libyara/exec: avoid unaliged access to 64bit variable #412
Conversation
The derefence of an unaligned 64bit variable results in a SIGBUS abort on 32bit SPARC. ARMv5 CPUs seem to perform the load but load garbish. This memcpy() workaround forces the compiler to do something that works on even if the data was not properly aligned. For X86 it means no change. ARM on other hand will produce slightly different code depending on the CPU used. While at it, I coverted all (ip + 1) references. Signed-off-by: Sebastian Andrzej Siewior <[email protected]>
|
Did you test YARA extensively on SPARC and ARM? I'm wondering if those are the only places where we have unaligned accesses. |
|
clamav in 0.99 is using yara however they pulled a modified version of it into their source. Before the patch clamav's testsuite failed on sparc: If you have a testsuite or something you want me to run on sparc to verify I would try to give it a try. |
|
There's a test suit in Python, but of course it requires building yara-python too. You can find it here: https://github.com/plusvic/yara-python/blob/master/tests.py |
|
Now we have the same tests in C (https://github.com/plusvic/yara/blob/master/tests/test-rules.c) and you can run them with "make check". Could you verify if the tests pass in SPARC and ARM? |
|
if fails on the sparc and the arm box. ARM says |
|
A lot of work as been done in #1724 that should fix most of the unaligned memory accesses. |
The derefence of an unaligned 64bit variable results in a SIGBUS abort
on 32bit SPARC. ARMv5 CPUs seem to perform the load but load garbish.
This memcpy() workaround forces the compiler to do something that works
on even if the data was not properly aligned. For X86 it means no
change. ARM on other hand will produce slightly different code depending
on the CPU used.
While at it, I coverted all (ip + 1) references.
Signed-off-by: Sebastian Andrzej Siewior [email protected]