tree e9194dbcdfd3504ac0e720abe0b1efa1e3740ae8
parent bf3cf5c8bcbfb5208b14c04ce5facf2d4f27af1e
author Vadim Yanitskiy <axilirator@gmail.com> 1532632996 +0700
committer Harald Welte <laforge@gnumonks.org> 1532696753 +0000

BTS: fix: pad LAPDm frames before sending via L1CTL

Thanks to Stefan Sperling, a critical bug was discovered in trxcon.
The problem was that length of LAPDm frames was not checked before
passing them to the libosmocoding API. So, if a received LAPDm
frame is shorter than expected (i.e. 23 bytes), then:

  - in case of xCCH, there was a heap overflow (detected by ASAN),
    so a short frame has been encoded together with some garbage
    outside the primitive buffer...

  - in case of FACCH, as the length != 23, a frame was recognised
    as a speech frame, and also encoded together with some garbage.

Since the bug is fixed (OS#3415), some TTCN-3 BTS tests started
to fail, because most likely it was assumed that trxcon would
pad the frames automatically, but it doesn't and shouldn't.

Let's automatically pad LAPDm frames with 0x2b before sending.

Change-Id: I16cba4e4179456bebabf0638760af011a27fd333
Related: OS#3418