socket error on client...disconnect when payload > 4096 under protocol websocket

[ralight]

migrated from Bugzilla #464811
status RESOLVED severity normal in component Mosquitto for 1.4
Reported in version 1.4 on platform Other
Assigned to: Roger Light

Original attachment names and IDs:

• 464811.patch (ID 253027)

On 2015-04-16 11:40:23 -0400, stephen zeng wrote:

I use paho.javascript-1.0.1 to test mosquitto-1.4.1. And i found that under ws protocol, when the packet > 4096, the broker disconnect client, but fine under tcp protocol.
More specific, i checked func callback_mtqq, this callback only get 4096 size data one time but not all data, then mqtt3_packet_handle return MOSQ_ERR_PROTOCOL err code.
Finally,i figure out that i can change the protocols.rx_buffer_size to fix it, but i prefer config it in mosquitto.conf file, hope you can add this configuration.
GOODLUCK,BYE.

On 2015-04-17 16:26:59 -0400, Roger Light wrote:

Thanks for the report, could you confirm what value you set rx_buffer_size to, and what version of libwebsockets you are using?

On 2015-05-01 09:16:05 -0400, Arjen van Elteren wrote:

Created attachment 253027
patch to fix message size problems for websockets

On 2015-05-01 09:20:45 -0400, Arjen van Elteren wrote:

I can confirm this biug, I used latests mosquitto from git (SHA: 11d0d88) and mqttws31.js.

Default configuration for mosquitto, except enabling the websockets listener.

Messages larger than 4096 get mangled.

I've attached a patch to fix this problem. Libwebsocket will split large messages into several blocks, and although mosquitto does have almost complete handling for this case, it seems to miss a small part.

Simply returning from the <callback_mqtt> function if the message reading is incomplete will fix the problem (see patch).

On 2015-05-04 16:16:39 -0400, Roger Light wrote:

Hi Arjen,

Thanks for the patch. I need some more things from you in order to be able to accept it. The details are here:

http:https://git.eclipse.org/c/mosquitto/org.eclipse.mosquitto.git/tree/CONTRIBUTING.md

The ideal scenario is that you sign the Eclipse CLA. This is described here: http:https://www.eclipse.org/legal/CLA.php

You then need to sign-off on the Certificate of Origin (i.e. state that you are able to make this submission), more details here: https://wiki.eclipse.org/Development_Resources/Contributing_via_Git#via_Bugzilla

I have been authorised to accept very small patches such as this one without a signed CLA being in place. If you would prefer you only need to sign-off on the Certificate of Origin in this bug report. Without this as a minimum I cannot accept the patch.

On 2015-05-06 07:07:07 -0400, Arjen van Elteren wrote:

Hi Roger,

I've signed off on CLA already (I have the CLA button next to my name).

As for the Cerficate of orgin, I don't think I can add the line to the patch,
but I can do so in this comment:

This contribution complies with http:https://www.eclipse.org/legal/CoO.php

Is this sufficient?

Kind regards,

Arjen

On 2015-05-06 07:32:59 -0400, Roger Light wrote:

Hi Arjen,

That's great, thanks very much. I must have looked in the wrong place when I was checking for the CLA tick.

This will be part of 1.4.2, thanks for your contribution.

On 2015-05-06 17:42:17 -0400, Roger Light wrote:

I've actually fixed this in a subtly different way rather than with your patch. It makes a bit more sense this way.

Thanks again for the patch though, it was still very important pointing out where the problem was!