sliver-openvswitch.git
12 years agoDESIGN: Describe principles of in-band control.
Ben Pfaff [Mon, 23 Apr 2012 21:15:32 +0000 (14:15 -0700)]
DESIGN: Describe principles of in-band control.

These principles are drawn from an email I sent to the openflow-spec list
long ago.

Signed-off-by: Ben Pfaff <blp@nicira.com>
12 years agotests: Generalize 'sed' calls in MAC learning test to more than one digit.
Ben Pfaff [Mon, 23 Apr 2012 17:07:29 +0000 (10:07 -0700)]
tests: Generalize 'sed' calls in MAC learning test to more than one digit.

With "check-valgrind" the test can take more than 10 seconds to run, so
replacing only a single trailing digit with ? ends up with 1? which causes
the test to fail.

Signed-off-by: Ben Pfaff <blp@nicira.com>
12 years agotests: Avoid daemon race between pidfile creation and parent notification.
Ben Pfaff [Mon, 23 Apr 2012 17:05:50 +0000 (10:05 -0700)]
tests: Avoid daemon race between pidfile creation and parent notification.

This was causing "check-valgrind" to sometimes fail in this test.

Signed-off-by: Ben Pfaff <blp@nicira.com>
12 years agoofproto: Fix use-after-free error when ports disappear.
Ben Pfaff [Mon, 23 Apr 2012 16:16:18 +0000 (09:16 -0700)]
ofproto: Fix use-after-free error when ports disappear.

update_port() can delete the port for which it is called, if the underlying
network device has been destroyed, so HMAP_FOR_EACH is unsafe in
ofproto_run().

Less obviously, update_port() can delete unrelated ports.  For example,
suppose that initially device A is port 1 and device B is port 2.  If
update_port("A") runs just after this, then it will ofport_remove() both
ports, then ofport_install() A as the new port 2.

So this commit first assembles a list of ports to update, then updates them
in a separate loop.

Without this commit, running "ovs-dpctl del-dp" while ovs-vswitchd is
running consistently causes a crash for me within a few seconds.

Signed-off-by: Ben Pfaff <blp@nicira.com>
12 years agovswitchd: Make "cfm_health" column ephemeral.
Ben Pfaff [Fri, 20 Apr 2012 18:12:46 +0000 (11:12 -0700)]
vswitchd: Make "cfm_health" column ephemeral.

There's no need to log this to the on-disk database.

Spotted while examining "ovsdb-tool show-log" output.

Reported-by: Ethan Jackson <ethan@nicira.com>
Signed-off-by: Ben Pfaff <blp@nicira.com>
12 years agovswitchd: Make "cfm_fault_status" column ephemeral.
Ben Pfaff [Fri, 20 Apr 2012 18:03:14 +0000 (11:03 -0700)]
vswitchd: Make "cfm_fault_status" column ephemeral.

There's no need to log this to the on-disk database.

Spotted while examining "ovsdb-tool show-log" output.

Signed-off-by: Ben Pfaff <blp@nicira.com>
12 years agoautomake: ofp-errors.[c|inc] are actually inside srcdir
Ansis Atteka [Thu, 19 Apr 2012 19:54:16 +0000 (12:54 -0700)]
automake: ofp-errors.[c|inc] are actually inside srcdir

This patch fixes a build error when OVS is built inside
"./_debian" directory.

To reproduce this issue run "fakeroot debian/rules binary"
or "debuild binary" inside the git root directory.

Signed-off-by: Ansis Atteka <aatteka@nicira.com>
12 years agodocumantation: Trivial fix for kvm option error.
Yasuhito Takamiya [Fri, 20 Apr 2012 14:38:04 +0000 (07:38 -0700)]
documantation: Trivial fix for kvm option error.

Signed-off-by: Yasuhito Takamiya <yasuhito@gmail.com>
Signed-off-by: Ben Pfaff <blp@nicira.com>
12 years agoovs-vsctl: Speed up port management operations with many ports.
Ben Pfaff [Tue, 17 Apr 2012 23:56:21 +0000 (16:56 -0700)]
ovs-vsctl: Speed up port management operations with many ports.

This makes a sequence of 10,000 "add-port" operations on a single ovs-vsctl
command line about 4X faster.  It makes a sequence of 10,000 "del-port"
operations on a single command line over 2X faster.

It works by not repopulating the cache of relationships between bridges,
ports, and interfaces after most operations, instead updating them
incrementally in-place.

Signed-off-by: Ben Pfaff <blp@nicira.com>
12 years agoovs-vsctl: Remove 'ctrl', 'n_ctrl' from struct vsctl_bridge.
Ben Pfaff [Tue, 17 Apr 2012 20:56:38 +0000 (13:56 -0700)]
ovs-vsctl: Remove 'ctrl', 'n_ctrl' from struct vsctl_bridge.

Only the controller commands used these members and they didn't even help
those commands very much.

Signed-off-by: Ben Pfaff <blp@nicira.com>
12 years agoovs-vsctl: Remove 'fail_mode' member from struct vsctl_bridge.
Ben Pfaff [Thu, 19 Apr 2012 21:27:40 +0000 (14:27 -0700)]
ovs-vsctl: Remove 'fail_mode' member from struct vsctl_bridge.

It's only used in cmd_get_fail_mode(), which can easily look it up for
itself, so there's no benefit to storing it in every vsctl_bridge record.

Signed-off-by: Ben Pfaff <blp@nicira.com>
12 years agoovs-vsctl: Merge struct vsctl_info into struct vsctl_context.
Ben Pfaff [Wed, 18 Apr 2012 00:07:00 +0000 (17:07 -0700)]
ovs-vsctl: Merge struct vsctl_info into struct vsctl_context.

To speed up management operations with many ports, we need to preserve the
cache of bridges, ports, and interfaces from one operation to the next.
One necessary step is to push the "struct vsctl_info" that did the caching
up from the individual functions that need it into a more global structure.
This commit does that, merging it into struct vsctl_context.

This commit also modifies do_vsctl(), the top-level control code in
ovs-vsctl, to keep this part of the vsctl_context unchanged from running
one command to the next.

Signed-off-by: Ben Pfaff <blp@nicira.com>
12 years agoovs-vsctl: Verify VLAN bridge controllers in cmd_get_controller().
Ben Pfaff [Tue, 17 Apr 2012 20:50:53 +0000 (13:50 -0700)]
ovs-vsctl: Verify VLAN bridge controllers in cmd_get_controller().

A VLAN bridge uses its parent's controllers, so checking the controller
should verify the parent's set of controllers.

The change to verify_controllers() isn't necessary; it just deletes
the check for a null 'bridge' because verify_controllers() can no
longer be called with a null 'bridge'.

This fixes a bug, but it is unlikely to ever have caused a real problem for
users.

Found by inspection.

Signed-off-by: Ben Pfaff <blp@nicira.com>
12 years agoovs-vsctl: Verify correct record in cmd_get_fail_mode() for VLAN bridges.
Ben Pfaff [Tue, 17 Apr 2012 20:44:51 +0000 (13:44 -0700)]
ovs-vsctl: Verify correct record in cmd_get_fail_mode() for VLAN bridges.

A VLAN bridge uses its parent's fail-mode, so checking the fail-mode should
verify the parent's bridge record.

This fixes a bug, but it is unlikely to ever have caused a real problem for
users.

Found by inspection.

Signed-off-by: Ben Pfaff <blp@nicira.com>
12 years agodatapath: Use ETH_ALEN instead of VLAN_ETH_ALEN
Simon Horman [Thu, 19 Apr 2012 07:38:39 +0000 (16:38 +0900)]
datapath: Use ETH_ALEN instead of VLAN_ETH_ALEN

VLAN_ETH_ALEN will be removed in version 3.5 of the Linux kernel.

Signed-off-by: Simon Horman <horms@verge.net.au>
Signed-off-by: Jesse Gross <jesse@nicira.com>
12 years agodatapath: Include linux/completion.h
Simon Horman [Thu, 19 Apr 2012 07:38:20 +0000 (16:38 +0900)]
datapath: Include linux/completion.h

This is needed for compilation against Linux 2.6.19

Signed-off-by: Simon Horman <horms@verge.net.au>
Signed-off-by: Jesse Gross <jesse@nicira.com>
12 years agoofproto-dpif: Avoid extra flow copy in xlate_actions() for unneeded warnings.
Ben Pfaff [Mon, 16 Apr 2012 22:54:37 +0000 (15:54 -0700)]
ofproto-dpif: Avoid extra flow copy in xlate_actions() for unneeded warnings.

The copy of the extra flow copy here was showing up in profiles.  We only
need this copy if we end up doing a "trace" to warn the user.  Most runs
won't ever do that, so don't start making copies until we actually hit
such a case.

This has a small behavioral change in that we'll only get a warning on the
*second* time we hit the resubmit recursion limit, not on the first.  I
doubt that's really a problem.

Signed-off-by: Ben Pfaff <blp@nicira.com>
12 years agoofproto-dpif: Avoid extra flow copy in xlate_actions() if no mirrors.
Ben Pfaff [Mon, 16 Apr 2012 22:43:15 +0000 (15:43 -0700)]
ofproto-dpif: Avoid extra flow copy in xlate_actions() if no mirrors.

This was showing up on profiles.

Signed-off-by: Ben Pfaff <blp@nicira.com>
12 years agoofproto-dpif: Avoid calling get_ofp_port() twice in xlate_normal().
Ben Pfaff [Mon, 16 Apr 2012 22:22:52 +0000 (15:22 -0700)]
ofproto-dpif: Avoid calling get_ofp_port() twice in xlate_normal().

This was showing up in profiles.

Signed-off-by: Ben Pfaff <blp@nicira.com>
12 years agoofproto-dpif: Implement "flow setup governor" to speed up many short flows.
Ben Pfaff [Thu, 19 Apr 2012 00:11:10 +0000 (17:11 -0700)]
ofproto-dpif: Implement "flow setup governor" to speed up many short flows.

The cost of creating and initializing facets and subfacets and installing,
tracking, and uninstalling kernel flows is significant.  When most flows
have only one or a few packets, this overhead is higher than the cost of
handling each packet individually.  This commit introduces heuristics that
cheaply count (approximately) the number of packets seen in a flow and
skips most of this expensive bookkeeping until the packet count exceeds a
threshold (currently 5 packets).

Signed-off-by: Ben Pfaff <blp@nicira.com>
12 years agoofproto-dpif: Avoid malloc() in common case for "execute" operations.
Ben Pfaff [Wed, 18 Apr 2012 23:54:10 +0000 (16:54 -0700)]
ofproto-dpif: Avoid malloc() in common case for "execute" operations.

Signed-off-by: Ben Pfaff <blp@nicira.com>
12 years agoofproto-dpif: Make it easier to credit statistics for resubmits.
Ben Pfaff [Fri, 13 Apr 2012 21:09:10 +0000 (14:09 -0700)]
ofproto-dpif: Make it easier to credit statistics for resubmits.

Until now, crediting statistics to OpenFlow rules due to "resubmit" actions
has required setting up a "resubmit hook" with a callback function and
auxiliary data.  This commit makes it easier to do, by adding a member to
struct action_xlate_ctx that specifies statistics to credit to each
resubmitted rule.

This commit includes one small behavioral change as an optimization.
Previously, rule_execute() translated the rule twice: once to get the ODP
actions, then a second time after executing the ODP actions to credit
statistics to the rules.  After this commit, rule_execute() translates the
rule only once, crediting statistics as a side effect.  The difference only
becomes visible when executing the actions fails: previously the statistics
would not be incremented, after this commit they will be.  It is very
unusual for executing actions to fail (generally this indicates a bug) so
I'm not concerned about it.

Signed-off-by: Ben Pfaff <blp@nicira.com>
12 years agoclassifier: Optimize search of "catchall" table.
Ben Pfaff [Mon, 9 Apr 2012 22:49:22 +0000 (15:49 -0700)]
classifier: Optimize search of "catchall" table.

Most flow tables have some kind of "catchall" rule that matches every
packet.  For this table, the cost of copying, zeroing, and hashing the
input flow is significant.  This patch avoids these costs.

Signed-off-by: Ben Pfaff <blp@nicira.com>
12 years agomac-learning: Speed up mac_table_hash().
Ben Pfaff [Mon, 9 Apr 2012 21:33:56 +0000 (14:33 -0700)]
mac-learning: Speed up mac_table_hash().

Signed-off-by: Ben Pfaff <blp@nicira.com>
12 years agohash: Speed up hash_bytes().
Ben Pfaff [Mon, 9 Apr 2012 21:33:33 +0000 (14:33 -0700)]
hash: Speed up hash_bytes().

Signed-off-by: Ben Pfaff <blp@nicira.com>
12 years agoofproto-dpif: Avoid malloc() of "struct flow_miss".
Ben Pfaff [Sat, 7 Apr 2012 00:11:18 +0000 (17:11 -0700)]
ofproto-dpif: Avoid malloc() of "struct flow_miss".

In addition to avoid malloc() for struct flow_miss, this commit avoids
copying "struct flow" around, which is a significant benefit because
struct flow is currently 144 bytes.

Signed-off-by: Ben Pfaff <blp@nicira.com>
12 years agodpif: Make caller of dpif_recv() provide buffer space.
Ben Pfaff [Fri, 6 Apr 2012 23:23:28 +0000 (16:23 -0700)]
dpif: Make caller of dpif_recv() provide buffer space.

This improves performance under heavy flow setup loads.

Signed-off-by: Ben Pfaff <blp@nicira.com>
12 years agoofproto-dpif: Avoid malloc() in common case for allocating subfacets.
Ben Pfaff [Fri, 6 Apr 2012 22:30:21 +0000 (15:30 -0700)]
ofproto-dpif: Avoid malloc() in common case for allocating subfacets.

Usually a facet has exactly one subfacet that has the same lifetime as
the facet.  Allocating both the facet and its subfacet in a single memory
block improves performance.

Signed-off-by: Ben Pfaff <blp@nicira.com>
12 years agoofproto-dpif: Move definition of "struct subfacet" earlier.
Ben Pfaff [Fri, 6 Apr 2012 22:27:31 +0000 (15:27 -0700)]
ofproto-dpif: Move definition of "struct subfacet" earlier.

The next commit will add an instance of struct subfacet as a member of
struct facet, so struct subfacet must be declared first.

This commit moves around code without otherwise changing it.

Signed-off-by: Ben Pfaff <blp@nicira.com>
12 years agonetlink-socket: Make caller provide message receive buffers.
Ben Pfaff [Mon, 9 Apr 2012 22:35:29 +0000 (15:35 -0700)]
netlink-socket: Make caller provide message receive buffers.

Typically an nl_sock client can stack-allocate the buffer for receiving
a Netlink message, which provides a performance boost.

Signed-off-by: Ben Pfaff <blp@nicira.com>
12 years agonetlink-socket: Remove unnecessary #include.
Ben Pfaff [Fri, 6 Apr 2012 22:13:37 +0000 (15:13 -0700)]
netlink-socket: Remove unnecessary #include.

Signed-off-by: Ben Pfaff <blp@nicira.com>
12 years agonetlink: Postpone choosing sequence numbers until send time.
Ben Pfaff [Mon, 16 Apr 2012 23:01:01 +0000 (16:01 -0700)]
netlink: Postpone choosing sequence numbers until send time.

Choosing sequence numbers at time of creating a packet means that
nl_sock_transact_multiple() has to search for the sequence number
of a reply, because the sequence numbers of the requests aren't
necessarily sequential.  This commit makes it possible to avoid
the search, by deferring choice of sequence numbers until the
time that we send the packets.  It doesn't actually modify
nl_sock_transact_multiple(), which will happen in a later commit.

Previously, I was concerned about a theoretical race condition
described in a comment in the old versino of this code:

    This implementation uses sequence numbers that are unique
    process-wide, to avoid a hypothetical race: send request, close
    socket, open new socket that reuses the old socket's PID value,
    send request on new socket, receive reply from kernel to old
    socket but with same PID and sequence number.  (This race could be
    avoided other ways, e.g. by preventing PIDs from being quickly
    reused).

However, I no longer believe that this can be a real problem,
because Netlink operates synchronously.  The reply to a request
will always arrive before the socket can be closed and a new
socket opened with the old socket's PID.

Signed-off-by: Ben Pfaff <blp@nicira.com>
12 years agoofproto-dpif: Avoid computing flow hash multiple times.
Ben Pfaff [Wed, 18 Apr 2012 16:17:33 +0000 (09:17 -0700)]
ofproto-dpif: Avoid computing flow hash multiple times.

Signed-off-by: Ben Pfaff <blp@nicira.com>
12 years agodpif-linux: Avoid malloc() in dpif_linux_operate().
Ben Pfaff [Thu, 5 Apr 2012 23:49:22 +0000 (16:49 -0700)]
dpif-linux: Avoid malloc() in dpif_linux_operate().

Signed-off-by: Ben Pfaff <blp@nicira.com>
12 years agoofproto-dpif: Avoid malloc() in common cases for xlate_actions().
Ben Pfaff [Mon, 16 Apr 2012 20:52:09 +0000 (13:52 -0700)]
ofproto-dpif: Avoid malloc() in common cases for xlate_actions().

Signed-off-by: Ben Pfaff <blp@nicira.com>
12 years agoofpbuf: New functions ofpbuf_use_stub() and ofpbuf_get_uninit_pointer().
Ben Pfaff [Wed, 18 Apr 2012 04:55:27 +0000 (21:55 -0700)]
ofpbuf: New functions ofpbuf_use_stub() and ofpbuf_get_uninit_pointer().

Signed-off-by: Ben Pfaff <blp@nicira.com>
12 years agoofproto-dpif: Don't do an extra flow translation when removing facets.
Ben Pfaff [Thu, 5 Apr 2012 20:43:24 +0000 (13:43 -0700)]
ofproto-dpif: Don't do an extra flow translation when removing facets.

When we remove a facet, it seems superfluous to translate the flow
once more just to update the MAC learning table, since we know that
it was done within a second or so anyway (e.g. when the flow stats
were last updated).

Signed-off-by: Ben Pfaff <blp@nicira.com>
12 years agonetlink-socket: Avoid forcing a reply for final message in a transaction.
Ben Pfaff [Thu, 5 Apr 2012 19:58:42 +0000 (12:58 -0700)]
netlink-socket: Avoid forcing a reply for final message in a transaction.

Signed-off-by: Ben Pfaff <blp@nicira.com>
12 years agodpif: Include TCP flags in "ovs-dpctl dump-flows" output.
Ben Pfaff [Thu, 5 Apr 2012 17:24:56 +0000 (10:24 -0700)]
dpif: Include TCP flags in "ovs-dpctl dump-flows" output.

Signed-off-by: Ben Pfaff <blp@nicira.com>
12 years agoofproto-dpif: Batch flow uninstallations due to expiration.
Ben Pfaff [Wed, 18 Apr 2012 04:52:10 +0000 (21:52 -0700)]
ofproto-dpif: Batch flow uninstallations due to expiration.

Signed-off-by: Ben Pfaff <blp@nicira.com>
12 years agoofproto-dpif: Drop unneeded _SAFE from iteration in handle_flow_miss().
Ben Pfaff [Mon, 16 Apr 2012 21:07:56 +0000 (14:07 -0700)]
ofproto-dpif: Drop unneeded _SAFE from iteration in handle_flow_miss().

The loop never frees the packet that it operates on, so _SAFE is not
needed.

Signed-off-by: Ben Pfaff <blp@nicira.com>
12 years agoofproto-dpif: Keep subfacet "used" times more up-to-date.
Ben Pfaff [Fri, 13 Apr 2012 21:15:15 +0000 (14:15 -0700)]
ofproto-dpif: Keep subfacet "used" times more up-to-date.

handle_flow_miss() didn't update subfacet "used" times for packets
processed by userspace.  This commit fixes the problem.

Found by inspection.  I didn't verify the bug in testing.

Signed-off-by: Ben Pfaff <blp@nicira.com>
12 years agoovs-test: Enhancements to the ovs-test tool
Ansis Atteka [Fri, 30 Mar 2012 02:03:08 +0000 (19:03 -0700)]
ovs-test: Enhancements to the ovs-test tool

-Implemented support for ovs-test client, so that it could automatically
spawn an ovs-test server process from itself. This reduces the number of
commands the user have to type to get tests running.
-Automated creation of OVS bridges and ports (for VLAN and GRE tests), so that
user would not need to invoke ovs-vsctl manually to switch from direct, 802.1Q
and GRE tests.
-Fixed some pylint reported warnings.
-Fixed ethtool invocation so that we always try to query the physical interface
to get the driver name and version.
-and some others enhancements.

The new usage:
Node1:ovs-test -s 15531
Node2:ovs-test -c 127.0.0.1,1.1.1.1 192.168.122.151,1.1.1.2 -d -l 125 -t gre

Signed-off-by: Ansis Atteka <aatteka@nicira.com>
12 years agolacp: Remove heartbeat mode.
Ethan Jackson [Mon, 16 Apr 2012 22:01:09 +0000 (15:01 -0700)]
lacp: Remove heartbeat mode.

The LACP heartbeat mode was used to monitor interfaces for
connectivity.  It turns out that LACP is inferior to CFM for this
purpose.  For the sake of simplicity this patch removes the
feature.

Signed-off-by: Ethan Jackson <ethan@nicira.com>
12 years agolacp: Remove custom transmission intervals.
Ethan Jackson [Mon, 16 Apr 2012 21:55:58 +0000 (14:55 -0700)]
lacp: Remove custom transmission intervals.

Open vSwitch allowed users to set a custom LACP PDU transmission
interval.  This turned out to be an ill conceived idea which was
more confusing than useful.  This patch reverts Open vSwitch to the
behavior supported in the LACP specification: two transmission
intervals, fast and slow.

Signed-off-by: Ethan Jackson <ethan@nicira.com>
12 years agovswitch: Use consistent representation of DSCP bits.
Ethan Jackson [Mon, 16 Apr 2012 19:09:49 +0000 (12:09 -0700)]
vswitch: Use consistent representation of DSCP bits.

There are two sensible ways to represent the 6 DSCP bits of an IP
packet.  One could represent them as an integer in the range 0 to
63.  Or one could represent them as they would appear in the tos
field (0 to 63) << 2.  Before this patch, OVS had used the former
method for the DSCP bits in the Queue Table, and the latter for the
DSCP in the Controller and Manager tables.  Since the ability to
set DSCP bits in the Controller and Manager tables is so new that
it hasn't been released yet, this patch changes it to use the
existing style employed in the Queue table.  Hopefully this should
make the code and configuration less confusing.

Signed-off-by: Ethan Jackson <ethan@nicira.com>
12 years agosocket-util: Remove DSCP_INVALID.
Ethan Jackson [Mon, 16 Apr 2012 20:56:58 +0000 (13:56 -0700)]
socket-util: Remove DSCP_INVALID.

The DSCP_INVALID flag allowed callers to prevent socket-util from
modify the DSCP bits of newly created sockets.  However, the two
really important callers (implementations of the controller and
manager tables) never used it.  Furthermore, the other callers
would be fine always setting the DSCP bits to zero.  This patch
removes the DSCP_INVALID option in an effort to simplify the code.

Signed-off-by: Ethan Jackson <ethan@nicira.com>
12 years agolib: Pull dscp bits out of reconnect.
Ethan Jackson [Thu, 12 Apr 2012 02:52:46 +0000 (19:52 -0700)]
lib: Pull dscp bits out of reconnect.

The DSCP bits of a connection have nothing to do with the
reconnection state machine.  This pulls them up into jsonrpc which
needs them to properly establish connections.

Signed-off-by: Ethan Jackson <ethan@nicira.com>
12 years agosocket-util: Close socket on failed dscp modification.
Ethan Jackson [Mon, 16 Apr 2012 19:46:46 +0000 (12:46 -0700)]
socket-util: Close socket on failed dscp modification.

If socket-util failed to modify the dscp bits of an active
connection, it would fail to close the file descriptor potentially
causing a leak. Found by inspection.

Signed-off-by: Ethan Jackson <ethan@nicira.com>
12 years agodocumentation: Add INSTALL.Libvirt file
Ansis Atteka [Tue, 10 Apr 2012 21:27:31 +0000 (14:27 -0700)]
documentation: Add INSTALL.Libvirt file

Add documentation on how to use OVS integration
with the upcoming Libvirt 0.9.11 release.

Signed-off-by: Ansis Atteka <aatteka@nicira.com>
12 years agoovs-save: Add "dev" keyword before interface names in the ip commands
Ansis Atteka [Mon, 16 Apr 2012 06:30:15 +0000 (23:30 -0700)]
ovs-save: Add "dev" keyword before interface names in the ip commands

The "service force-reload-kmod" command did not work properly, if there
was a bridge, which name was a prefix of string "broadcast" (e.g. "br").

To reproduce:
ovs-vsctl add-br br
service openvswitch-switch force-reload-kmod

This patch will make sure that the generated ip commands are clearly
interpretable by the ip utility.

Issue #9841
Signed-off-by: Ansis Atteka <aatteka@nicira.com>
12 years agostream: By default disable probing on unix sockets.
Ethan Jackson [Thu, 12 Apr 2012 03:18:34 +0000 (20:18 -0700)]
stream: By default disable probing on unix sockets.

There isn't a lot of value in sending inactivity probes on unix
sockets.  This patch changes the default to disable them.

Signed-off-by: Ethan Jackson <ethan@nicira.com>
12 years agopython: Honor zero probe interval in reconnect.py
Ethan Jackson [Thu, 12 Apr 2012 03:42:01 +0000 (20:42 -0700)]
python: Honor zero probe interval in reconnect.py

The python reconnect library attempted to send a probe every 0
milliseconds instead of disabling probing when the probe_interval
was zero.

Signed-off-by: Ethan Jackson <ethan@nicira.com>
12 years agolearn: Make it possible to parse "load" actions wider than 64 bits.
Ben Pfaff [Sat, 14 Apr 2012 04:24:17 +0000 (21:24 -0700)]
learn: Make it possible to parse "load" actions wider than 64 bits.

The implementation of the "learn" action now properly implements
specifications such as 0x20010db885a308d313198a2e03707348->NXM_NX_IPV6_DST
but the parser used in ovs-ofctl and elsewhere could not generate such
specifications.  This commit adds that support.

Signed-off-by: Ben Pfaff <blp@nicira.com>
12 years agolearn: Fix bugs when learn actions use subfields wider than 64 bits.
Ben Pfaff [Thu, 12 Apr 2012 00:08:13 +0000 (17:08 -0700)]
learn: Fix bugs when learn actions use subfields wider than 64 bits.

Bug #10576.
Reported-by: James Schmidt <jschmidt@nicira.com>
Signed-off-by: Ben Pfaff <blp@nicira.com>
12 years agometa-flow: New functions for reading and writing generalized subfields.
Ben Pfaff [Wed, 11 Apr 2012 21:45:34 +0000 (14:45 -0700)]
meta-flow: New functions for reading and writing generalized subfields.

The existing functions for reading and writing the values of subfields only
handle subfields up to 64 bits wide.  These new functions handle subfields
of any width.

Also update some existing comments.

Signed-off-by: Ben Pfaff <blp@nicira.com>
12 years agoutil: New function bitwise_is_all_zeros().
Ben Pfaff [Sat, 14 Apr 2012 04:12:37 +0000 (21:12 -0700)]
util: New function bitwise_is_all_zeros().

Signed-off-by: Ben Pfaff <blp@nicira.com>
12 years agoutil: New function bitwise_one().
Ben Pfaff [Wed, 11 Apr 2012 19:07:51 +0000 (12:07 -0700)]
util: New function bitwise_one().

It's the obvious counterpart to bitwise_zero().

Signed-off-by: Ben Pfaff <blp@nicira.com>
12 years agobridge: Rate limit port creations and deletions.
Ethan Jackson [Sat, 7 Apr 2012 01:38:48 +0000 (18:38 -0700)]
bridge: Rate limit port creations and deletions.

In some datapaths, adding or deleting OpenFlow ports can take quite
a bit of time.  If there are lots of OpenFlow ports which needed to
be added in a run loop, this can cause Open vSwitch to lock up and
stop setting up flows while trying to catch up.  This patch lessons
the severity of the problem by only doing a few OpenFlow port adds
or deletions per run loop allowing other work to be done in
between.

Bug #10672.
Signed-off-by: Ethan Jackson <ethan@nicira.com>
12 years agobridge: Rate limit default address warnings.
Ethan Jackson [Wed, 11 Apr 2012 01:35:04 +0000 (18:35 -0700)]
bridge: Rate limit default address warnings.

This information is typically not more useful if displayed more
often.

Signed-off-by: Ethan Jackson <ethan@nicira.com>
12 years agoovsdb-idl: Improve documentation.
Ben Pfaff [Thu, 12 Apr 2012 15:27:56 +0000 (08:27 -0700)]
ovsdb-idl: Improve documentation.

Signed-off-by: Ben Pfaff <blp@nicira.com>
12 years agoovsdb-idl: Improve ovsdb_idl_txn_increment() interface.
Ben Pfaff [Thu, 12 Apr 2012 15:25:10 +0000 (08:25 -0700)]
ovsdb-idl: Improve ovsdb_idl_txn_increment() interface.

The previous interface was just bizarre.

Signed-off-by: Ben Pfaff <blp@nicira.com>
12 years agoovsdb-idl: Simplify transaction retry.
Ben Pfaff [Tue, 27 Mar 2012 17:16:52 +0000 (10:16 -0700)]
ovsdb-idl: Simplify transaction retry.

Originally the IDL transaction state machine had a return value
TXN_TRY_AGAIN to signal the client to wait for a change in the database and
then retry its transaction.  However, this logic was incomplete, because
it was possible for the database to change before the reply to the
transaction RPC was received, in which case the client would wait for a
further change.  Commit 4fdfe5ccf84c (ovsdb-idl: Prevent occasional hang
when multiple database clients race.) fixed the problem by breaking
TXN_TRY_AGAIN into two status codes, TXN_AGAIN_WAIT that meant to wait for
a further change and TXN_AGAIN_NOW that meant that a change had already
occurred so try again immediately.

This is correct enough, but it is more complicated than necessary.  It is
simpler and just as correct to use a single "try again" status that
requires the client to wait for a change relative to the database contents
*before* the transaction was committed.  This commit makes that change.
It also changes ovsdb_idl_run()'s return type from bool to void because
its return type is hardly useful anymore.

Signed-off-by: Ben Pfaff <blp@nicira.com>
12 years agoovs-monitor-ipsec: Use same pattern as ovs-xapi-sync.
Ben Pfaff [Thu, 8 Mar 2012 19:23:14 +0000 (11:23 -0800)]
ovs-monitor-ipsec: Use same pattern as ovs-xapi-sync.

This does not fix an actual bug in ovs-monitor-ipsec because it does not
ever commit any changes to the database.

Signed-off-by: Ben Pfaff <blp@nicira.com>
12 years agoPython binaries to write comment in db show-log while updating database
Arun Sharma [Mon, 9 Apr 2012 20:29:32 +0000 (13:29 -0700)]
Python binaries to write comment in db show-log while updating database

This is an improvement in python script ovs-xapi-sync to call add_comment()
method while updating database. This will help developer to debug binaries
and database from ovsdb-tool show-log command output and understand which
python binary is updating db.

Feature #10543
Signed-off-by: Arun Sharma <arun.sharma@calsoftinc.com>
Signed-off-by: Ben Pfaff <blp@nicira.com>
12 years agovswitchd: Remove port from datapath if it becomes non-operational
Ansis Atteka [Sat, 7 Apr 2012 00:52:56 +0000 (17:52 -0700)]
vswitchd: Remove port from datapath if it becomes non-operational

If kernel module rejects config changes then vswitchd sets the ofport
column to -1, but does not remove the non-operational port from the
datapath. This patch fixes this problem.

ovs-vsctl add-br ovsbr
ovs-vsctl add-port ovsbr p1
ovs-vsctl add-port ovsbr p2
ovs-vsctl set Interface p1 options:remote_ip=2.1.1.1 options:key=123 type=gre
ovs-vsctl set Interface p2 options:remote_ip=1.1.1.1 options:key=123 type=gre
ovs-vsctl set Interface p2 options:remote_ip=2.1.1.1 options:key=123 type=gre
ovs-dpctl show #observe that p2 does not appear here anymore

Signed-off-by: Ansis Atteka <aatteka@nicira.com>
12 years agodatapath: Do not send notification if ovs_vport_set_options() failed
Ansis Atteka [Sat, 7 Apr 2012 00:35:12 +0000 (17:35 -0700)]
datapath: Do not send notification if ovs_vport_set_options() failed

There is no need to send a notification if ovs_vport_set_options() failed
and ovs_vport_cmd_set() did not change anything.

Issue#10285

Signed-off-by: Ansis Atteka <aatteka@nicira.com>
12 years agoutil: New function set_program_name_version().
Ethan Jackson [Fri, 6 Apr 2012 18:47:51 +0000 (11:47 -0700)]
util: New function set_program_name_version().

With this function, users of the Open vSwitch libraries which
should not have the same version as Open vSwitch will have their
version displayed in unixctl and at the command line.

Signed-off-by: Ethan Jackson <ethan@nicira.com>
12 years agoAdded handling of previously ignored cfm faults.
Mehak Mahajan [Thu, 5 Apr 2012 00:36:00 +0000 (17:36 -0700)]
Added handling of previously ignored cfm faults.

The CFM packets that are out of sequence or contain invalid cfm_interval were
previously not ignored. The behavior is changed with this patch to not
process those CFM frames.

Signed-off-by: Mehak Mahajan <mmahajan@nicira.com>
12 years agodebian: Remove unused directory.
Gurucharan Shetty [Fri, 6 Apr 2012 15:20:36 +0000 (08:20 -0700)]
debian: Remove unused directory.

Installing the openvswitch-switch package creates a directory
'openvswitch-switch' in /etc. It is not used for any purpose.

Bug #10623.
Reported-by: Eric Lopez <elopez@nicira.com>
Signed-off-by: Gurucharan Shetty <gshetty@nicira.com>
12 years agoGranular link health statistics for cfm.
Mehak Mahajan [Thu, 5 Apr 2012 21:30:23 +0000 (14:30 -0700)]
Granular link health statistics for cfm.

The changes display the cfm_health of an interface.  The cfm_health
is an exponential weighted moving average of the health of all
remote_mpids.  The value can vary from 0 to 100, 100 being very healthy
and 0 being unhealthy.

Feature #10363
Requested-by: Ethan Jackson <ethan@nicira.com>
Signed-off-by: Mehak Mahajan <mmahajan@nicira.com>
12 years agoRevert "Granular link health statistics for cfm."
Mehak Mahajan [Thu, 5 Apr 2012 21:26:29 +0000 (14:26 -0700)]
Revert "Granular link health statistics for cfm."

Missed commiting one line of change.
This reverts commit c75b7e39d973cc9f4869c84d48eeb3b66afb2971.

Signed-off-by: Mehak Mahajan <mmahajan@nicira.com>
12 years agoGranular link health statistics for cfm.
Mehak Mahajan [Thu, 29 Mar 2012 21:34:51 +0000 (14:34 -0700)]
Granular link health statistics for cfm.

The changes display the cfm_health of an interface.  The cfm_health
is an exponential weighted moving average of the health of all
remote_mpids.  The value can vary from 0 to 100, 100 being very healthy
and 0 being unhealthy.

Feature #10363
Requested-by: Ethan Jackson <ethan@nicira.com>
Signed-off-by: Mehak Mahajan <mmahajan@nicira.com>
12 years agodatapath: tunnelling: capwap: UDP_ENCAP_CAPWAP is an abitrary value
Simon Horman [Wed, 4 Apr 2012 05:06:24 +0000 (14:06 +0900)]
datapath: tunnelling: capwap: UDP_ENCAP_CAPWAP is an abitrary value

It seems to me that as the value of UDP_ENCAP_CAPWAP is hardcoded
it would be more appropriate to describe it as arbitrary than
random.

Signed-off-by: Simon Horman <horms@verge.net.au>
Signed-off-by: Jesse Gross <jesse@nicira.com>
12 years agobugtool - Collect version information for all running Open vSwitch daemons.
Arun Sharma [Wed, 4 Apr 2012 20:23:19 +0000 (13:23 -0700)]
bugtool - Collect version information for all running Open vSwitch daemons.

This is an improvement in {ovs|xen}-bugtool archive output to determine the
version which was running for all the OVS daemons. It is implemented as a
plugin which internally calls "ovs-appctl -t <daemon> version" command for
daemons whose pid file is present in /var/run/openvswitch directory.

Feature #10383
Signed-off-by: Arun Sharma <arun.sharma@calsoftinc.com>
[blp@nicira.com updated debian/copyright.in]
Signed-off-by: Ben Pfaff <blp@nicira.com>
12 years agoAUTHORS: Add David S. Miller
Jesse Gross [Mon, 2 Apr 2012 20:48:24 +0000 (13:48 -0700)]
AUTHORS: Add David S. Miller

Signed-off-by: Jesse Gross <jesse@nicira.com>
12 years agodatapath: Stop using NLA_PUT*().
David S. Miller [Mon, 2 Apr 2012 20:25:21 +0000 (13:25 -0700)]
datapath: Stop using NLA_PUT*().

These macros contain a hidden goto, and are thus extremely error
prone and make code hard to audit.

Signed-off-by: David S. Miller <davem@davemloft.net>
[jesse: Additional transformations for code not upstream.]
Signed-off-by: Jesse Gross <jesse@nicira.com>
12 years agodatapath: Backport nla_put_be* functions.
Jesse Gross [Mon, 2 Apr 2012 20:41:20 +0000 (13:41 -0700)]
datapath: Backport nla_put_be* functions.

Linux 3.5 replaces the NLA_PUT_* functions with non-macro version,
which required adding the endian-typed versions.  This backports
those functions and drops the macro backports.

Signed-off-by: Jesse Gross <jesse@nicira.com>
12 years agounixctl.py: Allow callers to manually set unixctl version.
Ethan Jackson [Sat, 31 Mar 2012 06:11:05 +0000 (23:11 -0700)]
unixctl.py: Allow callers to manually set unixctl version.

Some clients of unixctl.py may want to use a different version than
the one supplied in ovs.version.VERSION.  This patch supports an
optional manual override of this value.

Feature #10383.
Signed-off-by: Ethan Jackson <ethan@nicira.com>
12 years agocheck-structs: Disallow uint<N>_t because ovs_be<N> should always be used.
Ben Pfaff [Mon, 2 Apr 2012 16:06:11 +0000 (09:06 -0700)]
check-structs: Disallow uint<N>_t because ovs_be<N> should always be used.

The header files that check-structs checks should only contain big-endian
data, never native-endian data, so disallow uint<N>_t entirely.  (We had
a couple of mistakes in this area until recently.)

uint8_t is an obvious exception.

Reported-by: Simon Horman <horms@verge.net.au>
Reviewed-by: Simon Horman <horms@verge.net.au>
Signed-off-by: Ben Pfaff <blp@nicira.com>
12 years agoDefinitions for Open Flow 1.2
Simon Horman [Fri, 30 Mar 2012 02:30:01 +0000 (11:30 +0900)]
Definitions for Open Flow 1.2

This is a first pass at adding include/openflow/openflow-1.2.h to
include enum and struct definitions for Open Flow 1.2 that
are not already covered by Open Flow 1.1.

Signed-off-by: Simon Horman <horms@verge.net.au>
Signed-off-by: Ben Pfaff <blp@nicira.com>
12 years agoAdd some missing Open Flow 1.1 definitions
Simon Horman [Fri, 30 Mar 2012 02:30:00 +0000 (11:30 +0900)]
Add some missing Open Flow 1.1 definitions

Signed-off-by: Simon Horman <horms@verge.net.au>
[blp@nicira.com added OFPRR_GROUP_DELETE to
 ofp_flow_removed_reason_to_string()]
Signed-off-by: Ben Pfaff <blp@nicira.com>
12 years agoAdd some missing Open Flow 1.0 definitions
Simon Horman [Fri, 30 Mar 2012 02:29:59 +0000 (11:29 +0900)]
Add some missing Open Flow 1.0 definitions

Signed-off-by: Simon Horman <horms@verge.net.au>
Signed-off-by: Ben Pfaff <blp@nicira.com>
12 years agoMove structures into openflow-common.h
Simon Horman [Fri, 30 Mar 2012 02:29:58 +0000 (11:29 +0900)]
Move structures into openflow-common.h

* Move ofp_packet_in_reason into openflow-common.h as it
  contains elements first defined in both the
  Openflow 1.0 and 1.2 specifications.

* Move ofp_flow_removed_reason into openflow-common.h as
  subsequent patches will add an element first defined in the
  Openflow 1.1 specification.

Signed-off-by: Simon Horman <horms@verge.net.au>
Signed-off-by: Ben Pfaff <blp@nicira.com>
12 years agoUse ovs_be16 in ofp_queue_prop_min_rate
Simon Horman [Fri, 30 Mar 2012 02:29:57 +0000 (11:29 +0900)]
Use ovs_be16 in ofp_queue_prop_min_rate

Use ovs_be16 in place of uint16_t.
This corrects an error made by me when
adding ofp_queue_prop_min_rate.

Signed-off-by: Simon Horman <horms@verge.net.au>
Signed-off-by: Ben Pfaff <blp@nicira.com>
12 years agoUse ovs_be32 in nx_flow_mod_table_id
Simon Horman [Fri, 30 Mar 2012 02:29:56 +0000 (11:29 +0900)]
Use ovs_be32 in nx_flow_mod_table_id

This corrects what appears to be an oversight
whereby uing32_t is used while all other structures
in the file make use of ovs_be* in place of uint*_t.

Signed-off-by: Simon Horman <horms@verge.net.au>
Signed-off-by: Ben Pfaff <blp@nicira.com>
12 years agoRearrange structures to better fit valgrind's memory leak heuristics.
Ben Pfaff [Wed, 28 Mar 2012 21:13:02 +0000 (14:13 -0700)]
Rearrange structures to better fit valgrind's memory leak heuristics.

valgrind's memory leak detector considers a pointer to the head of a memory
block to be "definitely" a pointer to that memory block but a pointer to
the interior of a memory block only "possibly" a pointer to that memory
block.  Open vSwitch hmap_node and list data structures can go anywhere
inside a structure; if they are in the middle of a structure then valgrind
considers pointers to them to be possible leaks.  Therefore, this commit
moves some of these from the middle of data structures to the head, to
reduce valgrind's uncertainty.

Signed-off-by: Ben Pfaff <blp@nicira.com>
12 years agotests: Gracefully shut down ovsdb-server, to avoid valgrind leak warnings.
Ben Pfaff [Wed, 28 Mar 2012 21:06:54 +0000 (14:06 -0700)]
tests: Gracefully shut down ovsdb-server, to avoid valgrind leak warnings.

When ovsdb-server exits gracefully it frees all the memory that it owns,
which keeps valgrind from warning about "possibly lost" blocks.

Signed-off-by: Ben Pfaff <blp@nicira.com>
12 years agotests: Suppress valgrind error reports from glibc timer_create() function.
Ben Pfaff [Wed, 28 Mar 2012 00:20:48 +0000 (17:20 -0700)]
tests: Suppress valgrind error reports from glibc timer_create() function.

valgrind reports a memory leak and a parameter error in glibc's
timer_create() in almost every program execution.  I don't know a way to
avoid them.

Signed-off-by: Ben Pfaff <blp@nicira.com>
12 years agoFix minor memory leaks found by valgrind.
Ben Pfaff [Tue, 27 Mar 2012 21:09:14 +0000 (14:09 -0700)]
Fix minor memory leaks found by valgrind.

All of these leaks are in normally short-lived programs, so none of them
is very important.

Signed-off-by: Ben Pfaff <blp@nicira.com>
12 years agoovsdb-idl: Fix memory leak writing synthetic rows in ovsdb_idl_txn_write().
Ben Pfaff [Wed, 28 Mar 2012 21:02:57 +0000 (14:02 -0700)]
ovsdb-idl: Fix memory leak writing synthetic rows in ovsdb_idl_txn_write().

This could cause a slow but steady memory leak in ovs-vswitchd.

Found by valgrind.

Signed-off-by: Ben Pfaff <blp@nicira.com>
12 years agoAdd some missing common Open Flow definitions
Simon Horman [Wed, 28 Mar 2012 00:44:14 +0000 (09:44 +0900)]
Add some missing common Open Flow definitions

Signed-off-by: Simon Horman <horms@verge.net.au>
Signed-off-by: Ben Pfaff <blp@nicira.com>
12 years agoAvoid possibly including an old vswitch-idl.h.
Ben Pfaff [Tue, 27 Mar 2012 22:57:52 +0000 (15:57 -0700)]
Avoid possibly including an old vswitch-idl.h.

Codes that uses #include "vswitch-idl.h" can get an older version of this
header, because this header file moved from vswitchd/ to lib/ and the
older generated file might still be present.

This helps out two ways:

     * "make clean" will delete the generated files from their old
       locations.

     * Use #include "lib/vswitch-idl.h" to explicitly avoid including the
       files from their old locations.

Reported-by: Justin Pettit <jpettit@nicira.com>
Signed-off-by: Ben Pfaff <blp@nicira.com>
12 years agodpif-netdev: Correct type of struct dp_netdev_flow's 'tcp_flags' member.
Ben Pfaff [Tue, 27 Mar 2012 17:20:56 +0000 (10:20 -0700)]
dpif-netdev: Correct type of struct dp_netdev_flow's 'tcp_flags' member.

TCP flags fit in 8 bits so this type seems more appropriate.

Fixes the following "sparse" warnings introduced by commit 734ec5ec1349
(packet: Add additional TCP flags extraction on IPv6.):

dpif-netdev.c:630: warning: incorrect type in assignment (different base types)
dpif-netdev.c:630:    expected unsigned char [unsigned] [usertype] tcp_flags
dpif-netdev.c:630:    got restricted __be16 [usertype] tcp_flags
dpif-netdev.c:979: warning: invalid assignment: |=
dpif-netdev.c:979:    left side has type restricted __be16
dpif-netdev.c:979:    right side has type unsigned char

CC: Jesse Gross <jesse@nicira.com>
Signed-off-by: Ben Pfaff <blp@nicira.com>
12 years agoAdd error codes for Open Flow v1.2
Simon Horman [Mon, 26 Mar 2012 20:46:35 +0000 (13:46 -0700)]
Add error codes for Open Flow v1.2

* Where Open Flow 1.2 breaks apart error codes defined
  in previous versions, provide all new definitions to
  previous versions and map the numeric error code to
  the first first definition supplied in ofp-errors.h.
  The case handled so far is:
  OFPERR_OFPBIC_BAD_EXP_TYPE -> { OFPERR_OFPBIC_BAD_EXPERIMENTER,
                                  OFPERR_OFPBIC_BAD_EXP_TYPE }

* Where Open Flow 1.2 adds error codes that were previously
  defined as Nicira extension errors define the later in terms
  of the new codes.

Signed-off-by: Simon Horman <horms@verge.net.au>
[blp@nicira.com added better error checking in extract-ofp-errors, added
 unit tests, miscellaneous cleanup]
Signed-off-by: Ben Pfaff <blp@nicira.com>
Acked-by: Simon Horman <horms@verge.net.au>
12 years agoofp-errors: Use OF1.1+ in place of OF1.1 throughout.
Ben Pfaff [Mon, 26 Mar 2012 18:25:52 +0000 (11:25 -0700)]
ofp-errors: Use OF1.1+ in place of OF1.1 throughout.

In general, I guess that the common case is for most error codes to be
retained without change in future versions of OpenFlow, so to me it seems
best to use the "+" version "by default".

Signed-off-by: Ben Pfaff <blp@nicira.com>
Acked-by: Simon Horman <horms@verge.net.au>
12 years agoofp-errors: Rename "OF" to "OF1.0+", "NX" to "NX1.0+".
Ben Pfaff [Mon, 26 Mar 2012 18:22:42 +0000 (11:22 -0700)]
ofp-errors: Rename "OF" to "OF1.0+", "NX" to "NX1.0+".

This seems like a more reasonable way to do things given that we will soon
need "OpenFlow 1.1 and later" and "OpenFlow 1.1 only" keywords.

Signed-off-by: Ben Pfaff <blp@nicira.com>
Reviewed-by: Simon Horman <horms@verge.net.au>
12 years agoextract-ofp-errors: Fix error message.
Ben Pfaff [Mon, 26 Mar 2012 19:39:12 +0000 (12:39 -0700)]
extract-ofp-errors: Fix error message.

The error message should mention the keyword that caused the error, not
some other random keyword from previous iterations.

Signed-off-by: Ben Pfaff <blp@nicira.com>
Reviewed-by: Simon Horman <horms@verge.net.au>
12 years agopacket: Add additional TCP flags extraction on IPv6.
Jesse Gross [Mon, 26 Mar 2012 19:56:14 +0000 (12:56 -0700)]
packet: Add additional TCP flags extraction on IPv6.

Commit 11460e2316b88f0bd0ea0005d94338d800ea16bd
(flow: Enable retrieval of TCP flags from IPv6 traffic.) updated
one of the TCP flags extraction functions in userspace but missed
the other.  This updates that function and converts the other to
use it to reduce duplication.

Bug #10194

Signed-off-by: Jesse Gross <jesse@nicira.com>
12 years agoovs-bugtool: Add ability to prioritize files by date.
Raju Subramanian [Fri, 23 Mar 2012 22:49:32 +0000 (15:49 -0700)]
ovs-bugtool: Add ability to prioritize files by date.

When size limit is reached in the middle of processing a dir,
the report ends up containing oldest files. This change adds
an optional param in the plugin to prioritize newer files.

Feature #9937
Requested-by: Ronald Lee <rlee@nicira.com>
Signed-off-by: Raju Subramanian <rsubramanian@nicira.com>
Signed-off-by: Ben Pfaff <blp@nicira.com>