Hi,
Now that all the 2.4<->2.6 synchronizing stuff is done, we are moving
forward with the cleanup plan. This set is similar to the previous
set I sent on 8/27, but it is based on the latest version that was
accepted into the kernel with the seq_file changes, a few bug fixes
and a bit more cleanup stuff. This set is very comprehensive and
touches almost all the code. The set is broken down to many patches
for better tracking. It was already tested by me for functionality
and is undergoing a more thorough set of testing by our QA group for
any corner case bugs. A set that cleans up the 802.3ad code will
follow shortly.
This patch applies on 2.4.23-pre5. It would also apply on 2.6.0 after
Amir's patch 2/10 from the "[bonding 2.6] propagating master's
settings to slaves" set is accepted by Jeff and applied on 2.6.
patch set can be downloaded from:
http://osdn.dl.sourceforge.net/sourceforge/bonding/bonding-cleanup-2.4.23-pre5.tar.bz2
This will update the following files:
Documentation/networking/bonding.txt
Documentation/networking/ifenslave.c
drivers/net/bonding/bond_3ad.c
drivers/net/bonding/bond_alb.c
drivers/net/bonding/bond_alb.h
drivers/net/bonding/bonding.h
drivers/net/bonding/bond_main.c
include/linux/if_bonding.h
Description:
patch 1 - ifenslave lite - No more IP settings to slaves, unified
printing format, code re-org and broken to more functions.
patch 2 - convert all debug prints to use the dprintk macro and
consolidate format of all prints (e.g. "bonding: Error:
...").
patch 3 - death of typedef. eliminate bonding_t/slave_t types and
consolidate casting.
patch 4 - remove dead code, old compatibility stuff and redundant
checks.
patch 5 - consolidate timers initialization, error checking and
re-queuing.
patch 6 - convert too long if-else to a switch-case. Fix all locations
that handles bond->primary.
patch 7 - eliminate the multicast_mode module param. settings are now
done only according to mode.
patch 8 - slave list iteration - bond is no longer part of the list,
added cyclic list iteration macros.
patch 9 - consolidate function declarations:
o all functions begin with bond_
o return value, function name and all params are on the same
line.
o change some function names to be more descriptive.
patch 10 - consolidate names of function params and variables (e.g.
bond_dev instead of dev/master/master_dev).
patch 11 - change names/types for some of the members in struct
bonding. change position of members.
patch 12 - consolidate return values of functions.
patch 13 - put curly braces around all if, else, for, while, switch
statements. change conditions to short format.
e.g. (ptr == NULL) ==> (!ptr)
patch 14 - consolidate error handling in all xmit functions.
patch 15 - chomp all trailing white space.
patch 16 - remove duplicate empty lines. add empty lines where needed
to improve readability.
patch 17 - fix indentations.
patch 18 - code re-organization in bond_main.c according to context
(e.g. module initialization, bond initialization, device
entry points, monitoring, etc.). some last minute minor
changes and fixes.
--
| Shmulik Hen Advanced Network Services |
| Israel Design Center, Jerusalem |
| LAN Access Division, Platform Networking |
| Intel Communications Group, Intel corp. |
>patch 7 - eliminate the multicast_mode module param. settings are now
> done only according to mode.
This goes a bit beyond straight cleanup; could you explain the
rationale for this change? Also, unless I'm missing something, the
patch does not appear to update bonding.txt to reflect the fact that
the module parameter is no more.
-J
---
-Jay Vosburgh, IBM Linux Technology Center, [email protected]
> patch set can be downloaded from:
> http://osdn.dl.sourceforge.net/sourceforge/bonding/bonding-cleanup-2.4.23-pre5.tar.bz2
>
> This will update the following files:
>
> Documentation/networking/bonding.txt
> Documentation/networking/ifenslave.c
> drivers/net/bonding/bond_3ad.c
> drivers/net/bonding/bond_alb.c
> drivers/net/bonding/bond_alb.h
> drivers/net/bonding/bonding.h
> drivers/net/bonding/bond_main.c
> include/linux/if_bonding.h
>
> Description:
> patch 1 - ifenslave lite - No more IP settings to slaves, unified
> printing format, code re-org and broken to more functions.
> patch 2 - convert all debug prints to use the dprintk macro and
> consolidate format of all prints (e.g. "bonding: Error:
> ...").
> patch 3 - death of typedef. eliminate bonding_t/slave_t types and
> consolidate casting.
> patch 4 - remove dead code, old compatibility stuff and redundant
> checks.
I'm a bit concerned about doing some of this stuff in the 2.4 series. That
compatibility stuff is there for a reason, and was set to be removed in
2.6. Perhaps we shouldn't be doing stuff this drastic until 2.6 because of
the risk of breaking users.
Chad
On Thursday 25 September 2003 07:22 pm, Jay Vosburgh wrote:
> >patch 7 - eliminate the multicast_mode module param. settings are
> > now done only according to mode.
>
> This goes a bit beyond straight cleanup; could you explain
> the rationale for this change? Also, unless I'm missing something,
> the patch does not appear to update bonding.txt to reflect the fact
> that the module parameter is no more.
>
>
That question rings a bell :)
I had this discussion with Deniel Laurent on 8/30 and it sounded like
that wouldn't be a problem after all.
[from that thread]
> Do you know that this breaks upward compatibility ?
[snip]
> But if everyone is OK ... (I'd be since I currently only use
> multicast_mode = 1 ;-).
The rationale is that the propagation code already made this module
param obsolete, so the next step was to remove it entirely since it
had no effect anyway.
According to the propagation RFC we sent on 2/6, multicast list,
allmulti flag and promisc flag are all controlled the same way, and
that is according to the USES_PRIMARY macro. When the bond uses the
current slave as a primary interface, it, and only it, is supposed to
have the bond's properties, while in aggregation modes all slaves
have the same settings. There is no point in settings other slaves,
that are not supposed to be receiving in the first place, to have
loose filtering. Otherwise the stack will be flooded by duplicate
packets. The situation is bad enough now since bonding has no
solution for broadcast packets, but that's for another thread.
--
| Shmulik Hen Advanced Network Services |
| Israel Design Center, Jerusalem |
| LAN Access Division, Platform Networking |
| Intel Communications Group, Intel corp. |
On Thursday 25 September 2003 07:47 pm, Chad N. Tindel wrote:
> > patch 4 - remove dead code, old compatibility stuff and redundant
> > checks.
>
> I'm a bit concerned about doing some of this stuff in the 2.4
> series. That compatibility stuff is there for a reason, and was
> set to be removed in 2.6. Perhaps we shouldn't be doing stuff this
> drastic until 2.6 because of the risk of breaking users.
That's the word I got from Jay in response to the " [Kernel-janitors]
old ioctl definitions in 2.5" thread.
>Jay Vosburgh <[email protected]> wrote:
> I was going to add it on to the end of the clean up set, but
> if you want to do it, go ahead. Nobody seems to have objected to
> removing the _OLD stuff, which I view as a good thing.
--
| Shmulik Hen Advanced Network Services |
| Israel Design Center, Jerusalem |
| LAN Access Division, Platform Networking |
| Intel Communications Group, Intel corp. |