Hi Anton,

Thinking about this some more:
Did you try the "retain" option? it is the xor mask to set operation.

Examples:
----
#At offset 3 retain the first nibble but set the second to 0xA
action pedit munge offset 3 u8 set 0xA0 retain 0xF
#At offset 9, retain the second nibble but set the first to 0xA
action pedit munge offset 9 u8 set 0x0A retain 0xF0
---

So if you were to point to the nibble holding the cos field
in your match, then you can set it to a new value while retaining
bitfields you dont want to change...

I was going to ask you to test the retain in addition to the
extra mask you are adding on different sets - but i am beginning 
to think the mask you are adding is not needed at all.

Here is an example of other tests i was going to ask you to 
run (this time with 16bit field):

#At offset 0, invert/clear/preserve the first 16 bits
action pedit munge offset 0 u16 invert

original data:
45000054 00004000 40013ca7 7f000001
modified data (invert):
baff0054 00004000 40013ca7 7f000001
modified data (clear):
00000054 00004000 40013ca7 7f000001
retain
45000054 00004000 40013ca7 7f000001
set 0x1234
12340054 00004000 40013ca7 7f000001

You can pipe to a mirror action to send to dummy0 and then run tcpdump
on dummy to watch the edited fields. I can describe a simple setup
i use if this still doesnt make sense.

As a side note: I think we need an action to work on VLANs since
they are such an important field. The action would push/pop and edit
VLAN fields. If you are interested in this, I can help provide guidance
(It is one of those things on my todo longtimenow but i cant seem to get
around to it).

cheers,
jamal


cheers,
jamal

On Mon, 2012-02-27 at 17:15 +0300, Anton 'EvilMan' Danilov wrote:

Hi, Jamal.

2012/2/27 jamal [off-list ref]:

quoted

Looks like a reasonable feature to have - but it has to continue working
as before for things that dont need the mask.

quoted

diff --git a/tc/m_pedit.c b/tc/m_pedit.c
index 7499846..f081eff 100644
--- a/tc/m_pedit.c
+++ b/tc/m_pedit.c

quoted

      tkey->val = htonl(tkey->val & retain);
-     tkey->mask = htonl(tkey->mask | ~retain);
+     tkey->mask = htonl(~tkey->mask | ~retain);

A change like the above worries me that this may work for your use case
but will break other things that used to work. I'd like to run some
tests but dont have the cycles for the next few days. Alternatively, I
could explain some tests to you and you could validate and fix whatever
breaks.
Let me know what is best for you.

Can You explain tests to me? I'll check behavior with my changes and fix breaks