Jarno Rajahalme [Tue, 25 Aug 2015 20:55:03 +0000 (13:55 -0700)]
classifier: Retire partitions.
Classifier partitions allowed skipping subtables when if was known
from the flow's metadata field that the subtable cannot possibly
match. This functionality was later implemented in a more general
fashion by staged lookup, where the first stage also covers the
metadata field, among the rest of the non-packet fields in the struct
flow. While in theory skipping a subtable on the basis of the
metadata field alone could produce more effective wildcards, on the
basis of our testsuite coverage it does not seem to be the case, as
removing the partitioning feature did not result in any test failures.
Removing the partitioning feature makes classifier lookups roughly 20%
faster when a wildcard mask is not needed, and roughly 10% faster when
a wildcard mask is needed, as tested with the test-classifier
benchmark with one lookup thread.
Found by profiling with 'perf'.
Signed-off-by: Jarno Rajahalme <jrajahalme@nicira.com> Acked-by: Ben Pfaff <blp@nicira.com>
Jarno Rajahalme [Tue, 25 Aug 2015 20:55:03 +0000 (13:55 -0700)]
flow: Add struct flowmap.
Struct miniflow is now sometimes used just as a map. Define a new
struct flowmap for that purpose. The flowmap is defined as an array of
maps, and it is automatically sized according to the size of struct
flow, so it will be easier to maintain in the future.
It would have been tempting to use the existing struct bitmap for this
purpose. The main reason this is not feasible at the moment is that
some flowmap algorithms are simpler when it can be assumed that no
struct flow member requires more bits than can fit to a single map
unit. The tunnel member already requires more than 32 bits, so the map
unit needs to be 64 bits wide.
Performance critical algorithms enumerate the flowmap array units
explicitly, as it is easier for the compiler to optimize, compared to
the normal iterator. Without this optimization a classifier lookup
without wildcard masks would be about 25% slower.
With this more general (and maintainable) algorithm the classifier
lookups are about 5% slower, when the struct flow actually becomes big
enough to require a second map. This negates the performance gained
in the "Pre-compute stage masks" patch earlier in the series.
Requested-by: Ben Pfaff <blp@nicira.com> Signed-off-by: Jarno Rajahalme <jrajahalme@nicira.com> Acked-by: Ben Pfaff <blp@nicira.com>
Alex Wang [Thu, 13 Aug 2015 18:31:49 +0000 (11:31 -0700)]
coverage: Add coverage_try_clear() for performance-critical threads.
For performance-critical threads like pmd threads, we currently make them
never call coverage_clear() to avoid contention over the global mutex
'coverage_mutex'. So, even though pmd thread still keeps updating their
thread-local coverage count, the count is never attributed to the global
total. But it is useful to have them available.
This commit makes this happen by implementing a non-contending version
of the clear function, coverage_try_clear(). The function will use
the ovs_mutex_trylock() and return immediately if the mutex cannot
be acquired. Since threads like pmd thread are always busy-looping,
the lock will eventually be acquired.
Requested-by: Ilya Maximets <i.maximets@samsung.com> Signed-off-by: Alex Wang <alexw@nicira.com> Acked-by: Ilya Maximets <i.maximets@samsung.com> Acked-by: Ben Pfaff <blp@nicira.com> Acked-by: Daniele Di Proietto <diproiettod@vmware.com
Justin Pettit [Mon, 24 Aug 2015 22:39:37 +0000 (15:39 -0700)]
ofp-parse: Add "ipv4" and "ip4" shorthand notations.
This will allow ovs-ofctl to accept them on input and provides some
symmetry with the IPv6 versions. Documentation was not added, since
output will always be "ip".
Signed-off-by: Justin Pettit <jpettit@nicira.com> Acked-by: Russell Bryant <rbryant@redhat.com>
Justin Pettit [Mon, 24 Aug 2015 22:22:12 +0000 (15:22 -0700)]
ovs-ofctl: Document missing shorthand notations.
Some new shorthands were added to ovs-ofctl, but not documented in the
man page. This adds most of them. A couple were left out that would
print differently than they would be input. For example, we accept
"ip6" and "ipv6", but we would output it as "ipv6", so "ip6" is not
documented.
Signed-off-by: Justin Pettit <jpettit@nicira.com> Acked-by: Russell Bryant <rbryant@redhat.com>
Wenyu Zhang [Tue, 25 Aug 2015 03:56:44 +0000 (20:56 -0700)]
datapath: Make 100 percents packets sampled when sampling rate is 1.
When sampling rate is 1, the sampling probability is UINT32_MAX. The packet
should be sampled even the prandom32() generate the number of UINT32_MAX.
And none packet need be sampled when the probability is 0.
Signed-off-by: Wenyu Zhang <wenyuz@vmware.com> Acked-by: Pravin B Shelar <pshelar@nicira.com> Signed-off-by: David S. Miller <davem@davemloft.net>
Upstream: e05176a3283 ("openvswitch: Make 100 percents packets
sampled when sampling rate is 1.")
Signed-off-by: Pravin B Shelar <pshelar@nicira.com>
Joe Stringer [Wed, 5 Aug 2015 20:31:25 +0000 (13:31 -0700)]
ofp-print: Fix redundant expression.
'features' is always a valid pointer to something on the stack, so
checking (!features || ...) is the same as checking (false || ...).
Simplify the expression.
Found by MIT STACK undefined behaviour checker.
Signed-off-by: Joe Stringer <joestringer@nicira.com> Acked-by: Ben Pfaff <blp@nicira.com>
Alex Wang [Sat, 22 Aug 2015 16:26:54 +0000 (09:26 -0700)]
ovn-controller-vtep.at: Remove all 'Broken pipe' warning logs from
ovsdb-server.log.
This commit makes the OVN_CONTROLLER_VTEP_STOP remove all 'Broken pipe'
warning logs from ovsdb-server.log before running 'check_log'. This is in
that *ctl command (e.g. ovn-nbctl) exits right after committing the change
to database; however, in reaction, some daemon (e.g. ovn-controller-vtep)
may immediately update the database; this later update may cause database
sending update back to *ctl command if *ctl has not proceeded to exit yet;
and if *ctl command exits before database calling send, the send will fail
with 'Broken pipe' error.
Signed-off-by: Alex Wang <alexw@nicira.com> Acked-by: Russell Bryant <rbryant@redhat.com>
Ben Pfaff [Fri, 21 Aug 2015 00:07:40 +0000 (17:07 -0700)]
ofp-util: For OF1.0, don't wildcard PCP field when 802.1Q header absent.
OpenFlow 1.0.1 says:
The dl_vlan_pcp field must be ignored when the OFPFW_DL_VLAN wildcard
bit is set or when the dl_vlan value is set to OFP_VLAN_NONE. Fields
that are ignored don’t need to be wildcarded and should be set to 0.
Previously, OVS wildcarded the PCP field when dl_vlan was OFP_VLAN_NONE,
but this commit changes the behavior to that suggested above: the PCP
field should not be wildcarded (and should be set to 0, but the code
already did that).
This commit only changes the translation from OVS's internal flow format
to the OpenFlow 1.0 wire format. Translation in the other direction and
to other formats is unaffected.
Found by OFTest.
Signed-off-by: Ben Pfaff <blp@nicira.com> Acked-by: Jarno Rajahalme <jrajahalme@nicira.com>
In OVN, ovsdb-server is the daemon that manages the databases
and can be called as the central controller. So it would be
nice for ovsdb-server to be able to push its self-signed
certificate to all the other nodes where ovn-controller runs.
Signed-off-by: Gurucharan Shetty <gshetty@nicira.com> Acked-by: Ben Pfaff <blp@nicira.com>
Russell Bryant [Fri, 21 Aug 2015 17:42:18 +0000 (13:42 -0400)]
ovs-sandbox: Add support for ovn-controller-vtep.
When ovs-sandbox is run with ovn enabled, create the vtep database and
run ovn-controller-vtep. This lets you do some basic testing with
ovn-controller-vtep. For example:
$ make sandbox SANDBOXFLAGS="--ovn"
$ vtep-ctl add-ps ps0
After those commands, you can see that ovn-controller-vtep added a
Chassis row to OVN_Southbound for the physical switch.
Signed-off-by: Russell Bryant <rbryant@redhat.com> Signed-off-by: Ben Pfaff <blp@nicira.com>
Jesse Gross [Wed, 12 Aug 2015 01:46:14 +0000 (18:46 -0700)]
tun-metadata: Always set option present when copying data.
Whenever we write into a tunnel option field, we also need to mark
it as significant. If we don't, then the data will later be ignored.
We currently do this in every case except for flow metadata. This causes
us to not correctly serialize the tunnel metadata for Packet Ins to the
controller.
Rather than separately writing the data and marking the options as present,
it is better to combine the two steps to ensure that one can never be
done without the other.
Signed-off-by: Jesse Gross <jesse@nicira.com> Acked-by: Ben Pfaff <blp@nicira.com>
Russell Bryant [Tue, 18 Aug 2015 18:43:18 +0000 (11:43 -0700)]
hmap: Ensure iterator is NULL after iteration.
The HMAP_FOR_EACH_()* macros had a usability issue where the iterator
was only NULL at the completion of iteration if the hmap_node was the
first struct member. This change ensures that the iterator is set to
NULL when iteration ends normally without a 'break'.
Signed-off-by: Russell Bryant <rbryant@redhat.com> Signed-off-by: Ben Pfaff <blp@nicira.com>
<n_rules> - The number of rules to install for lookups. More rules
makes misses less likely.
<n_priorities> - How many different priorities to use. Using only 1
priority will force lookups to continue through all
subtables.
<n_subtables> - Number of subtables to use. Normally a classifier has
rules with different kinds of masks, resulting in
multiple subtables (one per mask). However, in some
special cases a table may consist of only one kind of
rules, so there will be only one subtable.
<n_threads> - How many lookup threads to use. Using one thread should
give less variance accross runs, but classifier
scaling can be tested with multiple threads.
<n_lookups> - How many lookups each thread should perform.
For testing the classifier is filled with <n_rules> rules using
<n_subtables> different mask patterns and <n_priorities> different
priorities. A random set of lookup flows are created, and <n_threads>
lookup threads are spawned to perform <n_lookups> lookups each. The
count of hits and misses, as well as the overall execution time is
reported.
Simon Horman [Thu, 20 Aug 2015 23:45:34 +0000 (16:45 -0700)]
ovs-ctl: do not attempt to restore flows when called with --delete-bridges
When called with --delete-bridges saved flows cannot be restored as the
bridges to which they belong no longer exist. This results in the following
error messages on restart.
ovs-ofctl: br0 is not a bridge or a socket
Restoring saved flows ... failed!
Although there is no effect of this error other than the message
it seems worth avoiding. This patch does so by skipping saving of flows
when --delete-bridges is in effect.
As flows are no longer saved when --delete-bridges is in effect
a side-effect of this change is that restart may be faster when
there are many flows.
Signed-off-by: Simon Horman <simon.horman@netronome.com> Signed-off-by: Gurucharan Shetty <gshetty@nicira.com> Acked-by: Ben Pfaff <blp@nicira.com>
Pravin B Shelar [Tue, 18 Aug 2015 19:18:47 +0000 (12:18 -0700)]
tunneling: Remove gre64 tunnel support.
GRE64 was introduced to extend gre key from 32-bit to 64-bit using
gre-key and sequence number field. But GRE64 is not standard
protocol. There are not many users of this protocol. Therefore we
have decided to remove it.
Signed-off-by: Pravin B Shelar <pshelar@nicira.com> Acked-by: Jesse Gross <jesse@nicira.com>
Ben Pfaff [Mon, 25 May 2015 22:35:53 +0000 (15:35 -0700)]
ofp-util: Use correct error for indirect group with more than one bucket.
OpenFlow 1.5 says:
If the group-mod request specifies more than one bucket for a group of
type Indirect, the switch must refuse to add the group entry and must
send an ofp_error_msg with OFPET_GROUP_MOD_FAILED type and
OFPGMFC_INVALID_GROUP code.
Older versions don't specify a particular error for this case, so we might
as well use it for older OpenFlow also.
Found by OFTest.
Signed-off-by: Ben Pfaff <blp@nicira.com> Acked-by: Jarno Rajahalme <jrajahalme@nicira.com>
This test is skipped if the 'ip' command cannot interpret the vxlan 'dstport'
option; this is used as a proxy for detecting native kernel support for this
tunnel type.
Signed-off-by: Joe Stringer <joestringer@nicira.com> Acked-by: Daniele Di Proietto <diproiettod@vmware.com>
Joe Stringer [Wed, 12 Aug 2015 21:01:29 +0000 (14:01 -0700)]
system-macros: Don't explicitly remove bridge.
Depending on the kernel in use, manually removing a bridge during
OVS_VSWITCHD_STOP can cause the kernel to send a route update that
refers to the bridge device which is being deleted. OVS can't make sense
of these messages, resulting in logs like the following:
route_table|DBG|Could not find interface name[702]: No such device or
address
netlink_notifier|WARN|received bad netlink message
One such example kernel is the Ubuntu 3.16.0-33-generic linux package.
The bridge doesn't need to be explicitly removed, as the OVSDB
configuration for each test will be cleared before executing the next.
Drop this extra step.
Signed-off-by: Joe Stringer <joestringer@nicira.com> Acked-by: Daniele Di Proietto <diproiettod@vmware.com>
Joe Stringer [Wed, 12 Aug 2015 21:01:28 +0000 (14:01 -0700)]
system-macros: Create ADD_BR variant.
This patch splits ADD_BR into two commands, so they can be used from
different contexts:
ADD_BR(...) is a standalone command to add a bridge to OVS, and allows
additional ovs-vsctl arguments to be passed. It uses _ADD_BR().
_ADD_BR(...) is the implementation-specific ovs-vsctl arguments to
set up the correct datapath type for userspace or kmod tests.
Signed-off-by: Joe Stringer <joestringer@nicira.com> Acked-by: Daniele Di Proietto <diproiettod@vmware.com>
Joe Stringer [Wed, 12 Aug 2015 21:01:26 +0000 (14:01 -0700)]
system-common-macros: Allow quotes in NS_EXEC().
This allows arbitrary commands to be passed into the NS_EXEC macro to be
executed within a namespace, including commands that have quotes and
commands chained together.
Signed-off-by: Joe Stringer <joestringer@nicira.com> Acked-by: Daniele Di Proietto <diproiettod@vmware.com>
A unit test for ovn-northd was hanging because ovn-northd
running on Windows failed to detach. The detach function
for Windows is implemented inside "service_start()".
Also handle Windows services.
Reported-by: Alex Wang <alexw@nicira.com> Signed-off-by: Gurucharan Shetty <gshetty@nicira.com> Acked-by: Alex Wang <alexw@nicira.com>
Glenn Griffin [Mon, 17 Aug 2015 14:50:51 +0000 (07:50 -0700)]
datapath: Fix L4 checksum handling when dealing with IP fragments
openvswitch modifies the L4 checksum of a packet when modifying
the ip address. When an IP packet is fragmented only the first
fragment contains an L4 header and checksum. Prior to this change
openvswitch would modify all fragments, modifying application data
in non-first fragments, causing checksum failures in the
reassembled packet.
Signed-off-by: Glenn Griffin <ggriffin.kernel@gmail.com> Acked-by: Pravin B Shelar <pshelar@nicira.com> Signed-off-by: David S. Miller <davem@davemloft.net>
Upstream: 3576fd794b3 ("openvswitch: Fix L4 checksum handling when
dealing with IP fragments").
Signed-off-by: Pravin B Shelar <pshelar@nicira.com>
Chris J Arges [Mon, 17 Aug 2015 14:40:43 +0000 (07:40 -0700)]
datapath: allocate nr_node_ids flow_stats instead of num_possible_nodes
Some architectures like POWER can have a NUMA node_possible_map that
contains sparse entries. This causes memory corruption with openvswitch
since it allocates flow_cache with a multiple of num_possible_nodes() and
assumes the node variable returned by for_each_node will index into
flow->stats[node].
Use nr_node_ids to allocate a maximal sparse array instead of
num_possible_nodes().
The crash was noticed after 3af229f2 was applied as it changed the
node_possible_map to match node_online_map on boot. Fixes: 3af229f2071f5b5cb31664be6109561fbe19c861 Signed-off-by: Chris J Arges <chris.j.arges@canonical.com> Acked-by: Pravin B Shelar <pshelar@nicira.com> Acked-by: Nishanth Aravamudan <nacc@linux.vnet.ibm.com> Signed-off-by: David S. Miller <davem@davemloft.net>
Upstream: bac541e4631 ("openvswitch: allocate nr_node_ids flow_stats
instead of num_possible_nodes")
Signed-off-by: Pravin B Shelar <pshelar@nicira.com>
Jiri Benc [Mon, 17 Aug 2015 14:38:01 +0000 (07:38 -0700)]
datapath: disable LRO
Currently, openvswitch tries to disable LRO from the user space. This does
not work correctly when the device added is a vlan interface, though.
Instead of dealing with possibly complex stacked cross name space relations
in the user space, do the same as bridging does and call dev_disable_lro in
the kernel.
Signed-off-by: Jiri Benc <jbenc@redhat.com> Acked-by: Flavio Leitner <fbl@redhat.com> Acked-by: Pravin B Shelar <pshelar@nicira.com> Signed-off-by: David S. Miller <davem@davemloft.net>
Upstream: 640b2b107c ("openvswitch: disable LRO").
Signed-off-by: Pravin B Shelar <pshelar@nicira.com>
Alex Wang [Sat, 4 Jul 2015 05:55:24 +0000 (22:55 -0700)]
ovn-controller-vtep: Add binding module.
This commit adds the binding module to ovn-controller-vtep. The
module will scan through the Port_Binding table in ovnsb. If there is
a port binding entry for a logical switch on the vtep gateway chassis's
"vtep_logical_switches", sets the port binding's chassis column to the
vtep gateway chassis.
Signed-off-by: Alex Wang <alexw@nicira.com> Acked-by: Russell Bryant <rbryant@redhat.com>
Alex Wang [Sun, 16 Aug 2015 07:06:24 +0000 (00:06 -0700)]
ovn-controller-vtep.at: Fix intermittent test failure.
When testing the recreation of 'chassis' table entry by 'ovn-controller-
vtep'. The removal of 'chassis' table entry by the 'ovn-sbctl' could
cause 'Broken pipe' warning in ovsdb-server.log. This is due to the
race between 'ovn-sbctl' exiting and 'ovn-controller-vtep' adding
the chassis back. So, if the 'ovn-sbctl' exits right when the
ovsdb-server tries to send update of readd of the deleted 'chassis',
the sending will fail with 'Broken pipe' error.
Signed-off-by: Alex Wang <alexw@nicira.com> Acked-by: Russell Bryant <rbryant@redhat.com>
Alex Wang [Sun, 16 Aug 2015 06:44:45 +0000 (23:44 -0700)]
ovn-controller-vtep.at: Fix intermittent test failure.
The test waits until grep no vlan '200' from the VTEP 'vlan_binding'
column. However, string '200' could also appear in other 'vlan_binding'
entry's uuid value. Instead, we should grep for '200='.
Signed-off-by: Alex Wang <alexw@nicira.com> Acked-by: Russell Bryant <rbryant@redhat.com>
Flavio Leitner [Thu, 13 Aug 2015 19:06:29 +0000 (16:06 -0300)]
rhel: add installed but not packaged OVN tools
This patch adds the following to OVN %files:
/usr/bin/ovn-controller-vtep
/usr/bin/ovn-sbctl
/usr/share/man/man8/ovn-controller-vtep.8.gz
/usr/share/man/man8/ovn-sbctl.8.gz
Simon Horman [Thu, 13 Aug 2015 07:55:32 +0000 (16:55 +0900)]
flow: Ignore invalid ICMPv6 fields when parsing packets
There is a miss-match between the handling of invalid ICMPv6 fields in the
implementations of parse_icmpv6() in user-space and in the kernel datapath.
This patch addresses that by modifying the user-space implementation to
match that of the kernel datapath; processing is terminated without
rather than with an error and partial information is cleared.
With these changes the user-space implementation of parse_icmpv6()
never returns an error. Accordingly the return type and caller have been
updated.
The original motivation for this is to allow matching the ICMPv6 type and
code of packets with invalid neighbour discovery options although only the
change around the '(!opt_len || opt_len > *sizep)' conditional is necessary
to achieve that goal.
Signed-off-by: Simon Horman <simon.horman@netronome.com> Signed-off-by: Jesse Gross <jesse@nicira.com>
Ethan Jackson [Tue, 4 Aug 2015 01:43:53 +0000 (18:43 -0700)]
ofproto: Allow in-place modifications of datapath flows.
There are certain use cases (such as bond rebalancing) where a
datapath flow's actions may change, while it's wildcard pattern
remains the same. Before this patch, revalidators would note the
change, delete the flow, and wait for the handlers to install an
updated version. This is inefficient, as many packets could get
punted to userspace before the new flow is finally installed.
To improve the situation, this patch implements in place modification
of datapath flows. If the revalidators detect the only change to a
given ukey is its actions, instead of deleting it, it does a put with
the MODIFY flag set.
Signed-off-by: Ethan J. Jackson <ethan@nicira.com> Acked-by: Jarno Rajahalme <jrajahalme@nicira.com>
Ethan Jackson [Wed, 12 Aug 2015 21:50:54 +0000 (14:50 -0700)]
ofproto-dpif-upcall: Make ukey actions modifiable with RCU.
Future patches will need to modify ukey actions in some instances.
This patch makes this possible by protecting them with RCU. It also
adds thread safety checks to enforce the new protection mechanism.
Signed-off-by: Ethan J. Jackson <ethan@nicira.com> Acked-by: Jarno Rajahalme <jrajahalme@nicira.com>
Alex Wang [Thu, 13 Aug 2015 04:29:06 +0000 (21:29 -0700)]
db-ctl-base: Allow print rows that weak reference to table in
'cmd_show_table'.
Sometimes, it is desirable to print the table with weak reference to
the table specified in 'struct cmd_show_table'. For example the
Port_Binding table rows in OVN_Southbound database that refer to the
same Chassis table row can be printed under the same chassis entry
in 'ovn-sbctl show' output.
To achieve it, this commit adds a new struct in 'struct cmd_show_table'
that allows users to print a table with weak reference to 'table'
specified in 'struct cmd_show_table'. The 'ovn-sbctl' which now prints
the Port_Binding entries with Chassis table, is the first user of this
new feature.
Requested-by: Justin Pettit <jpettit@nicira.com> Signed-off-by: Alex Wang <alexw@nicira.com> Acked-by: Justin Pettit <jpettit@nicira.com>
Jarno Rajahalme [Thu, 13 Aug 2015 00:03:07 +0000 (17:03 -0700)]
classifier: Do not use mf_value.
mf_value has grown bigger than needed for storing the biggest
supported prefix (IPv6 address length). Define a new type to be used
instead of mf_value.
This makes classifier lookups a bit faster.
Signed-off-by: Jarno Rajahalme <jrajahalme@nicira.com> Acked-by: Joe Stringer <joestringer@nicira.com>
Russell Bryant [Wed, 12 Aug 2015 23:53:09 +0000 (16:53 -0700)]
ovn: Add lflow-list to ovn-sbctl.
I frequently view the contents of the Logical_Flow table while working
on OVN. Add a command that can output the contents of this table in a
sorted way that makes it easier to read through. It's sorted by
logical datapath, pipeline, table id, priority, and match.
Signed-off-by: Russell Bryant <rbryant@redhat.com> Acked-by: Alex Wang <alexw@nicira.com>
Joe Stringer [Fri, 7 Aug 2015 21:00:20 +0000 (14:00 -0700)]
system-kmod-macros: Fix VSWITCHD_STOP.
This was renamed. Surprisingly, the tests still pass without this,
however the extra checks that this command performs were not executed.
Fix the macro definition.
Signed-off-by: Joe Stringer <joestringer@nicira.com> Acked-by: Daniele Di Proietto <diproiettod@vmware.com>
Alex Wang [Sun, 9 Aug 2015 07:02:47 +0000 (00:02 -0700)]
ovn-controller-vtep: Add gateway module.
This commit adds the gateway module to ovn-controller-vtep. The
module will register the physical switches to ovnsb as chassis and
constantly update the "vtep_logical_switches" column in Chassis table.
Limitation (Recorded in TODO file):
- Do not support reading multiple tunnel ips of physical switch.
Signed-off-by: Alex Wang <alexw@nicira.com> Acked-by: Russell Bryant <rbryant@redhat.com>
Alex Wang [Fri, 7 Aug 2015 22:55:32 +0000 (15:55 -0700)]
ovn-sb: Remove the "Gateway" table from the ovn-sb schema.
In a gateway like the VTEP L2 gateway, physical vlans belonging to
the same logical network form a "logical switch". Each logical switch
has a dedicated tunnel key and will keep records of all MACs learned
from the owned vlans. So user can just send packet to a "logical
switch" and the gateway will figure out the output port and vlan tag
automatically.
Therefore, it is really not necessary to keep record of the vlan map
for each gateway physical port in the OVN_Southbound database using
"gateway_ports" and to map each vlan to a unique ovn logical port.
Instead, we should simply map each logical switch to a ovn logical
port.
Thusly, this commit removes the "Gateway" table from the OVN_Southbound
database. In the "Chassis" table, the "gateway_ports" column is replaced
by "vtep_logical_switches" column which stores all vtep logical switch
names. The use of this column will be documented in later commit.
Signed-off-by: Alex Wang <alexw@nicira.com> Acked-by: Russell Bryant <rbryant@redhat.com>
will run "make check-system-userspace" in the vagrant launched VM.
It may be more convenient to run this tests inside a vm rather than in
the host, because they interact with system networking.
Suggested-by: Joe Stringer <joestringer@nicira.com> Signed-off-by: Daniele Di Proietto <diproiettod@vmware.com> Acked-by: Joe Stringer <joestringer@nicira.com>
The new system-userspace-testsuite, which can be launched via
`make check-system-userspace`, reuses the kmod tests on the userspace
datapath.
The userspace datapath is already tested by the main testsuite (and
that's not going to change), but having also the
system-userspace-testsuite has the following advantages:
* More complicated tests are possible: real client and server
applications can be used.
* The same tests run on both kernel and userspace datapath: this gives
us an easy way to make sure that the behaviour is consistent (e.g.
with the upcoming connection tracker integration)
The userspace datapath is able to use system network interfaces via an
AF_PACKET socket.
Signed-off-by: Daniele Di Proietto <diproiettod@vmware.com> Acked-by: Joe Stringer <joestringer@nicira.com>
Joe Stringer [Wed, 29 Jul 2015 23:28:59 +0000 (16:28 -0700)]
kmod-macros: Don't unload kmod in VSWITCHD_STOP.
We already queue the removal of the kernel module in OVS_VSWITCHD_START,
via an ON_EXIT() call. That command is executed in both the success and
failure cases, so it is unnecessary to unload the kernel module in
OVS_VSWITCHD_STOP.
Signed-off-by: Joe Stringer <joestringer@nicira.com> Acked-by: Daniele Di Proietto <diproiettod@vmware.com>
The file will be part of two different testsuites: one for the kernel
datapath and another for the userspace datapath. No functional change
in this commit.
Signed-off-by: Daniele Di Proietto <diproiettod@vmware.com> Acked-by: Joe Stringer <joestringer@nicira.com>
Ian Stokes [Thu, 6 Aug 2015 15:58:47 +0000 (16:58 +0100)]
INSTALL.DPDK.md: Add details of XL710 restrictions for DPDK
Currently there are restrictions regarding the use of the XL710 network
interface with OVS and DPDK. This patch details those restrictions in
INSTALL.DPDK.md.
Signed-off-by: Ian Stokes <ian.stokes@intel.com> Signed-off-by: Ethan Jackson <ethan@nicira.com> Acked-by: Ethan Jackson <ethan@nicira.com>
netdev-dpdk: Retry tx/rx queue setup until we don't get any failure.
It has been observed that some DPDK device (e.g intel xl710) report an
high number of queues but make some of them available only for special
functions (SRIOV). Therefore the queues will be counted in
rte_eth_dev_info_get(), but rte_eth_tx_queue_setup() will fail.
This commit works around the issue by retrying the device initialization
with a smaller number of queues, if a queue fails to setup.
Reported-by: Ian Stokes <ian.stokes@intel.com> Tested-by: Ian Stokes <ian.stokes@intel.com> Acked-by: Kevin Traynor <kevin.traynor@intel.com> Signed-off-by: Daniele Di Proietto <diproiettod@vmware.com> Signed-off-by: Ethan Jackson <ethan@nicira.com>
Jesse Gross [Tue, 30 Jun 2015 01:01:59 +0000 (18:01 -0700)]
dpif-netdev: Translate Geneve options per-flow, not per-packet.
The kernel implementation of Geneve options stores the TLV option
data in the flow exactly as received, without any further parsing.
This is then translated to known options for the purposes of matching
on flow setup (which will then install a datapath flow in the form
the kernel is expecting).
The userspace implementation behaves a little bit differently - it
looks up known options as each packet is received. The reason for this
is there is a much tighter coupling between datapath and flow translation
and the representation is generally expected to be the same. This works
but it incurs work on a per-packet basis that could be done per-flow
instead.
This introduces a small translation step for Geneve packets between
datapath and flow lookup for the userspace datapath in order to
allow the same kind of processing that the kernel does. A side effect
of this is that unknown options are now shown when flows dumped via
ovs-appctl dpif/dump-flows, similar to the kernel.
There is a second benefit to this as well: for some operations it is
preferable to keep the options exactly as they were received on the wire,
which this enables. One example is that for packets that are executed from
ofproto-dpif-upcall to the datapath, this avoids the translation of
Geneve metadata. Since this conversion is potentially lossy (for unknown
options), keeping everything in the same format removes the possibility
of dropping options if the packet comes back up to userspace and the
Geneve option translation table has changed. To help with these types of
operations, most functions can understand both formats of data and seamlessly
do the right thing.
dpif-netdev: Don't use metaflow to operate on userspace datapath fields.
If ofproto-dpif installs a flow into the userspace datapath that doesn't
include a mask, we need to synthesize an exact match one. This is currently
done using the metaflow infrastructure, iterating over each field and
setting it to all ones.
There is a conceptual mismatch here because metaflow is operating on
OpenFlow fields, not datapath ones. Even though they are generally very
similar, there are subtle differences, which is why it is necessary to
fix up the input port mask.
With Geneve options, the mapping is much more complicated and so the
situation is worse. The first issue is that the metaflow to flow
mapping can change over time, so we would need to do more revalidation
to track this. In addition, an upcoming patch will completely disconnect
the option format between ofproto-dpif and dpif-netdev, so the values
written by metaflow don't make sense at all.
When megaflows are turned off, ofproto-dpif internally generates masks
using flow_wildcards_init_for_packet(). Since that's the same as what
we want to do here, we can just use that instead of metaflow.
Ilya Maximets [Tue, 4 Aug 2015 19:36:37 +0000 (12:36 -0700)]
dpif-netdev: fix race for queues between pmd threads.
Currently pmd threads select queues in pmd_load_queues() according to
get_n_pmd_threads_on_numa(). This behavior leads to race between pmds,
beacause dp_netdev_set_pmds_on_numa() starts them one by one and
current number of threads changes incrementally.
As a result we may have the following situation with 2 pmd threads:
* dp_netdev_set_pmds_on_numa()
* pmd12 thread started. Currently only 1 pmd thread exists.
dpif_netdev(pmd12)|INFO|Core 1 processing port 'port_1'
dpif_netdev(pmd12)|INFO|Core 1 processing port 'port_2'
* pmd14 thread started. 2 pmd threads exists.
dpif_netdev|INFO|Created 2 pmd threads on numa node 0
dpif_netdev(pmd14)|INFO|Core 2 processing port 'port_2'
We have:
core 1 --> port 1, port 2
core 2 --> port 2
Fix this by starting pmd threads only after all of them have
been configured.
Cc: Daniele Di Proietto <diproiettod at vmware.com> Cc: Dyasly Sergey <s.dyasly at samsung.com> Acked-by: Flavio Leitner <fbl@sysclose.org> Acked-by: Daniele Di Proietto <diproiettod@vmware.com> Signed-off-by: Ilya Maximets <i.maximets at samsung.com> Signed-off-by: Ethan Jackson <ethan@nicira.com>
Joe Stringer [Wed, 29 Jul 2015 19:56:06 +0000 (12:56 -0700)]
kmod-traffic: Expand sanity tests.
The initial sanity test only checked IPv4 without IP fragments. This patch
adds additional tests using IPv6 and VLANs with IP fragments and expands
the existing test to be more strict.
Signed-off-by: Joe Stringer <joestringer@nicira.com> Acked-by: Daniele Di Proietto <diproiettod@vmware.com>
Ben Pfaff [Sun, 2 Aug 2015 18:51:32 +0000 (11:51 -0700)]
ofproto-dpif: Use a regular ref instead of try_ref for rule translation.
Until now, flow translation has had to use try_ref to take a reference on
a rule, because a competing thread might have released the last reference
and done an RCU-postponed deletion. Since classifier versioning was
introduced, however, the release of the last reference is itself
RCU-postponed, which means that it is always safe to take the reference
directly.
Changing try_ref to ref means that taking a reference can't fail, which
allows the caller to take a reference in cases where the need to take a
reference was previously passed along a call chain, which simplifies some
code.
Signed-off-by: Ben Pfaff <blp@nicira.com> Acked-by: Jarno Rajahalme <jrajahalme@nicira.com>
Ben Pfaff [Mon, 3 Aug 2015 23:38:12 +0000 (16:38 -0700)]
ovn: Change strategy for tunnel keys.
Until now, OVN has used "flat" tunnel keys, in which the STT tunnel key or
Geneve VNI contains a logical port number. Logical port numbers are unique
within an OVN deployment.
Flat tunnel keys have the advantage of simplicity. However, for packets
that are destined to logical ports on multiple hypervisors, they require
sending one packet per destination logical port rather than one packet per
hypervisor. They also make it hard to integrate with VXLAN-based hardware
switches, which use VNIs to designate logical networks instead of logical
ports.
This commit switches OVN to a different scheme. In this scheme, in Geneve
the VNI designates a logical network and a Geneve option specifies the
logical input and output ports, which are now scoped within the logical
network rather than globally unique. In STT, all three identifiers are
encoded in the tunnel key.
To allow for the reduced amount of traffic for packets destined to logical
ports on multiple hypervisors, this commit also introduces the concept
of a logical multicast group. The membership of these groups can be set
using a new Multicast_Group table in the southbound database (and
ovn-northd does use it starting in this commit).
With multicast groups alone, it would be difficult to implement ACLs,
because an ACL might disallow only some of the packets being sent to
a remote hypervisor. Thus, this commit also splits the OVN logical
pipeline into two pipelines: the "ingress" pipeline, which makes the
decision about the logical destination of a packet as a set of logical
ports or multicast groups, and the "egress" pipeline, which runs on the
destination hypervisor with the multicast group destination exploded into
individual ports and makes a final decision on whether to deliver the
packet. The "egress" pipeline can efficiently apply ACLs.
Until now, the OVN logical and physical pipeline implementation was not
adequately documented. This commit adds extensive documentation to
the OVN manpages to cover these issues.
Signed-off-by: Ben Pfaff <blp@nicira.com> Acked-by: Justin Pettit <jpettit@nicira.com>
Ben Pfaff [Sun, 19 Jul 2015 01:29:18 +0000 (18:29 -0700)]
actions: Allow caller to specify output table.
When an upcoming commit divides the pipeline up into ingress and egress
pipeline, it will become necessary to resubmit to different tables from
each of those pipelines to implement output. This commit makes that
possible.
Signed-off-by: Ben Pfaff <blp@nicira.com> Acked-by: Justin Pettit <jpettit@nicira.com>
Ben Pfaff [Mon, 3 Aug 2015 18:10:43 +0000 (11:10 -0700)]
ovn: Rename Pipeline table to Logical_Flow table.
The OVN pipeline is being split into two phases, which are most naturally
called "pipelines". I kept getting very confused trying to call them
anything else, and in the end it seems to make more sense to just rename
the Pipeline table.
Signed-off-by: Ben Pfaff <blp@nicira.com> Acked-by: Justin Pettit <jpettit@nicira.com>