On Sep 29, 2014, at 3:34 PM, Ankur Sharma wrote:
> +VOID
> +NlMsgAlignSize(const PNL_MSG_HDR nlh)
> +{
> +nlh->nlmsgLen = NLMSG_ALIGN(nlh->nlmsgLen);
One question I have here is what if nlmsgLen ends up being more than the
nlBufLen after adjusting for alignment?
Thanks,
-- Nithin
___
ovs-vsctl commands like 'ovs-vsctl list Interface p1' use the
'monitor' RPC method, which causes ovsdb sending updates to
the command session when changes are committed to the monitored
table. Since ovs-vsctl commands are short-lived, there is chance
that ovs-vsctl terminates the connection to ovs
On Sep 29, 2014, at 3:34 PM, Ankur Sharma wrote:
> In this patch we have added basic changes for
> handler registeration for FLOW_GET command.
>
> Signed-off-by: Ankur Sharma
LG.
Acked-by: Nithin Raju
___
dev mailing list
dev@openvswitch.org
http
OVS implements specific keywords like udp_src/udp_dst
to match on L4 header port in OpenFlow flows. However,
these keywords are not documented in the ovs-ofctl
manpage. This commit fixes it.
VMware-BZ: #1333815
Signed-off-by: Alex Wang
---
utilities/ovs-ofctl.8.in | 14 ++
1 fil
> diff --git a/datapath-windows/ovsext/Netlink/NetlinkBuf.c
> b/datapath-windows/ovsext/Netlink/NetlinkBuf.c
> index 918bddd..bc079ef 100644
> --- a/datapath-windows/ovsext/Netlink/NetlinkBuf.c
> +++ b/datapath-windows/ovsext/Netlink/NetlinkBuf.c
> @@ -291,7 +291,7 @@ NlBufAt(PNL_BUFFER nlBuf, UIN
hi Ankur,
Thanks for making the change. Looks good mostly, I had minor comments.
I'll ack the v2.
On Sep 29, 2014, at 3:34 PM, Ankur Sharma wrote:
> In this patch we have added following new APIs.
>
> NlMsgAlignSize => Aligns the size of netlink message.
> NlMsgSetSize => Sets the value of nlm
On Wed, Sep 17, 2014 at 4:12 PM, Madhu Challa
wrote:
>
> I got a chance to try out your code with the example flows and measure raw
> packet throughput with and without connection tracking. The code works and
> the performance numbers look pretty good.
>
Sorry that I missed this message earlier.
Thomas, is there a reason not to consider this patch as-is right now? It
seems like a straight-forward code cleanup with no direct association with
the later patches.
--Justin
On Fri, Sep 26, 2014 at 3:00 PM, Thomas Graf
wrote:
> Signed-off-by: Thomas Graf
> ---
> datapath/datapath.c |
On Sep 26, 2014, at 11:20 AM, Ben Pfaff wrote:
> On Wed, Sep 24, 2014 at 11:24:00AM -0700, Jarno Rajahalme wrote:
>> Aligned 64-bit memory accesses in i586 are atomic. By using an SSE
>> register we can make such memory accesses in one instruction without
>> bus-locking. Need to compile with -
We already update the count field as the last step of these functions,
so returning the current count is very cheap. Callers that care about
the count become a bit more efficient, as they avoid extra
non-inlineable function call.
Signed-off-by: Jarno Rajahalme
---
lib/cmap.c | 25
flow inserts and removals are simplified:
- No need for classifier internal mutex, as dpif-netdev already has a
'flow_mutex'
- Number of memory allocations/frees can be halved
Lookup code path is a bit more effcient as well, as we can rely on
netdev_flow_key always having inline data.
Signed-o
Depending on the CPU, count_1bits() may be more or less expensive, so
it is better to avoid unnecessary calls to it. When the map has
consecutive 1-bits, those can be cleared without counting.
Signed-off-by: Jarno Rajahalme
---
lib/flow.h |3 ++-
1 file changed, 2 insertions(+), 1 deletion(
miniflow_extract_from_flow() is faster than miniflow_init(), since it
does not scan the fields not significant for the packet type. More
importantly, it produces the same layout as miniflow_extract, which
does not try to compress zero-valued packet header fields.
Signed-off-by: Jarno Rajahalme
--
Working with netdev_flow_keys is simpler when the key has the fields
present even when they are all-zeroes.
Signed-off-by: Jarno Rajahalme
---
lib/flow.c |2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/lib/flow.c b/lib/flow.c
index a42895e..14a3350 100644
--- a/lib/flow.c
Previously we masked labels not present in the incoming packet.
Signed-off-by: Jarno Rajahalme
---
lib/flow.c | 36 ++--
lib/meta-flow.c |4 ++--
lib/odp-util.c | 14 --
ofproto/ofproto-dpif-xlate.c
Thx for the reminding,
I checked the branch-2.3 code, we use rwlock to protect the classifier
add/remove/lookup on ofproto level and dpif-netdev level.
So, I think we are fine there,
Alex Wang,
On Wed, Oct 1, 2014 at 1:10 PM, Ben Pfaff wrote:
> Does it need to be backported?
>
> On Wed, Oct
On Wed, Oct 1, 2014 at 1:38 PM, Jesse Gross wrote:
> I agree that the memory handling is too fragile here. Maybe we should
> just make people attach the dst before calling into tunneling routines
> though. It might be easier considering the multiple entry points (i.e.
> GRE calls iptunnel_xmit() d
On Wed, Oct 1, 2014 at 1:02 AM, Andy Zhou wrote:
> OVS_VPORT_TYPE_GENEVE is currently hard coded to 6. This is not
> necessary since slot 5 has not been taken yet. Drop the hard
> coded value to before upstreaming GENEVE support to Linux kernel.
>
> Signed-off-by: Andy Zhou
looks good.
Acked-by:
Ah, I missed the don't card case. Thanks for pointing it out. I will
drop the patch.
On Wed, Oct 1, 2014 at 1:48 PM, Jesse Gross wrote:
> On Wed, Oct 1, 2014 at 1:02 AM, Andy Zhou wrote:
>> TUNNEL_OPTIONS_PRESENT should always be checked against the tun_flgs
>> bits in a flow key, for both flow
On Wed, Oct 1, 2014 at 1:02 AM, Andy Zhou wrote:
> TUNNEL_OPTIONS_PRESENT should always be checked against the tun_flgs
> bits in a flow key, for both flow and mask serialization.
>
> Signed-off-by: Andy Zhou
I'm not sure that this is the right thing to do. In the case where you
want to match on
I agree that the memory handling is too fragile here. Maybe we should
just make people attach the dst before calling into tunneling routines
though. It might be easier considering the multiple entry points (i.e.
GRE calls iptunnel_xmit() directly).
Several callers are definitely expecting iptunnel
> > -ASSERT(KeGetCurrentIrql() == DISPATCH_LEVEL);
> > OvsAcquireDatapathRead(datapath, &dpLockState, dispatch);
>
> We can pass NDIS_RWL_AT_DISPATCH_LEVEL instead of 'dispatch', and let the
> ASSERT be.
>
> SV: Here, I will leave the use of the 'dispatch' variable, due to
It turns out rout entry will not be attached to skb until
iptunnel_xmit, which does not return negative value. So there is no
bug here.
I will drop this patch, and modify the upstream vport-geneve.c accordingly.
It is pretty subtle that the correctness of free depends on the
implementation detail
Does it need to be backported?
On Wed, Oct 01, 2014 at 01:03:11PM -0700, Alex Wang wrote:
> Thx for the review, applied to master,~
>
> On Wed, Oct 1, 2014 at 12:52 PM, Ben Pfaff wrote:
>
> > On Wed, Oct 01, 2014 at 12:03:30PM -0700, Alex Wang wrote:
> > > Currently, the cmap_destroy() directly
Thx for the review, applied to master,~
On Wed, Oct 1, 2014 at 12:52 PM, Ben Pfaff wrote:
> On Wed, Oct 01, 2014 at 12:03:30PM -0700, Alex Wang wrote:
> > Currently, the cmap_destroy() directly frees the cmap memory.
> > Some callers of cmap_destroy() (e.g. destroy_subtable()) still
> > allows o
On Wed, Oct 01, 2014 at 12:03:30PM -0700, Alex Wang wrote:
> Currently, the cmap_destroy() directly frees the cmap memory.
> Some callers of cmap_destroy() (e.g. destroy_subtable()) still
> allows other threads (e.g. pmd threads) accessing the cmap at
> the same time (e.g. via classifier_lookup_min
As of now, when someone passes a punix:foo/bar as a connection type
in Windows, we create a TCP server using 127.0.0.1 and save the kernel
assigned port number in the file foo/bar. The connection name
as obtained through pstream_get_name() would be ptcp:127.0.0.1:$PORT.
This was okay if pstream_get
On Wed, Oct 01, 2014 at 11:54:09AM -0700, Daniele Di Proietto wrote:
> When the bridge datapath_type is changed, ofproto is destroyed and immediately
> recreated. This involves closing and reopening the mgmt socket. If the
> destruction of the 'connmgr' is postponed, a race condition might happen,
Hi Nithin,
Thanks for your review. Please see my comments inline.
-Original Message-
From: Nithin Raju [mailto:nit...@vmware.com]
Sent: Wednesday, October 1, 2014 8:55 PM
To: Sorin Vinturis
Cc: dev@openvswitch.org
Subject: Re: [ovs-dev] [PATCH] datapath-windows: Incorrect assumption of t
When the bridge datapath_type is changed, ofproto is destroyed and immediately
recreated. This involves closing and reopening the mgmt socket. If the
destruction of the 'connmgr' is postponed, a race condition might happen, where
we first recreate the socket and then try to destroy it.
Reported-by
Currently, the cmap_destroy() directly frees the cmap memory.
Some callers of cmap_destroy() (e.g. destroy_subtable()) still
allows other threads (e.g. pmd threads) accessing the cmap at
the same time (e.g. via classifier_lookup_miniflow_batch()),
which could cause segfault.
To fix the above issue
Test the switch over to inactive ovs-vswitchd process when user
kill the currently active ovs-vswitchd.
Signed-off-by: Alex Wang
---
tests/ovs-vswitchd.at | 41 +
1 file changed, 41 insertions(+)
diff --git a/tests/ovs-vswitchd.at b/tests/ovs-vswitchd.a
Sending out meeting minutes for future reference.
Participants: Alin Serdean, Sorin Vinturis, Nithin Raju.
On Sep 30, 2014, at 6:54 PM, Nithin Raju wrote:
> hi Alin/Sorin,
> There were the items we had to discuss:
>
> 1. Netlink task status: vport, flow, events, packet.
Discussed status.
High
hi Sorin,
I had minor comments. Pls. see inline.
> #define NETLINK_FAMILY_NAME_LEN 48
>
> +#ifndef IFNAMSIZ
> +#define IFNAMSIZ 16
> +#endif
This is defined in OvsDpInterface.h. We can avoid an extra definition. See
commit - f92156ae2c86cf1b7c2076f2bca88b6fa10ec632.
> +
>
> /*
> * Netlink me
On Wed, Oct 1, 2014 at 1:02 AM, Andy Zhou wrote:
> Route entry will be free on error by vport when freeing skb.
> additional error check and free after xmit() will cause double free.
>
> Signed-off-by: Andy Zhou
I'm not sure that the route entry is attached to the skb in all cases.
If it's not t
On Tue, Sep 30, 2014 at 10:09 PM, Joe Stringer wrote:
> On 1 October 2014 11:59, Joe Stringer wrote:
>>
>>
>>
>> On 1 October 2014 11:55, Pravin Shelar wrote:
>>>
>>> On Tue, Sep 30, 2014 at 3:15 PM, Joe Stringer
>>> wrote:
>>> >
>>> >
>>> > On 1 October 2014 06:56, Pravin Shelar wrote:
>>> >>
Acquiring a spin lock raises the IRQL level to DISPATCH_LEVEL.
But in many places of the code, while holding a spin lock, there
are useless checks for the current IRQL against DISPATCH_LEVEL.
Also, the dispatch flag is not correctly set when calling
NdisAcquireRWLockXXX() functions, which causes an
On Oct 1, 2014, at 8:40 AM, Ben Pfaff wrote:
> On Tue, Sep 30, 2014 at 11:23:46PM -0700, Pritesh Kothari wrote:
>> Signed-off-by: Pritesh Kothari
>> ---
>> v2:
>> * Changed html to mardown syntax instead.
>> * Fix references to README in various places.
>> * Fix whitespace issues.
>
> Thanks.
Thanks, I applied this to master.
On Tue, Sep 30, 2014 at 06:00:03PM -0700, Jarno Rajahalme wrote:
> I buy that,
>
> Acked-by: Jarno Rajahalme
>
> > On Sep 30, 2014, at 5:03 PM, Ben Pfaff wrote:
> >
> > The C standard allows compilers to do type-based alias analysis, which
> > means that the
On Tue, Sep 30, 2014 at 11:23:46PM -0700, Pritesh Kothari wrote:
> Signed-off-by: Pritesh Kothari
> ---
> v2:
> * Changed html to mardown syntax instead.
> * Fix references to README in various places.
> * Fix whitespace issues.
Thanks. It looks nicer like this.
I found a few more references to
On Wed, Oct 01, 2014 at 02:42:52PM +, Flynn, Dennis R (Dennis) wrote:
> This commit provides the initial delivery of support for the Auto-Attach
> standard to Open vSwitch. This standard describes a compact method of using
> IEEE 802.1AB Link Layer Discovery Protocol (LLDP) with a IEEE 802.1aq
This is the final commit in the series of commits that deliver initial support
for Auto-Attach. Specifically this commit delivers auto-attach support to the
OVS bridge layer as well as the new auto-attach commands. The OVSDB schema is
modified to define the new auto-attach entries. The man pages a
diff --git a/lib/odp-util.h b/lib/odp-util.h
index 11b54dd..61b71b1 100644
--- a/lib/odp-util.h
+++ b/lib/odp-util.h
@@ -40,6 +40,7 @@ struct pkt_metadata;
SPR(SLOW_BFD,"bfd","Consists of BFD packets") \
SPR(SLOW_LACP, "lacp", "Consists of LACP packets")
This patch sequence provides OVS support for the IETF Auto-Attach SPBM draft
standard. This standard describes a compact method of using IEEE 802.1AB Link
Layer Discovery Protocol (LLDP) together with a IEEE 802.1aq Shortest Path
Bridging (SPB) network to automatically attach network devices to in
DPDK rings don't need one TX 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 ignores, for
DPDK ring netdevs, the provided queue id (= PMD thread core id) in
ne
Hi Joe,
I re sent the patch via my private email to avoid the legal warning.
The diff was generated with 'svn diff' based on change from the OVS 2.3.0 tar
ball.
I tested that meters get cleared.
My project does not deal with groups. I added a flush for them per Ben's
request but I could not test
In fail-open mode on disconnect from controller rules are flushed. It makes
sense to flush groups and meters as well.
Signed-off-by: Gur Stavi
---
Original issue was discussed here:
https://www.mail-archive.com/discuss@openvswitch.org/msg10930.html
Resending to avoid email client auto sig
Hi Gur, looks straightforward. I have a few broad feedback points though:
If you haven't read it yet, please take a look at the CONTRIBUTING file in
the top of the source tree.
I'm not sure what format this patch is in, but git won't accept it in my
setup. I suspect it's due to the footer message
TUNNEL_OPTIONS_PRESENT should always be checked against the tun_flgs
bits in a flow key, for both flow and mask serialization.
Signed-off-by: Andy Zhou
---
datapath/flow_netlink.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/datapath/flow_netlink.c b/datapath/flow_netlink.
OVS_VPORT_TYPE_GENEVE is currently hard coded to 6. This is not
necessary since slot 5 has not been taken yet. Drop the hard
coded value to before upstreaming GENEVE support to Linux kernel.
Signed-off-by: Andy Zhou
---
datapath/linux/compat/include/linux/openvswitch.h | 2 +-
1 file changed, 1
Route entry will be free on error by vport when freeing skb.
additional error check and free after xmit() will cause double free.
Signed-off-by: Andy Zhou
---
datapath/vport-vxlan.c | 2 --
1 file changed, 2 deletions(-)
diff --git a/datapath/vport-vxlan.c b/datapath/vport-vxlan.c
index 8689853
A few small issues found in preparing for GENEVE support upstreaming
*** BLURB HERE ***
Andy Zhou (3):
datapath: avoid hard code OVS_VPORT_TYPE_GENEVE
datapath: avoid double free routing entry in vxlan_port xmit
datapath: serializing GENEVE options based on flow key's tun_flags
datapath/f
On 30 September 2014 10:24, Ben Pfaff wrote:
> I suspect that check_recirc() and check_uid() should pass
> DPIF_FP_MODIFY as well as DPIF_FP_CREATE, just in case a previous run
> of OVS was killed at just the right time to leave a straggler probe
> flow in the datapath.
>
I checked on this, and
> Thanks for fixing this.
>
> Acked-by: Joe Stringer
applied, thank you.
YAMAMOTO Takashi
___
dev mailing list
dev@openvswitch.org
http://openvswitch.org/mailman/listinfo/dev
54 matches
Mail list logo
