In fact, this is not a regression: The test case previously failed not because
of the invalid content on the second line, but because of the trailing newline
on the first one - the BPF parser never read more than a single line from the
provided bytecode file.

What has changed with the rebase is that the parser now removes any trailing
newline from that first line. This is a good thing per se, becase editors
commonly add that when editing a file (even if it contains just a single line)
anyway.

So in this particular case, the test case should be adjusted to append the
invalid bytecode part to the same line, not write a new line to the file.

Nevertheless I have submitted a patch upstream which enables the parser to
detect consecutive lines of bytecode and treat them correctly (which
implicitly also allows to split bytecode files over multiple lines):

http://marc.info/?l=linux-netdev&m=150167869213237&w=2

Cheers, Phil

Upstream accepted my patch:

commit 3da3ebfca85b8f1e8252b898453d8cb383c5c398
Author: Phil Sutter <phil>
Date:   Wed Aug 2 14:57:56 2017 +0200

    bpf: Make bytecode-file reading a little more robust
    
    bpf_parse_string() will now correctly handle:
    
    - Extraneous whitespace,
    - OPs on multiple lines and
    - overlong file names.
    
    The added feature of allowing to have OPs on multiple lines (like e.g.
    tcpdump prints them) is rather a side effect of fixing detection of
    malformed bytecode files having random content on a second line, like
    e.g.:
    
    | 4,40 0 0 12,21 0 1 2048,6 0 0 262144,6 0 0 0
    | foobar
    
    Cc: Daniel Borkmann <daniel>
    Signed-off-by: Phil Sutter <phil>
    Acked-by: Daniel Borkmann <daniel>

Hi Phil,

unfortunately I found a regression in the patch. It has stopped to work on correct bytecode file. See the reproducer I have attached.

iproute-4.11.0-1.el7a (old)

# cat /tmp/bytecode.txt 
7,40 0 0 12,21 0 4 2048,37 3 0 1500,48 0 0 14,21 0 1 66,6 0 0 262144,6 0 0 0,
# tc filter add dev TestIface parent 1:0 bpf run bytecode "$(cat /tmp/bytecode.txt)" flowid 1:11
# tc filter add dev TestIface parent 1:0 bpf run bytecode-file /tmp/bytecode.txt flowid 1:13


iproute-4.11.0-2.el7a (new)

# cat /tmp/bytecode.txt 
7,40 0 0 12,21 0 4 2048,37 3 0 1500,48 0 0 14,21 0 1 66,6 0 0 262144,6 0 0 0,
# tc filter add dev TestIface parent 1:0 bpf run bytecode "$(cat /tmp/bytecode.txt)" flowid 1:11
# tc filter add dev TestIface parent 1:0 bpf run bytecode-file /tmp/bytecode.txt flowid 1:13
Real program length exceeds encoded length parameter!

Created attachment 1319118 [details]
Regression reproducer.

Bug found, seems to happen on certain architectures only. Fix submitted upstream:
https://marc.info/?l=linux-netdev&m=150401943515102&w=2

Since the problem described in this bug report should be
resolved in a recent advisory, it has been closed with a
resolution of ERRATA.

For information on the advisory, and where to find the updated
files, follow the link below.

If the solution does not work for you, open a new bug report.

https://access.redhat.com/errata/RHEA-2017:3172