On Wed, 18 May 2011 00:18:12 +0200
Eric Dumazet [off-list ref] wrote:

Le lundi 02 mai 2011 à 15:27 -0700, David Miller a écrit :

quoted

From: Stephen Hemminger <redacted>
Date: Thu, 28 Apr 2011 08:43:37 -0700

quoted

On Thu, 28 Apr 2011 10:56:07 +0200
Eric Dumazet [off-list ref] wrote:

quoted

Four years ago, Patrick made a change to hold rtnl mutex during netlink
dump callbacks.

I believe it was a wrong move. This slows down concurrent dumps, making
good old /proc/net/ files faster than rtnetlink in some situations.

This occurred to me because one "ip link show dev ..." was _very_ slow
on a workload adding/removing network devices in background.

All dump callbacks are able to use RCU locking now, so this patch does
roughly a revert of commits :

1c2d670f366 : [RTNETLINK]: Hold rtnl_mutex during netlink dump callbacks
6313c1e0992 : [RTNETLINK]: Remove unnecessary locking in dump callbacks

This let writers fight for rtnl mutex and readers going full speed.

It also takes care of phonet : phonet_route_get() is now called from rcu
read section. I renamed it to phonet_route_get_rcu()

Signed-off-by: Eric Dumazet <redacted>
Cc: Patrick McHardy <redacted>
Cc: Remi Denis-Courmont <redacted>

Acked-by: Stephen Hemminger <redacted>

Applied, thanks everyone!

I think we probably have to make some fixes, because rtnl_fill_ifinfo()
can access fields that were previously protected with RTNL

Let's see if it's doable, before considering re-adding rtnl_lock() in
rtnl_dump_ifinfo() ?

First step : dev->ifalias

Its late here, I'll continue the work tomorrow.

Thanks

[PATCH net-next-2.6] net: add rcu protection to netdev->ifalias

Avoid holding RTNL in show_ifalias() and rtnl_fill_ifinfo() to
manipulate netdev->ifalias

Signed-off-by: Eric Dumazet <redacted>

I am beginning to think the idea of optimizing case of "ip link show"
running in background with adding/removing network devices is silly. 
And switching to RCU leads to all sorts of corner cases where only
partial data will be read. Is it really worth it?