lists.openwall.net   lists  /  announce  owl-users  owl-dev  john-users  john-dev  passwdqc-users  yescrypt  popa3d-users  /  oss-security  kernel-hardening  musl  sabotage  tlsify  passwords  /  crypt-dev  xvendor  /  Bugtraq  Full-Disclosure  linux-kernel  linux-netdev  linux-ext4  linux-hardening  linux-cve-announce  PHC 
Open Source and information security mailing list archives
 
Hash Suite: Windows password security audit tool. GUI, reports in PDF.
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <063D6719AE5E284EB5DD2968C1650D6DCFE6E5BB@AcuExch.aculab.com>
Date:   Thu, 23 Feb 2017 16:04:10 +0000
From:   David Laight <David.Laight@...LAB.COM>
To:     'Xin Long' <lucien.xin@...il.com>
CC:     network dev <netdev@...r.kernel.org>,
        "linux-sctp@...r.kernel.org" <linux-sctp@...r.kernel.org>,
        Marcelo Ricardo Leitner <marcelo.leitner@...il.com>,
        Neil Horman <nhorman@...driver.com>,
        "Vlad Yasevich" <vyasevich@...il.com>,
        "davem@...emloft.net" <davem@...emloft.net>
Subject: RE: [PATCH net-next 2/2] sctp: add support for MSG_MORE

From: Xin Long
> Sent: 23 February 2017 03:46
> On Tue, Feb 21, 2017 at 10:27 PM, David Laight <David.Laight@...lab.com> wrote:
> > From: Xin Long
> >> Sent: 18 February 2017 17:53
> >> This patch is to add support for MSG_MORE on sctp.
> >>
> >> It adds force_delay in sctp_datamsg to save MSG_MORE, and sets it after
> >> creating datamsg according to the send flag. sctp_packet_can_append_data
> >> then uses it to decide if the chunks of this msg will be sent at once or
> >> delay it.
> >>
> >> Note that unlike [1], this patch saves MSG_MORE in datamsg, instead of
> >> in assoc. As sctp enqueues the chunks first, then dequeue them one by
> >> one. If it's saved in assoc,the current msg's send flag (MSG_MORE) may
> >> affect other chunks' bundling.
> >
> > I thought about that and decided that the MSG_MORE flag on the last data
> > chunk was the only one that mattered.
> > Indeed looking at any others is broken.
> >
> > Consider what happens if you have two small chunks queued, the first
> > with MSG_MORE set, the second with it clear.
> >
> > I think that sctp_outq_flush() will look at the first chunk and decide it
> > doesn't need to do anything because sctp_packet_transmit_chunk()
> > returns SCTP_XMIT_DELAY.
> > The data chunk with MSG_MORE clear won't even be looked at.
> > So the data will never be sent.

> It's not that bad as you thought, in sctp_packet_can_append_data():
> when inflight == 0 || sctp_sk(asoc->base.sk)->nodelay, the chunks
> would be still sent out.

One of us isn't understanding the other :-)

IIRC sctp_packet_can_append_data() is called for the first queued
data chunk in order to decide whether to generate a message that
consists only of data chunks.
If it returns SCTP_XMIT_OK then a message is built collecting the
rest of the queued data chunks (until the window fills).

So if I send a message with MSG_MORE set (on an idle connection)
SCTP_XMIT_DELAY is returned and a message isn't sent.

I now send a second small message, this time with MSG_MORE clear.
The message is queued, then the code looks to see if it can send anything.

sctp_packet_can_append_data() is called for the first queued chunk.
Since it has force_delay set SCTP_XMIT_DELAY is returned and no
message is built.
The second message isn't even looked at.

> What MSG_MORE flag actually does is ignore inflight == 0 and
> sctp_sk(asoc->base.sk)->nodelay to delay the chunks, but still
> it has to respect the original logic (like !chunk->msg->can_delay
> || !sctp_packet_empty(packet) || ...)
> 
> To delay the chunks with MSG_MORE set even when inflight is 0
> it especially important here for users.

I'm not too worried about that.
Sending the first message was a cheap way to ensure something got
sent if the application lied and didn't send a subsequent message.

The change has hit Linus's tree, I'll should be able to test that
and confirm what I think is going on.

	David

Powered by blists - more mailing lists

Powered by Openwall GNU/*/Linux Powered by OpenVZ