Simon Horman [Sat, 8 Nov 2014 21:53:52 +0000 (13:53 -0800)]
datapath: Rename last_action() as nla_is_last() and move to netlink.h
The original motivation for this change was to allow the helper to be used
in files other than actions.c as part of work on an odp select group
action.
It was as pointed out by Thomas Graf that this helper would be best off
living in netlink.h. Furthermore, I think that the generic nature of this
helper means it is best off in netlink.h regardless of if it is used more
than one .c file or not. Thus, I would like it considered independent of
the work on an odp select group action.
Cc: Thomas Graf <tgraf@suug.ch> Cc: Pravin Shelar <pshelar@nicira.com> Cc: Andy Zhou <azhou@nicira.com> Signed-off-by: Simon Horman <simon.horman@netronome.com> Acked-by: Thomas Graf <tgraf@noironetworks.com> Signed-off-by: David S. Miller <davem@davemloft.net> Acked-by: Pravin B Shelar <pshelar@nicira.com>
Jarno Rajahalme [Mon, 27 Oct 2014 17:57:28 +0000 (10:57 -0700)]
classifier: Constify fields.
Some struct cls_match and cls_subtable fields were already documented
of being const. Make them const and use CONST_CAST where appropriate
to initialize them.
This will help catch future errors modifying those fields after
initialization.
Signed-off-by: Jarno Rajahalme <jrajahalme@nicira.com> Acked-by: Ben Pfaff <blp@nicira.com>
Jarno Rajahalme [Mon, 3 Nov 2014 17:56:54 +0000 (09:56 -0800)]
classifier: Make classifier_find_rule_exactly() lockless.
struct cls_match 'list' member was recently changed to an rculist.
This allows classifier_find_rule_exactly() to be made lockless.
Since subtable's 'max_priority' member would still require a lock, we
no longer check it before calling find_equal(). This adds a hash
table lookup in cases where the subtable may already be known to not
contain any rule of the target priority.
classifier_find_rule_exactly() is never called on the fastpath, so
this should not be significant.
Signed-off-by: Jarno Rajahalme <jrajahalme@nicira.com> Acked-by: Ben Pfaff <blp@nicira.com>
The systemd requires some actions when the package is
installed, upgraded or removed. This patch adds the
needed RPM sections with the missing systemd actions.
There is a change in behavior - the service is not enabled
or started by default.
Signed-off-by: Flavio Leitner <fbl@redhat.com> Signed-off-by: Ben Pfaff <blp@nicira.com>
Flavio Leitner [Thu, 6 Nov 2014 18:45:29 +0000 (16:45 -0200)]
fedora-spec: add openvswitch-test package
Put the utilities that are useful to diagnose performance
and connectivity issues in Open vSwitch setup into another
package since they are not needed in most cases.
Signed-off-by: Flavio Leitner <fbl@redhat.com> Signed-off-by: Ben Pfaff <blp@nicira.com>
Ben Pfaff [Thu, 6 Nov 2014 21:48:39 +0000 (13:48 -0800)]
FAQ: Describe how to add new fields and new actions.
We get these questions from time to time and it would be nice to just be
able to cut and paste the answer. (And possibly some people might actually
read the answer straight from the FAQ.)
Signed-off-by: Ben Pfaff <blp@nicira.com> Acked-by: Andy Zhou <azhou@nicira.com>
Simon Horman [Wed, 5 Nov 2014 23:55:47 +0000 (15:55 -0800)]
ovs-ofctl: Only allow usable protocols for group commands
parse_ofp_group_mod_str() may limit the usable protocols according
to the group and in particular its actions. However, without this
change ovs-ofctl ignores this calculation.
Signed-off-by: Simon Horman <simon.horman@netronome.com> Acked-by: Jarno Rajahalme <jrajahalme@nicira.com>
Andy Zhou [Thu, 16 Oct 2014 22:23:11 +0000 (15:23 -0700)]
bridge: Store datapath version into ovsdb
OVS userspace are backward compatible with older Linux kernel modules.
However, not having the most up-to-date datapath kernel modules can
some times lead to user confusion. Storing the datapath version in
OVSDB allows management software to check and optionally provide
notifications to users.
Signed-off-by: Andy Zhou <azhou@nicira.com> Acked-by: Ben Pfaff <blp@nicira.com>
Pravin B Shelar [Wed, 29 Oct 2014 09:45:48 +0000 (02:45 -0700)]
datapath: Convert dp rcu read operation to locked operations
dp read operations depends on ovs_dp_cmd_fill_info(). This API
needs to looup vport to find dp name, but vport lookup can
fail. Therefore to keep vport reference alive we need to
take ovs lock.
Found by code inspection.
Signed-off-by: Pravin B Shelar <pshelar@nicira.com> Acked-by: Andy Zhou <azhou@nicira.com>
Ben Pfaff [Tue, 4 Nov 2014 19:17:11 +0000 (11:17 -0800)]
ofproto-dpif: Make ofproto/trace a bit more like real packet translation.
Until now, ofproto/trace has looked up the flow itself. xlate_actions()
can do the flow lookup internally and, since that is what happens when a
packet arrives, having it do its own packet lookup makes a lot of sense.
I noticed this in connection with the actset_output field, which
xlate_actions() should set to OFPP_UNSET at the beginning of translation
before looking up the flow. ofproto/trace didn't do that, so it looked
up a rule with actset_output=0 instead. By having xlate_actions() do the
lookup, the behavior can be consistent and correct.
Signed-off-by: Ben Pfaff <blp@nicira.com> Acked-by: Jarno Rajahalme <jrajahalme@nicira.com>
Motonori Shindo [Tue, 4 Nov 2014 16:12:18 +0000 (01:12 +0900)]
netflow: Fix interpretation of flow_seq.
'flow_seq" field in NetFlow v5 header should represent a number of NetFlow
flow records exported while it is representing the number of NetFlow
packets exported in the current code. This patch fixes this problem.
Signed-off-by: Motonori Shindo <motonori@shin.do> Signed-off-by: Ben Pfaff <blp@nicira.com>
David Verbeiren [Tue, 14 Oct 2014 17:01:49 +0000 (19:01 +0200)]
netdev-dpdk: Fix DPDK rings broken by multi queue
DPDK rings don't need one queue per PMD thread and don't support multiple
queues (set_multiq function is undefined). To fix operation with DPDK rings,
this patch ignores EOPNOTSUPP error on netdev_set_multiq() and provides, for
DPDK rings, a netdev send() function that ignores the provided queue id
(= PMD thread core id).
Suggested-by: Maryam Tahhan <maryam.tahhan@intel.com> Signed-off-by: David Verbeiren <david.verbeiren@intel.com> Acked-by: Pravin B Shelar <pshelar@nicira.com>
Ben Pfaff [Fri, 10 Oct 2014 05:57:47 +0000 (22:57 -0700)]
nx-match: Add support for multiple OXM field assignments for one field.
actset_output, to be added in an upcoming commit, has one OXM assignment
in OpenFlow 1.3 and another one in OpenFlow 1.5. This commit allows both
of them to be supported in appropriate OpenFlow versions.
This feature is difficult to test on its own, so the same commit that adds
actset_output support also tests this feature.
Signed-off-by: Ben Pfaff <blp@nicira.com> Acked-by: Jarno Rajahalme <jrajahalme@nicira.com>
Jarno Rajahalme [Mon, 3 Nov 2014 22:40:22 +0000 (14:40 -0800)]
ofproto-dpif-xlate: Do not clear ports with frags in xlate_actions().
Transport ports are already cleared for lookup if need be, and they
should be available for action processing, so do not clear them in
xlate_actions().
A drop flow takes care of dropping the packet, so no special case is
needed in xlate_actions().
Signed-off-by: Jarno Rajahalme <jrajahalme@nicira.com> Acked-by: Ben Pfaff <blp@nicira.com>
Jarno Rajahalme [Mon, 3 Nov 2014 22:40:22 +0000 (14:40 -0800)]
tests/ofproto-dpif - fragment handling: Add detail.
Add transport port modifications to the existing frags handling test
case.
This demonstrates incorrect behavior by not moving the destination
port number to source port in normal mode for first fragment, as the
transport port numbers have been zeroed and the move has no effect,
and by allowing moves to/from transport ports on later fragments
(which do not have a transport header).
Next patches fix these problems.
Signed-off-by: Jarno Rajahalme <jrajahalme@nicira.com> Acked-by: Ben Pfaff <blp@nicira.com>
Pravin B Shelar [Thu, 16 Oct 2014 18:38:16 +0000 (11:38 -0700)]
route-table: extract gw information.
Routing table will be used by ovs userspace tunneling, it need to
know gw address, following commit extract gw information from
netlink message so that ovs can populate it in ovs route table.
Signed-off-by: Pravin B Shelar <pshelar@nicira.com> Acked-by: Thomas Graf <tgraf@noironetworks.com>
Pravin B Shelar [Thu, 16 Oct 2014 18:38:12 +0000 (11:38 -0700)]
route-table: Use classifier to store routing table.
Rather than using hmap for storing routing entries we can directly use
classifier which has support for priority and wildcard entries.
This makes route lookup lockless. This help when we use route lookup
for native tunneling.
Signed-off-by: Pravin B Shelar <pshelar@nicira.com> Acked-by: Thomas Graf <tgraf@noironetworks.com> Acked-by: Jarno Rajahalme <jrajahalme@nicira.com>
Pravin B Shelar [Thu, 16 Oct 2014 18:38:06 +0000 (11:38 -0700)]
route-table: get rid of name-table
name table maintains device ifindex to name mapping. On any name
table changes it invalidate name table and routing table. So
rather than building two tables this patch moves dev name to routing
entry and build routing table on any name table changes.
Signed-off-by: Pravin B Shelar <pshelar@nicira.com> Acked-by: Thomas Graf <tgraf@noironetworks.com>
Ben Pfaff [Fri, 3 Oct 2014 22:49:23 +0000 (15:49 -0700)]
xenserver: Turn on SSE and SSE2 for the build, for atomic 64-bit ops.
The ovs-atomic-i586 implementation of atomic operations can implement
64-bit atomics more efficiently when SSE is supported. XenServer runs only
on 64-bit capable processors, in 32-bit mode, so we know on XenServer that
SSE and SSE2 are supported because they are architectural for amd64. Thus,
this commit enables SSE and SSE2 when building for XenServer to get the
improved atomics support.
I tested that this successfully adds -msse -msse2 to the compiler flags
inside a XenServer DDK, but I didn't actually run it on a real XenServer
install.
Signed-off-by: Ben Pfaff <blp@nicira.com> Acked-by: Jarno Rajahalme <jrajahalme@nicira.com>
Jesse Gross [Fri, 31 Oct 2014 23:47:45 +0000 (16:47 -0700)]
packets: Use WORDS_BIGENDIAN for struct geneve_opt.
The macro LITTLE_ENDIAN is a constant, not a test for endianness,
so it doesn't actually tell us anything about the machine.
WORDS_BIGENDIAN is both correct and defined by configure so it is
more portable.
Reported-by: Ben Pfaff <blp@nicira.com> Signed-off-by: Jesse Gross <jesse@nicira.com> Acked-by: Ben Pfaff <blp@nicira.com>
Jarno Rajahalme [Fri, 31 Oct 2014 23:22:23 +0000 (16:22 -0700)]
classifier: Use rculist.
The list of identical, but lower priority rules is not currently used
in classifier lookup. A later patch introducing conjunctive matches
needs to access the list during lookups, so we must make the list RCU.
Signed-off-by: Jarno Rajahalme <jrajahalme@nicira.com> Acked-by: Ben Pfaff <blp@nicira.com>
Joe Stringer [Fri, 31 Oct 2014 21:05:46 +0000 (14:05 -0700)]
odp-util: Fix segfault in MPLS attribute parsing.
Just because the ethertype is MPLS, this doesn't mean that the datapath
understands and provides OVS_KEY_ATTR_MPLS attributes for the flow.
Previously we would check the size of the OVS_KEY_ATTR_MPLS attribute
before checking whether the attribute is present. This would cause a
segfault in nl_attr_get_size(), usually triggered from a handler thread.
This patch brings the MPLS parsing code more in line with the rest of
the parse_l2_5_onward() function, by only processing MPLS if the
attribute is present.
Reported-by: Pravin B Shelar <pshelar@nicira.com> Signed-off-by: Joe Stringer <joestringer@nicira.com> Acked-by: Ben Pfaff <blp@nicira.com>
Simon Horman [Fri, 31 Oct 2014 08:14:46 +0000 (17:14 +0900)]
ovs-ofctl: Free group buckets.
To some extent this is cosmetic because ovs-ovfctl will immediately
exit. However, it does seem to be in keeping with freeing 'gms' in
ofctl_group_mod_file().
Found by inspection using make check-valgrind.
Signed-off-by: Simon Horman <simon.horman@netronome.com> Signed-off-by: Ben Pfaff <blp@nicira.com>
Nithin Raju [Wed, 29 Oct 2014 22:18:19 +0000 (15:18 -0700)]
datapath-windows: Increase the maximum size of port name.
In userspace, port name sizes are restricted to IFNAMSIZ which is
defined to IF_NAME_SIZE in:
C:\Program Files (x86)\Windows Kits\8.1\Include\shared\netioapi.h
In the kernel, since IFNAMSIZ was not available, we previously defined a
value of 16 for the kernel. This is restrictive for Openstack
integration where we use UUID as the name.
In this patch, we make the kernel code also use the same value as the
userspace.
Also updated is the OVS.psm1 powershell script which now allows friendly
names to be upto 48 bytes.
Nithin Raju [Thu, 30 Oct 2014 20:17:43 +0000 (13:17 -0700)]
datapath-windows: Don't leak NBLs with multiple NBs.
Currently, if we receive an NBL with multiple NBs from NDIS, we just
ASSERT() and not do anything. The right thing to do obviously is to
process the NBL. This is a work in progress.
In the meantime, we should complete the NBL and not just leak it.
Signed-off-by: Nithin Raju <nithin@vmware.com> Acked-by: Eitan Eliahu <eliahue@vmware.com> Tested-by: Alin Gabriel Serdean <aserdean@cloudbasesolutions.com> Acked-by: Alin Gabriel Serdean <aserdean@cloudbasesolutions.com> Signed-off-by: Ben Pfaff <blp@nicira.com>
Ben Pfaff [Thu, 30 Oct 2014 18:40:07 +0000 (11:40 -0700)]
classifier: Change type used for priorities from 'unsigned int' to 'int'.
OpenFlow has priorities in the 16-bit unsigned range, from 0 to 65535.
In the classifier, it is sometimes useful to be able to have values below
and above this range. With the 'unsigned int' type used for priorities
until now, there were no values below the range, so some code worked
around it by converting priorities to 64-bit signed integers. This didn't
seem so great to me given that a plain 'int' also had the needed range.
This commit therefore changes the type used for priorities to int.
The interesting parts of this change are in pvector.h and classifier.c,
where one can see the elimination of the use of int64_t.
Signed-off-by: Ben Pfaff <blp@nicira.com> Acked-by: Jarno Rajahalme <jrajahalme@nicira.com>
Ben Pfaff [Thu, 30 Oct 2014 17:38:06 +0000 (10:38 -0700)]
netdev-linux: Avoid depending on kernel definition of rtnl_link_stats64.
We have to define our own with some kernel headers, so we might as well do
it everywhere, especially since there seems to be a problem with detecting
the presence of the definition with at least some kernels.
Reported-by: Wang Sheng-Hui <shhuiw@gmail.com> Signed-off-by: Ben Pfaff <blp@nicira.com> Acked-by: Joe Stringer <joestringer@nicira.com>
This patch adds bash command-line completion script for ovs-appctl,
ovs-dpctl, ovs-ofctl and ovsdb-tool command. Right now, the script
can do the following:
- display available completion or complete on unfinished user input
(long option, subcommand, and argument).
- once the subcommand (e.g. ofproto/trace) has been given, the
script will print the subcommand format.
- the script can convert between keywords like 'bridge/port/interface/dp'
and the available record in ovsdb.
The limitations are:
- only support small set of important keywords
(dp, datapath, bridge, switch, port, interface, iface).
- does not support parsing of nested options
(e.g. ovsdb-tool create [db [schema]]).
- does not support expansion on repeatitive argument
(e.g. ovs-dpctl show [dp...]).
- only support matching on long options, and only in the format
(--option [arg], i.e. should not use --option=[arg]).
To use the script, either copy it inside /etc/bash_completion.d/
or manually run it via . ovs-command-compgen.bash.
Also, a unit testsuite is provided as ovs-command-compgen-test.bash.
It is suggested that this test script be run only inside
tutorial/sandbox directory.
For more info please refer to utilities/ovs-command-compgen.INSTALL.md.
Signed-off-by: Alex Wang <alexw@nicira.com> Acked-by: Ben Pfaff <blp@nicira.com>
Jarno Rajahalme [Wed, 29 Oct 2014 18:34:40 +0000 (11:34 -0700)]
ovs_assert, tests: Support NDEBUG.
./configure accepts --enable-ndebug option. Make ovs_assert() honor
it, and make sure all test programs disable it.
The order of include files in test programs is also made uniform:
1. #include <config.h>
2. #undef NDEBUG
3. Include file of the test subject (to make sure it itself has
sufficient include directives).
4. System includes in alphapetical order.
5. OVS includes in aplhapetical order.
Suggested-by: Ben Pfaff <blp@nicira.com> Signed-off-by: Jarno Rajahalme <jrajahalme@nicira.com> Acked-by: Ben Pfaff <blp@nicira.com>
Thomas Graf [Thu, 30 Oct 2014 12:04:14 +0000 (13:04 +0100)]
doc: Add more cross references between docs
Suggested-by: Lori Jakab <lojakab@cisco.com> Signed-off-by: Thomas Graf <tgraf@noironetworks.com> Reviewed-by: Flavio Leitner <fbl@redhat.com> Signed-off-by: Ben Pfaff <blp@nicira.com>
Ben Pfaff [Wed, 29 Oct 2014 21:59:48 +0000 (14:59 -0700)]
vswitch.xml: Improve IPFIX documentation.
A coworker came by yesterday with a question about IPFIX in Open vSwitch.
I did not know the answer, so I read the documentation. The documentation
was harder to understand than necessary. This change should make it easier
to understand by providing an overview of the two forms of configuration
and grouping together the columns that apply only to one form of
configuration.
Ben Pfaff [Wed, 29 Oct 2014 21:11:07 +0000 (14:11 -0700)]
ovs-vsctl: Allow referring to rows whose names are UUIDs by those names.
Every row in the database has a UUID, generated by the database server.
Rows in most tables also have a user-assigned name (e.g. a bridge or port
or interface name). The ovs-vsctl database commands (e.g. "set", "get",
"list", ...) accept both UUIDs and names, but until now if a command's
argument had the form of a UUID, then it had to be the database-assigned
UUID for a row; that is, ovs-vsctl did not check whether it was the name
of a row. This commit changes that: a UUID argument to a database command
is now first checked against database UUIDs then, if it is not a database
UUID, it is checked as the name of a row.
This is prompted by Windows integration with OpenStack, which uses UUIDs
as port names.
Jarno Rajahalme [Wed, 29 Oct 2014 16:59:57 +0000 (09:59 -0700)]
tests/test-classifier: Properly use ovsrcu_postpone.
Following patches add stricter checks of RCU memory management of
rules removed from a classifier. This patch properly postpones
freeing of 'struct cls_rule's that have been removed from a
classifier.
Also remove all the rules from classifier before destructing it in
test_rule_replacement().
Signed-off-by: Jarno Rajahalme <jrajahalme@nicira.com> Acked-by: Ben Pfaff <blp@nicira.com>
Jarno Rajahalme [Wed, 29 Oct 2014 16:59:57 +0000 (09:59 -0700)]
lib/rculist: New RCU-iterator, single-writer doubly-linked list.
rculist allows concurrent lockless list iteration, while a writer may
be modifying the list. Multiple writers can be supported by using a
mutex in addition to rculist.
First user will be added in a following patch.
Signed-off-by: Jarno Rajahalme <jrajahalme@nicira.com> Acked-by: Ben Pfaff <blp@nicira.com>
Alex Wang [Thu, 16 Oct 2014 20:27:32 +0000 (13:27 -0700)]
command-line: Add function to print command usage.
This commit adds a new variable in 'struct command' for
recording the command usage. Also, a new function is
added to print the usage given the array of defined
commands.
Later patch will use the output in bash command-line
completion script.
Signed-off-by: Alex Wang <alexw@nicira.com> Acked-by: Ben Pfaff <blp@nicira.com>
Alex Wang [Thu, 16 Oct 2014 20:26:07 +0000 (13:26 -0700)]
dpctl: Implement 'list-commands' function.
This commit implements the 'list-commands' command for ovs-dpctl
and ovs-appctl dpctl/* commands. The function will print the
usage string for each subcommand.
Signed-off-by: Alex Wang <alexw@nicira.com> Acked-by: Ben Pfaff <blp@nicira.com>
Alex Wang [Fri, 17 Oct 2014 18:11:36 +0000 (11:11 -0700)]
ovs-appctl: Rename 'help' to 'list-commands'.
Having 'ovs-appctl help' and 'ovs-appctl --help' print different
output is confusing. This commit renames the 'help' to 'list-commands'.
Also, future patches will add the 'list-commands' to other ovs-*
commands, and the output will be used by bash command-line completion
script.
Signed-off-by: Alex Wang <alexw@nicira.com> Acked-by: Ben Pfaff <blp@nicira.com>
Alex Wang [Fri, 12 Sep 2014 00:24:35 +0000 (17:24 -0700)]
command-line: Add function to print all options.
This commit adds a function that prints (both long and short)
options of a ovs-* command. To use this function, option
'--option' is added to ovs-appctl/dpctl/ofctl and ovsdb-tool
commands. A future patch will use the option output to
conduct bash command-line completion.
Signed-off-by: Alex Wang <alexw@nicira.com> Acked-by: Ben Pfaff <blp@nicira.com>
Nithin Raju [Fri, 24 Oct 2014 18:41:17 +0000 (11:41 -0700)]
datapath-windows: Allow encapsulation if source is bridge-internal port.
It has been observed that when userspace generates and executes that
packet, the source port of such a packet is set to the bridge-internal
port. Currently, we allow encapsulation only if the source port is a VIF
port or no port. We relax the check in this patch.
Signed-off-by: Nithin Raju <nithin@vmware.com> Acked-by: Ankur Sharma <ankursharma@vmware.com> Acked-by: Alin Gabriel Serdean <aserdean@cloudbasesolutions.com> Tested-by: Alin Gabriel Serdean <aserdean@cloudbasesolutions.com> Signed-off-by: Ben Pfaff <blp@nicira.com>