[net/dns] Use only NullDnsSocketPool.

[net/dns] Use only NullDnsSocketPool.

This CL disables the source UDP port randomization in DnsClient
on Windows. The randomization requires preallocation of large
socket pools which is suspected to lead to long winsock calls.

BUG=107413, 165382
Committed: https://src.chromium.org/viewvc/chrome?view=rev&revision=194816

[mmenke1, 2013-04-17 21:04:20]

Is there a plan for getting rid of the DNS socket pool code?  We can always
revive it if needed.

[szym, 2013-04-17 21:07:54]

On 2013/04/17 21:04:20, mmenke (incorrect) wrote:
> Is there a plan for getting rid of the DNS socket pool code?  We can always
> revive it if needed.

My primary concern is http://crbug.com/165382 
The hang during UDPSocketWin::Close() could be a blocker for AsyncDns.
This CL is only intended to discover any correlation between socket pooling and
the observed distribution of Net.UDPSocketWinClose and ThreadWatcher crashes.

[szym, 2013-04-17 21:08:49]

On 2013/04/17 21:07:54, szym wrote:
> On 2013/04/17 21:04:20, mmenke (incorrect) wrote:
> > Is there a plan for getting rid of the DNS socket pool code?  We can always
> > revive it if needed.
> 
> My primary concern is http://crbug.com/165382 
> The hang during UDPSocketWin::Close() could be a blocker for AsyncDns.
> This CL is only intended to discover any correlation between socket pooling
and
> the observed distribution of Net.UDPSocketWinClose and ThreadWatcher crashes.

However, if we feel that the added security of the pooling is of limited value
in general, then we should remove DnsSocketPool completely for now.

[mmenke, 2013-04-17 21:32:09]

On 2013/04/17 21:08:49, szym wrote:
> On 2013/04/17 21:07:54, szym wrote:
> > On 2013/04/17 21:04:20, mmenke (incorrect) wrote:
> > > Is there a plan for getting rid of the DNS socket pool code?  We can
always
> > > revive it if needed.
> > 
> > My primary concern is http://crbug.com/165382 
> > The hang during UDPSocketWin::Close() could be a blocker for AsyncDns.
> > This CL is only intended to discover any correlation between socket pooling
> and
> > the observed distribution of Net.UDPSocketWinClose and ThreadWatcher
crashes.
> 
> However, if we feel that the added security of the pooling is of limited value
> in general, then we should remove DnsSocketPool completely for now.

LGTM

[commit-bot: I haz the power, 2013-04-18 04:03:09]

CQ is trying da patch. Follow status at
https://chromium-status.appspot.com/cq/szym@chromium.org/14225015/1

[commit-bot: I haz the power, 2013-04-18 06:51:39]

Change committed as 194816