aug 2011 20:39:02 +0000 (13:39 -0700)]
ofp-util: New function for parsing key-value pairs.

This will soon have a new user, but it's a worthwhile cleanup on its own.

12 years agoofp-util: Add type-safe functions for serializing actions.
Ben Pfaff [Tue, 16 Aug 2011 23:30:57 +0000 (16:30 -0700)]
ofp-util: Add type-safe functions for serializing actions.

12 years agoofp-parse: Refactor action parsing to improve compiler warnings.
Ben Pfaff [Tue, 16 Aug 2011 23:08:24 +0000 (16:08 -0700)]
ofp-parse: Refactor action parsing to improve compiler warnings.

When a new action is added, compiler warnings show most of the places that
need new code to handle that action.  The action parsing code in
ofp-parse.c was the one remaining missing case.  This commit fixes that.

12 years agoofp-util: Further abstract definitions of action properties.
Ben Pfaff [Wed, 17 Aug 2011 18:01:17 +0000 (11:01 -0700)]
ofp-util: Further abstract definitions of action properties.

This commit primarily moves the OFPAT_ACTION and NXAST_ACTION invocations
into a new file ofp-util.def.  This allows multiple places in the source to
use them.

This commit also adds a new function ofputil_action_code_from_name().
The following commit will add the first user.

12 years agoofp-util: New functions for parsing and formatting OpenFlow port numbers.
Ben Pfaff [Tue, 16 Aug 2011 22:26:18 +0000 (15:26 -0700)]
ofp-util: New functions for parsing and formatting OpenFlow port numbers.

These functions were previously used only in ofp-parse.c and ofp-print.c,
but they are more generally useful and future commits will add more users.

12 years agopackets: Add more utility functions for IPv4 and IPv6 addresses.
Ben Pfaff [Wed, 17 Aug 2011 17:55:15 +0000 (10:55 -0700)]
packets: Add more utility functions for IPv4 and IPv6 addresses.

We had these functions scattered around the source tree anyway.  packets.h
is a good place to centralize them.

I do plan to introduce some additional callers.

12 years agoclassifier: Change cls_rule_set_nd_target() to take a pointer.
Ben Pfaff [Mon, 12 Sep 2011 17:57:28 +0000 (10:57 -0700)]
classifier: Change cls_rule_set_nd_target() to take a pointer.

The other cls_rule_*() functions that take IPv6 addresses take a pointer
to an in6_addr, so cls_rule_set_nd_target() should as well for consistency.
Possibly this is more efficient also, although I guess it doesn't really
make much of a difference either way.

12 years agobridge: Don't update CFM on synthetic interfaces.
Ethan Jackson [Tue, 13 Sep 2011 05:39:01 +0000 (22:39 -0700)]
bridge: Don't update CFM on synthetic interfaces.

Synthetic interfaces don't have database records so it doesn't make
sense to update them.  In some situations this could cause a
segmentation fault.

Reported-by: Paul Ingram <paul@nicira.com>
Bug #7278.

12 years agotests: Test ofproto-dpif set_tunnel translation.
Ethan Jackson [Tue, 13 Sep 2011 00:19:13 +0000 (17:19 -0700)]
tests: Test ofproto-dpif set_tunnel translation.

This patch adds a unit test which would have caught the issue fixed
in Commit 2446268e "ofproto-dpif: set_tunnel when required to."

12 years agoofproto-dpif: Emit set_tunnel when required to.
Ethan Jackson [Mon, 12 Sep 2011 23:56:21 +0000 (16:56 -0700)]
ofproto-dpif: Emit set_tunnel when required to.

ofproto-dpif assumed that the datapath initialized the tun_id of a
flow on egress, to its tun_id on ingress.  For this reason, if
OpenFlow actions set the tun_id to a flow's ingress tun_id,
ofproto-dpif would fail to emit a set_tunnel action.

Reported-by: Igor Ganichev <iganichev@nicira.com>
Reported-by: Pankaj Thakkar <thakkar@nicira.com>
12 years agodatapath: Strip down vport interface : OVS_VPORT_ATTR_MTU
Pravin Shelar [Tue, 13 Sep 2011 00:12:52 +0000 (17:12 -0700)]
datapath: Strip down vport interface : OVS_VPORT_ATTR_MTU

There is no need to have vport attribute MTU (OVS_VPORT_ATTR_MTU) as
linux net-dev-ioctl can be used to get/set MTU for linux device.
Following patch removes OVS_VPORT_ATTR_MTU from datapath protocol.

This patch also adds netdev_set_mtu interface. So that MTU adjustments
can be done from OVS userspace. get_mtu() interface is also changed, now
get_mtu() returns EOPNOTSUPP rather than returning 0 and setting *pmtu
to INT_MAX in case there is no MTU attribute for given device.

Signed-off-by: Pravin B Shelar <pshelar@nicira.com>
Acked-by: Jesse Gross <jesse@nicira.com>
12 years agoofproto-dpif: Fix check for 802.1Q header in commit_odp_actions().
Ben Pfaff [Mon, 12 Sep 2011 17:49:55 +0000 (10:49 -0700)]
ofproto-dpif: Fix check for 802.1Q header in commit_odp_actions().

The 'vlan_tci' member of struct flow has value 0, not OFP_VLAN_NONE, when
there is no 802.1Q header.

Fixes a problem introduced in commit d9065a90b6b "datapath: VLAN actions
should use push/pop semantics."

Found by sparse.

CC: Pravin Shelar <pshelar@nicira.com>
12 years agoNEWS: Mention support for keys in CAPWAP.
Jesse Gross [Mon, 12 Sep 2011 17:43:46 +0000 (10:43 -0700)]
NEWS: Mention support for keys in CAPWAP.

12 years agoDebian: fail gracefully if modules can't be loaded on install
Simon Horman [Sat, 10 Sep 2011 06:09:31 +0000 (15:09 +0900)]
Debian: fail gracefully if modules can't be loaded on install

By registering an error-handler for the init script used
in openvswitch-switch.postinst and detecting if module insertion fails,
it is possible to avoid failure to install in the case where the
openvswitch_mod module is not available.

This is done without altering the behaviour that the start target
of the openvswitch-switch init script will fail if module insertion fails.

This patch also adds a friendly hint as as to why starting
openvswitch-switch has failed if it is due to failure to insert
the openvswtich_mod. This message is displayed as necessary both
on package install and other calls to the start target of the
init script.

[Ben Pfaff fixed up == to = in postinst]

12 years agodatapath: add key support to CAPWAP tunnel
Valient Gough [Wed, 24 Aug 2011 06:12:06 +0000 (15:12 +0900)]
datapath: add key support to CAPWAP tunnel

Add tunnel key support to CAPWAP vport.  Uses the optional WSI field in a
CAPWAP header to store a 64bit key.  It can also be used without keys, in which
case it is backward compatible with the old code.  Documentation about the
WSI field format is in CAPWAP.txt.

Signed-off-by: Valient Gough <vgough@pobox.com>
[horms@verge.net.au: Various minor fixes (v4.1)]
Signed-off-by: Simon Horman <horms@verge.net.au>
[jesse: Additional parsing fixes]
Signed-off-by: Jesse Gross <jesse@nicira.com>
12 years agodatapath: Improve kernel hash table
Pravin Shelar [Sat, 10 Sep 2011 02:09:47 +0000 (19:09 -0700)]
datapath: Improve kernel hash table

 Currently OVS uses its own hashing implmentation for hash tables
which has some problems, e.g. error case on deletion code.
Following patch replaces that with hlist based hash table which is
consistent with other kernel hash tables. As Jesse suggested, flex-array
is used for allocating hash buckets, So that we can have large
hash-table without large contiguous kernel memory.

Signed-off-by: Pravin B Shelar <pshelar@nicira.com>
Acked-by: Jesse Gross <jesse@nicira.com>
12 years agodatapath: VLAN actions should use push/pop semantics
Pravin Shelar [Sat, 10 Sep 2011 01:13:26 +0000 (18:13 -0700)]
datapath: VLAN actions should use push/pop semantics

Currently the kernel vlan actions mirror those used by OpenFlow 1.0.
i.e. MODIFY and STRIP. More flexible approach is to have an action to
push a tag and pop a tag off, so that it can handle multiple levels of vlan
tags. Plus it aligns with newer version of OpenFlow.
        As this patch replaces MODIFY with PUSH semantic, action
mapping done in userpace is fixed accordingly.
        GSO handling for multiple levels of vlan tags is also added as
Jesse suggested before.

Signed-off-by: Pravin B Shelar <pshelar@nicira.com>
Acked-by: Jesse Gross <jesse@nicira.com>
12 years agodatapath: Fix br_nlmsg_size
Pravin Shelar [Sat, 10 Sep 2011 00:07:35 +0000 (17:07 -0700)]
datapath: Fix br_nlmsg_size

I missed this in last vport iflink patch.
    As IFLA_LINK is not be passed in netlink msg there is no need to
allocate space for it.

Signed-off-by: Pravin B Shelar <pshelar@nicira.com>
Acked-by: Jesse Gross <jesse@nicira.com>
12 years agonetlink-socket: Avoid use-after-free in nl_lookup_genl_mcgroup().
Ben Pfaff [Fri, 9 Sep 2011 17:21:49 +0000 (10:21 -0700)]
netlink-socket: Avoid use-after-free in nl_lookup_genl_mcgroup().

Commit e408762f "netlink-socket: New function nl_lookup_genl_mcgroup()"
modified do_lookup_genl_family() to return the Netlink attributes to the
caller, but it still freed the Netlink message itself, which meant that
the attributes pointed into freed memory.  This commit fixes the problem.

This commit is not a minimal fix.  It refactors do_lookup_genl_family(),
changing the return value from "negative errno value or positive genl
family id" to the more common "zero or positive errno value".

Found by valgrind.

12 years agobridge: Avoid reading uninitialized data in bridge_pick_local_hw_addr().
Ben Pfaff [Fri, 9 Sep 2011 17:12:32 +0000 (10:12 -0700)]
bridge: Avoid reading uninitialized data in bridge_pick_local_hw_addr().

Commit 3a48ace3 "bridge: Make bridge_pick_local_hw_addr() easier to reason"
didn't initialize 'ea' before trying to compare against it.  We need to
check that an address has been found.

Found by valgrind.

12 years agocfm: Write remote MPIDs to the database.
Ethan Jackson [Fri, 26 Aug 2011 21:54:34 +0000 (14:54 -0700)]
cfm: Write remote MPIDs to the database.

A controller may want to know which MPIDs are reachable from an
interface configured with CFM.  This patch regularly writes this
information to the database.

Bug #7014.

12 years agocfm: Eight byte MPIDs in extended mode.
Ethan Jackson [Thu, 1 Sep 2011 20:28:25 +0000 (13:28 -0700)]
cfm: Eight byte MPIDs in extended mode.

802.1ag only allows for MPIDs in the range [1, 8191].  This is
restrictive enough to make assignment of MPIDs to instances of OVS
awkward.  This patch allows eight byte MPIDs when running in
extended mode.

Bug #7014.

12 years agocfm: Allow accurate transmission intervals in extended mode.
Ethan Jackson [Wed, 31 Aug 2011 00:37:15 +0000 (17:37 -0700)]
cfm: Allow accurate transmission intervals in extended mode.

The standard CFM protocol only allows a handful of transmission
rates.  This is particularly problematic if you want to support a
transmission rate slower than 100 ms and faster than 1000 ms.

This patch allows arbitrary transmission rates (between 1 ms and
65535 ms).  It does this by commandeering parts of a reserved
"zero" field in the ccm message.  This breaks wire compatibility
with standard 802.1ag implementations, and thus is only supported
in extended mode.

Bug #7014.

12 years agocfm: New cfm extended mode.
Ethan Jackson [Wed, 31 Aug 2011 00:37:06 +0000 (17:37 -0700)]
cfm: New cfm extended mode.

The new extended mode introduced in this patch will be used for
features which break wire compatibility with 802.1ag compliant
implementations.

Bug #7014.

12 years agocfm: Remove cfm_remote_mpid configuration.
Ethan Jackson [Fri, 19 Aug 2011 20:58:56 +0000 (13:58 -0700)]
cfm: Remove cfm_remote_mpid configuration.

According to the 802.1ag specification, users should be able to
configure the CFM module with a list of remote endpoints with which
the local endpoint should have connectivity.  Commit 93b8df3853
"cfm: Remove Maintenance_Point and Monitor tables." changed the
behavior so that only one remote endpoint could be specified.  This
commit takes it further, by disallowing specification of any
remote endpoints.

Due to this change, the semantics of the fault flag are slightly
different.  Before, a fault was triggered if any of the configured
remote endpoints were unreachable (or with RDI), or if any
unconfigured remote endpoints were reachable.  Now a fault is
triggered if no remote endpoints are reachable at all, or if
reachable endpoints have set their RDI.

Bug #7014.

12 years agocfm: Trigger fault on unexpected CCM reception.
Ethan Jackson [Tue, 30 Aug 2011 01:56:28 +0000 (18:56 -0700)]
cfm: Trigger fault on unexpected CCM reception.

12 years agobridge: Clear fault when CFM is not configured.
Ethan Jackson [Wed, 7 Sep 2011 00:13:24 +0000 (17:13 -0700)]
bridge: Clear fault when CFM is not configured.

12 years agobridge: Write CFM changes more aggressively.
Ethan Jackson [Thu, 8 Sep 2011 23:15:36 +0000 (16:15 -0700)]
bridge: Write CFM changes more aggressively.

This patch no longer rate limits database updates due to CFM
changes.  Due to recent changes, the fault status of CFM only
changes once per 3.5 tx_interval seconds.  There doesn't seem to be
a good reason to add an additional rate limit on top of this.

12 years agobridge: ovsdb_idl_omit_alert() on additional columns.
Ethan Jackson [Thu, 8 Sep 2011 23:19:10 +0000 (16:19 -0700)]
bridge: ovsdb_idl_omit_alert() on additional columns.

The bridge owns the lacp_current and cfm_fault columns and should
not be alerted when they change.

12 years agoofproto: Document that ->rule_construct() should uninitialize victim rules.
Ben Pfaff [Fri, 9 Sep 2011 19:45:15 +0000 (12:45 -0700)]
ofproto: Document that ->rule_construct() should uninitialize victim rules.

The comments didn't say how this should work, so this clarifies it.

12 years agoofproto: Fix documentation for calls to ->rule_destruct().
Ben Pfaff [Thu, 8 Sep 2011 23:22:38 +0000 (16:22 -0700)]
ofproto: Fix documentation for calls to ->rule_destruct().

This documented what I intended to implement and what I thought I had
implemented, but not what the code actually did.  It is a little easier to
fix the documentation to match the implementation than vice versa, so this
commit does so.

Reported-by: Hao Zheng <hzheng@nicira.com>
12 years agoofproto-dpif: Drop special packets more judiciously.
Ethan Jackson [Wed, 31 Aug 2011 01:19:48 +0000 (18:19 -0700)]
ofproto-dpif: Drop special packets more judiciously.

Before this patch, ofproto-dpif would not forward any LACP or CFM
packets regardless of whether or not the relevant modules were
configured on the in_port.

Feature #6316.

12 years agolacp: Clean up LACP module interface.
Ethan Jackson [Wed, 31 Aug 2011 01:17:27 +0000 (18:17 -0700)]
lacp: Clean up LACP module interface.

There's no particular reason to force users of the LACP module to
be aware of the lacp_pdu structure.  This patch hides that
information in the LACP module implementation.  This results in
slightly cleaner code which is more consistent with the CFM
module.

12 years agovswitch.xml: Whitespace cleanup.
Ethan Jackson [Wed, 31 Aug 2011 01:46:50 +0000 (18:46 -0700)]
vswitch.xml: Whitespace cleanup.