on ppc64le credentials are wrongly parsed/obtained

[judovana]

Hello!

When running on ppc64le, the test

jnr-unixsocket/src/test/java/jnr/unixsocket/CredentialsFunctionalTest.java

Line 79 in 0e180ee

assertEquals("Current PID should match client credentials",

fails correctly, expecting pid, but getting 1.
If the test would continue, it would fail anyway, as all three pid, uid and gid are parsed wrongly during credential initialisation

linux x linux
myPid: 42170 # three real values
myUid: 1001
myGid: 1001

cc uid: 16383 # wrong client credentials
cc pid: 1
cc gid: -2070423904
sc uid: 16383 # samely wrong server credentials
sc pid: 1
sc gid: -2070421584

The values above are exemplar. They differs, canbe positive, but error is obvious.

I had debugged it into:

jnr-unixsocket/src/main/java/jnr/unixsocket/Credentials.java

Line 73 in 0e180ee

static Credentials getCredentials(int fd) {

Which leads to jnr-ffi 's : https://github.com/jnr/jnr-ffi/blob/05cbacc8528e85b1dbff30be5b88cde4c6a9c6a7/src/main/java/jnr/ffi/Struct.java#L865 wehre it leaves my current knowlege a bit.

Suspicion is going to endians or basic size type. Will continue digging.

Note, I had run all jnr-* tests on this ppc64le machine and no other projects (including ffi) had any issues. Only this one unixsocket's CredentialsFunctionalTest

[judovana]

Hmm, the aliases in ffi
are weird:

uid/gid:
uint32
uint
6

but pid:
sint32
sint
5

As read in: https://github.com/jnr/jnr-ffi/blob/05cbacc8528e85b1dbff30be5b88cde4c6a9c6a7/src/main/java/jnr/ffi/Struct.java#L882

... and are correct as by https://stackoverflow.com/a/1922775/3048996

From time to time I can see on that machine also: pid of 65735 > 2^16 but afaik it have no meaning as __s32 can be much bigger (?)

[headius]

I would also suspect we are not handling endianness correctly.

I do not currently have a way to test on ppc64le but I will try to help as best I can!

[judovana]

The only trace on ppc x endians I found was: https://github.com/jnr/jnr-ffi/blob/05cbacc8528e85b1dbff30be5b88cde4c6a9c6a7/src/main/java/jnr/ffi/Platform.java#L221
However if that would be wrong, the jff would be dead, isnt it? Or maybe this is not followed just unix-socket?

Tahnx!