On Thu, Jan 9, 2014 at 1:09 AM, Ben Pfaff wrote:
> On Wed, Jan 08, 2014 at 04:43:46PM +0100, Helmut Schaa wrote:
>> ifi_flags is unsigned, the local equivalents should do the same.
>>
>> Signed-off-by: Helmut Schaa
>
> In the GNU C library, ifru_flags in struct ifreq, in net/if.h, has
> type shor
On Thu, Jan 9, 2014 at 1:59 AM, Ben Pfaff wrote:
> On Wed, Jan 08, 2014 at 04:43:47PM +0100, Helmut Schaa wrote:
>> Due to a race condition when bringing up an internal port on Linux
>> some interface flags (e.g. IFF_MULTICAST) are falsely reset. This
>> happens because netlink events may be proce
The patch "Implement OpenFlow support for MPLS, for up to 3 labels"
allows ovs-vswtichd to fully represent an MPLS label stack containing
up to three labels in a flow.
MPLS push and pop actions are not permitted by that patch in the case where
there are more than three labels present or more than
On Thu, Jan 09, 2014 at 01:04:33AM -0200, Flavio Leitner wrote:
> There is a chicken and egg issue where common OVS
> configuration uses a controller which is only accessible
> via the network. So starting OVS before network.target
> would break those configurations.
>
> However, the network doesn
On Thu, Jan 09, 2014 at 01:43:59AM -0200, Flavio Leitner wrote:
> I wrote a script to have ovs-vsctl bash completion. Right now it is
> missing a few table commands, but otherwise it should work just fine.
That's awesome! I started work on something like this years ago, but
was never satisfied en
On Wed, Jan 08, 2014 at 02:36:03PM -0500, Jesse Gross wrote:
> On Tue, Jan 7, 2014 at 11:03 PM, Simon Horman wrote:
> > On Tue, Jan 07, 2014 at 09:30:54PM -0500, Jesse Gross wrote:
> >> I was talking about the case where there is a single pass and you have
> >> multiple MPLS pop actions. This prev
Hi,
I wrote a script to have ovs-vsctl bash completion. Right now it is
missing a few table commands, but otherwise it should work just fine.
It's available here:
http://people.redhat.com/fleitner/ovs-vsctl/ovs-vsctl.sh
Just copy it into /etc/bash_completion.d directory.
(you might need to start
I was out of office (vacation), so sorry for the delay to reply.
On Thu, Jan 09, 2014 at 01:04:33AM -0200, Flavio Leitner wrote:
> Changelog:
> V2:
>- added checking for the service unit in ifup/ifdown scripts
>- brcompat removed
>- added the new sysconfig file
Thanks,
fbl
_
There is a chicken and egg issue where common OVS
configuration uses a controller which is only accessible
via the network. So starting OVS before network.target
would break those configurations.
However, the network doesn't come up after boot because
OVS isn't started until after the network scri
Sounds good, I'll merge this shortly.
Ethan
On Wed, Jan 8, 2014 at 6:52 PM, Alex Wang wrote:
> Sorry, forgot to add the v2,
>
> One additional point, we cannot directly call "bfd_set_state(bfd,
> bfd->state, bfd->diag)" since that will not work when we reset the
> cpath_down.
>
> Thanks,
>
>
> O
Sorry, forgot to add the v2,
One additional point, we cannot directly call "bfd_set_state(bfd,
bfd->state, bfd->diag)" since that will not work when we reset the
cpath_down.
Thanks,
On Wed, Jan 8, 2014 at 6:51 PM, Alex Wang wrote:
> Commit ccc09689 (bfd: Implement Bidirectional Forwarding Det
Yeah! thanks for the key point!
On Wed, Jan 8, 2014 at 6:41 PM, Ethan Jackson wrote:
> One problem with this, is that if some other code later calls
> bfd_set_state() with something other than DIAG_NONE, we lose our
> cpath_down status. I think instead, we should change bfd_set_state()
> to al
Commit ccc09689 (bfd: Implement Bidirectional Forwarding Detection.)
set the bfd local diagnostic to "Concatenated Path Down" in response
to the set of cpath_down only when the current local diagnostic is
"None". However, since the bfd local diagnostic is not reset when
the bfd state is restored f
One problem with this, is that if some other code later calls
bfd_set_state() with something other than DIAG_NONE, we lose our
cpath_down status. I think instead, we should change bfd_set_state()
to always override with cpath_down instead of just when DIAG_NONE.
Then whenever cpath_down changes we
Commit ccc09689 (bfd: Implement Bidirectional Forwarding Detection.)
set the bfd local diagnostic to "Concatenated Path Down" in response
to the set of cpath_down only when the current local diagnostic is
"None". However, since the bfd local diagnostic is not reset when
the bfd state is restored f
Thanks, I have pushed this bug fix to master, branch-2.1, branch-2.0 and
branch-1.11.
On Wed, Jan 8, 2014 at 2:42 PM, Ben Pfaff wrote:
> On Wed, Jan 08, 2014 at 02:32:24PM -0800, Andy Zhou wrote:
> > When vlan-splinter is enabled, ovs receives non-vlan flows from the
> > kernel vlan ports, vlan
Interestingly, when running with 10 handler threads, and 4
revalidators, this doesn't seem to help much. I still see us spending
roughly 20% of the time in the kernel, and the performance benefit is
only about 2k tps. We'll need to investigate further.
Ethan
On Wed, Jan 8, 2014 at 4:15 PM, Jar
On Wed, Jan 08, 2014 at 04:57:28PM -0800, Pravin Shelar wrote:
> On Fri, Dec 27, 2013 at 8:03 PM, Ben Pfaff wrote:
> > For now, we use exactly two threads. Presumably at some point we will want
> > to make this configurable.
> >
> > Signed-off-by: Ben Pfaff
>
> Look good.
>
> Thanks.
Thanks.
Thx,~
On Wed, Jan 8, 2014 at 5:07 PM, Ben Pfaff wrote:
> On Wed, Jan 08, 2014 at 04:52:15PM -0800, Alex Wang wrote:
> > This commit makes the ofproto/trace show the megaflow fields
> > for each resubmit.
> >
> > Signed-off-by: Alex Wang
>
> All applied, thanks!
>
__
On Wed, Jan 08, 2014 at 03:19:45PM -0800, Ethan Jackson wrote:
> For pthreads, should the init function set flag->b = false?
Yes. Ha, that's funny, I went to all this trouble to add init
functions and didn't actually implement the initialization in the
single case where it's really needed.
I fix
On Wed, Jan 08, 2014 at 04:52:15PM -0800, Alex Wang wrote:
> This commit makes the ofproto/trace show the megaflow fields
> for each resubmit.
>
> Signed-off-by: Alex Wang
All applied, thanks!
___
dev mailing list
dev@openvswitch.org
http://openvswitch
On Wed, Jan 8, 2014 at 3:57 PM, Ben Pfaff wrote:
> On Wed, Jan 08, 2014 at 03:32:25PM -0800, Pravin Shelar wrote:
>> On Fri, Dec 27, 2013 at 8:03 PM, Ben Pfaff wrote:
>> > This will allow for parallelism in multithreaded forwarding in an upcoming
>> > commit.
>> >
>> > Signed-off-by: Ben Pfaff
>
On Wed, Jan 08, 2014 at 04:43:47PM +0100, Helmut Schaa wrote:
> Due to a race condition when bringing up an internal port on Linux
> some interface flags (e.g. IFF_MULTICAST) are falsely reset. This
> happens because netlink events may be processed after the according
> netdev has been brought up (
On Fri, Dec 27, 2013 at 8:03 PM, Ben Pfaff wrote:
> For now, we use exactly two threads. Presumably at some point we will want
> to make this configurable.
>
> Signed-off-by: Ben Pfaff
Look good.
Thanks.
> ---
> lib/dpif-netdev.c | 210
> +++-
This commit makes the ofproto/trace show the megaflow fields
for each resubmit.
Signed-off-by: Alex Wang
---
ofproto/ofproto-dpif.c | 31 ---
tests/classifier.at| 16
2 files changed, 28 insertions(+), 19 deletions(-)
diff --git a/ofproto/ofp
This commit changes the output field name from "Relevant fields"
to "Megaflow".
Signed-off-by: Alex Wang
---
ofproto/ofproto-dpif.c |2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/ofproto/ofproto-dpif.c b/ofproto/ofproto-dpif.c
index cf6e213..e7b74f7 100644
--- a/ofproto/o
This commit removes the unused variables in ofproto_trace() function.
Signed-off-by: Alex Wang
---
ofproto/ofproto-dpif.c |4
1 file changed, 4 deletions(-)
diff --git a/ofproto/ofproto-dpif.c b/ofproto/ofproto-dpif.c
index 03f7840..cf6e213 100644
--- a/ofproto/ofproto-dpif.c
+++ b/ofp
Commit bcd2633a (ofproto-dpif: Store relevant fields for wildcarding
in facet.) introduced the bug that uses original input flow as final
flow to compute the "Relevant fields" in ofproto/trace ouput. This
commit fixes this bug.
Signed-off-by: Alex Wang
---
ofproto/ofproto-dpif.c |2 +-
1 fi
Most per-cpu stats are never used and remain at zero. Avoid the
locking overhead for such stats. In the unlikely event of a
concurrent writer holding the lock for the first write on the
stats we'll skip spinning on the lock and do not get those stats at
this time, but will get the stats on the ne
Prefetch flow stats as early as possible. This may have an impact on
throughput when many cores hammer on the same flows.
From: Jarno Rajahalme
Signed-off-by: Jarno Rajahalme
---
datapath/datapath.c | 13 +
datapath/flow_table.c | 15 ++-
2 files changed, 27 inser
The stats are taken before execution, but flow stats are updated only
after execution in hope that stats prefetch is complete by then.
Signed-off-by: Jarno Rajahalme
---
datapath/datapath.c | 34 +-
datapath/flow.c | 20 +---
datapath/f
Keep kernel flow stats for each NUMA node rather than each (logical)
CPU. This almost removes the kernel-side OVS locking overhead
otherwise on the top of perf report and improves TCP_CRR test scores
by roughly 50% (with 4 handler and 2 revalidator threads with a TCP
port flow forcing all connecti
'perf' report on current master shows that kernel-side locking is
taking more than 20% of the overall OVS execution time under TCP_CRR
tests with flow tables that make all new connections go to userspace
and that create huge amount of kernel flows (~6) on a 32-CPU
server. It turns out that dis
On Wed, Jan 08, 2014 at 04:43:46PM +0100, Helmut Schaa wrote:
> ifi_flags is unsigned, the local equivalents should do the same.
>
> Signed-off-by: Helmut Schaa
In the GNU C library, ifru_flags in struct ifreq, in net/if.h, has
type short int. I see that netdev-linux.c uses 'int' and 'unsigned
On Wed, Jan 08, 2014 at 04:40:50PM +0100, Daniele Venturino wrote:
> The command shown in the man page to disable the STP protocol on a bridge is:
>
> ovs-vsctl clear Bridge br0 stp_enable
>
> Calling that, the following warning message is returned:
>
> ovs-vsctl: "clear" operation c
On Wed, Jan 08, 2014 at 03:32:25PM -0800, Pravin Shelar wrote:
> On Fri, Dec 27, 2013 at 8:03 PM, Ben Pfaff wrote:
> > This will allow for parallelism in multithreaded forwarding in an upcoming
> > commit.
> >
> > Signed-off-by: Ben Pfaff
...
> > +static struct dp_netdev_flow *
> > +dp_netdev_f
On Fri, Dec 27, 2013 at 8:03 PM, Ben Pfaff wrote:
> This will allow for parallelism in multithreaded forwarding in an upcoming
> commit.
>
> Signed-off-by: Ben Pfaff
> ---
> lib/dpif-netdev.c | 480
> +++--
> 1 file changed, 355 insertions(+), 12
For pthreads, should the init function set flag->b = false?
Acked-by: Ethan Jackson
On Mon, Dec 30, 2013 at 12:47 PM, Ben Pfaff wrote:
> Oops. I've folded this in:
>
> diff --git a/lib/ovs-atomic-gcc4+.h b/lib/ovs-atomic-gcc4+.h
> index 4476162..4938261 100644
> --- a/lib/ovs-atomic-gcc4+.h
>
Acked-by: Ethan Jackson
On Fri, Dec 27, 2013 at 8:03 PM, Ben Pfaff wrote:
> Normally all the ports have the same mtu anyhow, so there is little
> advantage in keeping track of the maximum mtu on a per-bridge basis. In
> upcoming commits, tracking mtu will require more locking and present
> eve
Seems cleaner at any rate.
Acked-by: Ethan Jackson
On Fri, Dec 27, 2013 at 8:03 PM, Ben Pfaff wrote:
> The goal is to make it easy to divide the ports into groups for handling
> by threads. It seems easy enough to do that by hash value, and a little
> harder otherwise.
>
> This commit has the
On Wed, Jan 08, 2014 at 02:32:24PM -0800, Andy Zhou wrote:
> When vlan-splinter is enabled, ovs receives non-vlan flows from the
> kernel vlan ports, vlan tag is then added to the incoming flow before
> xlating, so that they look like those received from a trunk port.
>
> In case megaflow is enabl
On Wed, Jan 08, 2014 at 01:22:25PM -0800, Pravin Shelar wrote:
> On Fri, Dec 27, 2013 at 8:03 PM, Ben Pfaff wrote:
> > This is analogous to the split between rule and rule_actions in
> > ofproto. As there, it will allow retaining a reference to a rule's
> > actions, while processing them, without
When vlan-splinter is enabled, ovs receives non-vlan flows from the
kernel vlan ports, vlan tag is then added to the incoming flow before
xlating, so that they look like those received from a trunk port.
In case megaflow is enabled, xlating may set vlan masks during rule
processing as usual. If th
Looks good to me,
On Thu, Dec 5, 2013 at 9:29 PM, Ben Pfaff wrote:
> OpenFlow 1.1 through 1.4 say that the out_group in a flow_mod, for commands
> other than OFPFC_DELETE or OFPFC_DELETE_STRICT, should be ignored. OVS
> actually reported an error. This fixes the problem.
>
> Reported-by: chen
On Fri, Dec 27, 2013 at 8:03 PM, Ben Pfaff wrote:
> This is analogous to the split between rule and rule_actions in
> ofproto. As there, it will allow retaining a reference to a rule's
> actions, while processing them, without having to retain a reference
> to the rule itself.
>
> Signed-off-by:
Looks good to me,
On Mon, Nov 25, 2013 at 2:14 PM, Ben Pfaff wrote:
> These would have found the problem fixed in the previous patch.
>
> Signed-off-by: Ben Pfaff
> ---
> ofproto/connmgr.c |3 ++-
> ofproto/fail-open.c |3 +++
> ofproto/fail-open.h |6 +++---
> 3 files changed,
On Wed, Jan 8, 2014 at 12:25 PM, Ben Pfaff wrote:
> On Wed, Jan 08, 2014 at 11:55:15AM -0800, Pravin B Shelar wrote:
>> If dpif is closed without deleting it, dpif-provider resources are
>> not freed.
>>
>> Signed-off-by: Pravin B Shelar
>
> I think that this will cause the kernel datapath to be
On Wed, Jan 08, 2014 at 11:07:38AM -0800, Alex Wang wrote:
> I suspect that this approach to a rwlock is faster when lots of threads
> > try to take the read lock at once, because it should avoid cross-core
> > contention on cache lines.
>
>
>
> For avoiding the contention, is that only because
On Wed, Jan 8, 2014 at 9:50 AM, James Page
>
wrote:
> -BEGIN PGP SIGNED MESSAGE-
> Hash: SHA256
>
> On 08/01/14 14:34, Jesse Gross wrote:
>>> Note: ipv6 enablement with vxlan is not supported by this
>>> change.
Signed-off-by: James Page >
>> I think there's some more compatibilit
On Wed, Jan 08, 2014 at 11:55:15AM -0800, Pravin B Shelar wrote:
> If dpif is closed without deleting it, dpif-provider resources are
> not freed.
>
> Signed-off-by: Pravin B Shelar
I think that this will cause the kernel datapath to be deleted, which
I believe will have bad side effects. For e
On Wed, Jan 08, 2014 at 07:06:32PM +0200, Lori Jakab wrote:
> On 12/30/13 9:28 PM, Ben Pfaff wrote:
> >On Tue, Dec 24, 2013 at 04:02:35PM +0200, Lorand Jakab wrote:
> >>This commit relaxes the assumption that all packets have an Ethernet
> >>header, and adds support for layer 3 flows. For each pac
On Fri, Dec 27, 2013 at 8:03 PM, Ben Pfaff wrote:
> By making "destroyed" own a reference, we can treat dp_netdev's ref_cnt
> like any other in Open vSwitch.
>
> Signed-off-by: Ben Pfaff
Looks good to me.
> ---
> lib/dpif-netdev.c | 23 ++-
> 1 file changed, 14 insertions(
On Fri, Dec 27, 2013 at 8:03 PM, Ben Pfaff wrote:
> This is a thin wrapper around an atomic_uint. It is useful anyhow because
> each ovs_refcount_ref() or ovs_refcount_unref() call saves a few lines of
> code.
>
> This commit also changes all the potential direct users over to use the new
> data
If dpif is closed without deleting it, dpif-provider resources are
not freed.
Signed-off-by: Pravin B Shelar
---
ofproto/ofproto-dpif.c |1 +
1 files changed, 1 insertions(+), 0 deletions(-)
diff --git a/ofproto/ofproto-dpif.c b/ofproto/ofproto-dpif.c
index 91ffe23..1d60359 100644
--- a/ofp
On Tue, Jan 7, 2014 at 11:03 PM, Simon Horman wrote:
> On Tue, Jan 07, 2014 at 09:30:54PM -0500, Jesse Gross wrote:
>> I was talking about the case where there is a single pass and you have
>> multiple MPLS pop actions. This previously wasn't possible because
>> userspace only supported a single l
I'm trying to learn, want to ask few questions
I suspect that this approach to a rwlock is faster when lots of threads
> try to take the read lock at once, because it should avoid cross-core
> contention on cache lines.
For avoiding the contention, is that only because each thread will have it
On Wed, Jan 08, 2014 at 09:53:51AM -0200, Flavio Leitner wrote:
> On Tue, Jan 07, 2014 at 06:17:59PM -0800, Ben Pfaff wrote:
> > On Tue, Jan 07, 2014 at 11:12:41PM -0200, Flavio Leitner wrote:
> > > The fedora's rpm policy doesn't allow unpackaged files, so
> > > currently the build fails when pyth
It's orthogonal I think.
On Wed, Jan 08, 2014 at 11:44:43AM +0200, Philip Roth wrote:
> this could also be used for 802.1AD right?
>
>
> On Tue, Jan 7, 2014 at 7:33 AM, Simon Horman wrote:
>
> > If VLAN acceleration is used when the kernel receives a packet
> > then the outer-most VLAN tag wil
On Wed, Jan 08, 2014 at 09:48:45AM -0800, Pravin Shelar wrote:
> I found atomic_init without destroy for following var.
> clock->slow_path
That one doesn't have a call to atomic_destroy() because the timeval
module doesn't ever get uninitialized.
> rule->ref_count
Thanks. I folded this in:
dif
Bump kernel support for datapath module to include 3.12.
Make use of native vxlan kernel module for Linux >= 3.12 or if
the kernel in use supports the required vxlan functions.
Make use of native iptunnels capability for Linux >= 3.12.
Note: ipv6 enablement with vxlan is not supported by this ch
I found atomic_init without destroy for following var.
clock->slow_path
rule->ref_count
otherwise looks good.
On Fri, Dec 27, 2013 at 8:03 PM, Ben Pfaff wrote:
> C11 is able to require that atomics don't need to be destroyed, but some
> of the OVS implementations do.
>
> Signed-off-by: Ben Pfaff
On 12/30/13 9:28 PM, Ben Pfaff wrote:
On Tue, Dec 24, 2013 at 04:02:35PM +0200, Lorand Jakab wrote:
This commit relaxes the assumption that all packets have an Ethernet
header, and adds support for layer 3 flows. For each packet received on
the Linux kernel datapath the l2 and l3 members of str
ifi_flags is unsigned, the local equivalents should do the same.
Signed-off-by: Helmut Schaa
---
Otherwise the follow-up patch throws a warning during conmpilation.
lib/netdev-linux.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/lib/netdev-linux.c b/lib/netdev-linux.c
in
Due to a race condition when bringing up an internal port on Linux
some interface flags (e.g. IFF_MULTICAST) are falsely reset. This
happens because netlink events may be processed after the according
netdev has been brought up (which sets interface flags).
Fix this by reading the interface flags
The command shown in the man page to disable the STP protocol on a bridge is:
ovs-vsctl clear Bridge br0 stp_enable
Calling that, the following warning message is returned:
ovs-vsctl: "clear" operation cannot be applied to column stp_enable of
table
Bridge, which is no
On Wed, Jan 8, 2014 at 9:49 AM, Zoltan Kiss wrote:
> Hi,
>
> I've tried the latest net-next on a Xenserver install with 1.9.3 userspace,
> and it seems this patch series broke it (at least after reverting that
> locally it works now). I haven't went too far yet checking what's the
> problem, but i
-BEGIN PGP SIGNED MESSAGE-
Hash: SHA256
On 08/01/14 14:34, Jesse Gross wrote:
>> Note: ipv6 enablement with vxlan is not supported by this
>> change.
>>>
>>> Signed-off-by: James Page
> I think there's some more compatibility code that we can turn off
> in 3.12 - GRE transmit support and
Hi,
I've tried the latest net-next on a Xenserver install with 1.9.3
userspace, and it seems this patch series broke it (at least after
reverting that locally it works now). I haven't went too far yet
checking what's the problem, but it seems the xenbrX device doesn't
really receive too much
On Wed, Jan 8, 2014 at 7:03 AM, James Page wrote:
> Bump kernel support for datapath module to include 3.12.
>
> Make use of native vxlan kernel module for Linux >= 3.12 or if
> the kernel in use supports the required vxlan functions.
>
> Note: ipv6 enablement with vxlan is not supported by this c
On Wed, Jan 8, 2014 at 5:13 AM, Wei Yongjun wrote:
> From: Wei Yongjun
>
> memory allocated by kmem_cache_alloc() should be freed using
> kmem_cache_free(), not kfree().
>
> Fixes: e298e5057006 ('openvswitch: Per cpu flow stats.')
> Signed-off-by: Wei Yongjun
Good catch, thanks.
I'll let David
Bump kernel support for datapath module to include 3.12.
Make use of native vxlan kernel module for Linux >= 3.12 or if
the kernel in use supports the required vxlan functions.
Note: ipv6 enablement with vxlan is not supported by this change.
Signed-off-by: James Page
---
FAQ
On Tue, Jan 07, 2014 at 06:17:59PM -0800, Ben Pfaff wrote:
> On Tue, Jan 07, 2014 at 11:12:41PM -0200, Flavio Leitner wrote:
> > The fedora's rpm policy doesn't allow unpackaged files, so
> > currently the build fails when python-twisted-conch is installed.
[...]
> Just now, I sent a request for fe
Following OpenFlow 1.4 semantics, make barriers wait for
flow monitor replies. This should fix a race in
"ofproto - flow monitoring pause and resume" test.
Signed-off-by: YAMAMOTO Takashi
---
ofproto/connmgr.c | 27 +++
ofproto/connmgr.h | 1 +
ofproto/ofproto.c | 16 ++
From: Wei Yongjun
memory allocated by kmem_cache_alloc() should be freed using
kmem_cache_free(), not kfree().
Fixes: e298e5057006 ('openvswitch: Per cpu flow stats.')
Signed-off-by: Wei Yongjun
---
net/openvswitch/flow_table.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git
this could also be used for 802.1AD right?
On Tue, Jan 7, 2014 at 7:33 AM, Simon Horman wrote:
> If VLAN acceleration is used when the kernel receives a packet
> then the outer-most VLAN tag will not be present in the packet
> when it is received by netdev-linux. Rather, it will be present
> in
75 matches
Mail list logo
