Modify

Opened 21 months ago

Last modified 21 months ago

#22313 new defect

Broadcast packets gettting dropped when using wpa_supplicant

Reported by: xback Owned by: developers
Priority: normal Milestone: Bugs Paradise
Component: packages Version: Chaos Calmer 15.05
Keywords: Cc:

Description

The supplied hostapd package contains a bug were ~80% of the UDP Broadcast packets (ex. OLSR) get dropped randomly in IBSS mode.

Manually upgrading the hostapd package to the latest commit from wpa_supplicant GIT solved these issues.

Changes done to upgrade the package to the latest wpa_sup commit:

  • OWRT Pkg Makefile: Rename PKG_VERSION:=2016... to PKG_VERSION:=master
  • OWRT Pkg Makefile: Rename PKG_REV:=... to f4830bed661f4adff51f50a0d37c64ceb748e780
  • OWRT Pkg Makefile: Remove/Comment CONFIG_ACS flag (line 67)
  • Remove all patches in Pkg Patches folder

Basically, this package could use an upgrade to a newer version.

If more info is required, please let me know.

Thanks,

xb

Attachments (0)

Change History (5)

comment:1 Changed 21 months ago by anonymous

Yes, do you know which exact wpa_supplicant commi fixed the issue?
(bonus points for finding the commit that broke it). You may want
to look into "git bisect" for this. I don't think that you can
simply update to a later wpa_supplicant without it breaking
something else... So having just a "minimal version" and backport
it would be the way to go.

comment:2 Changed 21 months ago by anonymous

Also, post your findings on openwrt-devel@…
(including what hardware you use and relevant extracts from
the configs (/etc/config/wireless, /etc/config/network, ...)

Bonus points: if you look and find a discussion about it on
the hostapd/wpa_supplicant's list hostap@….
Again, since this was fixed, there must have been a patch
with a description and maybe a post from Jouni Malinen.

(Again: a simple call for an updated version is not enough,
since people will just say: "Well use the Trunk" then...).
You have to document the regression and fix properly. And
if you want it merged quickly: make a proper patch and publish
it too. (Again, make it easy to verify your claims, make
testable and robust patches and get a bit involved in fixing
this - after that's done you can stick around or continue
with your own projects).

comment:3 Changed 21 months ago by anonymous

How do I know I am affected by this? I have wpad-mini 2016-01-15-2, is that affected too?

comment:4 Changed 21 months ago by xback

This ticket can be closed.

I found the rootcause:

  • I have a single thread fetching incoming wifi broadcast packets (own OLSR model application)
  • When a packet arrived, occasionally the RF stats are gathered using netlink
  • It seems now that my netlink return buffer was a few bytes to small, causing bad reply and my thread spinning to fetch for netlink data until it timed out and return.

During this period the incoming HW queue got filled up .. and dropped some packets.

Updating some packages caused the netlink return buffer to be a few bytes smaller and fit in the buffer.

Simply increasing the buffer size, and adding some code to dynamically increase the buffer size on full-detection solved all my issues.

Apologies if anyone lost time reading/replying to this.

Thanks again,

xback

comment:5 Changed 21 months ago by anonymous

What is he talking about?

Add Comment

Modify Ticket

Action
as new .
Author


E-mail address and user name can be saved in the Preferences.

 
Note: See TracTickets for help on using tickets.