This commit changes how fko_new() deals with FKO context initialization
to not set ctx->initval back to zero (uninitialized) imediately after
calling each fko_set_... function and before checking the fko_set_... return
value. The reason for this change is that fko_destroy() checks for
context initialization via ctx->initval before calling free() against
any heap allocated context member. So, if fko_set_... returns an error,
fko_destroy() (previous to this commit) would have no opportunity to
free such members.
This bug was found with fault injection testing provided by libfiu
together with valgrind. Specifically the following test suite command
exposes the problem (from the test/ directory):
./test-fwknop.pl --enable-complete --include "fault injection.*libfko"
In the resulting output/2.test file valgrind reports the following:
==27941== LEAK SUMMARY:
==27941== definitely lost: 264 bytes in 1 blocks
==27941== indirectly lost: 28 bytes in 3 blocks
==27941== possibly lost: 0 bytes in 0 blocks
==27941== still reachable: 1,099 bytes in 12 blocks
==27941== suppressed: 0 bytes in 0 blocks
After this commit is applied, this changes to:
==7137== LEAK SUMMARY:
==7137== definitely lost: 0 bytes in 0 blocks
==7137== indirectly lost: 0 bytes in 0 blocks
==7137== possibly lost: 0 bytes in 0 blocks
==7137== still reachable: 1,099 bytes in 12 blocks
==7137== suppressed: 0 bytes in 0 blocks
Note that 'definitely lost' in valgrind output means there is a real
memory leak that needs to be fixed whereas 'still reachable' is most
likely not a real problem according to:
http://valgrind.org/docs/manual/faq.html#faq.deflost
This is a significant commit to add the ability to leverage libfko fault
injections from both the fwknop client and server command lines via a
new option '--fault-injection-tag <tag name>'. This option is used by
the test suite with the tests/fault_injection.pl tests.
This commit adds support for diff'ing before and after gcov/lcov results
to see when new function/line coverage is added by the test suite. Here
is an example of its output:
Sun Jun 1 22:28:00 2014 CMD: ./coverage_diff.py
[+] Coverage: /home/mbr/git/fwknop.git/server/config_init.c
[+] new 'fcns' coverage: usage()
[+] new 'lines' coverage: 1015
[+] new 'lines' coverage: 1017
[+] new 'lines' coverage: 1019
[+] new 'lines' coverage: 1059
[+] new 'lines' coverage: 979
[+] Coverage: /home/mbr/git/fwknop.git/server/fw_util_iptables.c
[+] new 'lines' coverage: 560
[+] new 'lines' coverage: 561
This commit adds a couple of suppressions for known issues that valgrind
finds in libcap, and then makes a significant change to how the test
suite deals with any valgrind errors (in --enable-valgrind mode) that
are outside of these suppressions. That is, any new valgrind errors
that are discovered will cause the test that triggers them to fail.
Previous to this commit, the final valgrind "flagged functions" test
attmpted to do this by comparing valgrind output across test runs. This
worked well enough for a while, but this latest commit enforces a
stricter stance for valgrind validation of the fwknop code base.
