This backport fixes 'kmtest_.exe SeQueryInfoToken'
on all testers: VBox x86, KVM x86, WHS x86, Win2003_x64.
And according to Thomas description may also prevent a buffer overrun when executing that formerly broken test.
Afterwards all 76 tests of this suite do complete on all those builders.
Before the patch only 74 of those tests succeeded, 2 failed.
The fix is a squashed backport of the following 6 commits from Thomas Faber:
0.4.16-dev-11-g 44bdafa17e [KMTESTS:SE] Fix failing tests (#5308)
0.4.16-dev-10-g bf6af0f52e [NTOS:SE] Mark output parameters as such (#5308)
0.4.16-dev-9-g 156053cafd [NDK] Match AUX_ACCESS_DATA definition with publicly available version. - if you allocated only sizeof(AUX_ACCESS_DATA), the test would crash with a 4 byte buffer overflow. (#5308)
0.4.16-dev-8-g ff410211e9 [KMTESTS:SE] Don't modify internal data structure, this might cause buffer overrun (#5308)
0.4.16-dev-7-g 206df96bc4 [KMTESTS:SE] Correctly allocate PrivilegeSet buffers (#5308)
0.4.16-dev-6-g 64a6bd4c3e [KMTESTS:SE] Avoid use of uninitialized pool and hardcoded offsets (#5308)
WHS x86 before-and-after-state, the after-test had a few fixes from Timos unrelated PR7343 inside unfortunately:
https://reactos.org/testman/compare.php?ids=97640,97871
(This is added to prove the test being wrong)
I tested it also successfully on my local 2k3sp2 x86 with the releases/0.4.15 afterstate, built with RosBEWin2.2.2 GCC8.4.0dbg x86.
Win2003_x64 0.4.16-dev-11-g44bdafa at 2024-09-12 15:19 (after-state):
https://reactos.org/testman/compare.php?ids=97791
0.4.16-dev-5-g2913ef5 vs. 0.4.16-dev-11-g44bdafa vs. 0.4.16-dev-23-g53b304e:
VBox x86 https://reactos.org/testman/compare.php?ids=97795,97806,97877
0.4.16-dev-5-g2913ef5 vs. 0.4.16-dev-20-g144a8b5 vs. 0.4.16-dev-21-g2af6fd4:
KVM x86 https://reactos.org/testman/compare.php?ids=97793,97855,97856
Since we do touch the NTOS and NDK here the fix is not guaranteed to be side-effect-free,
but since we are so early in the RC-phase, I dared to pick it, especially since the alternative would have
been to disable the test altogether in the releases/0.4.15 which would have been a pity, if we can also have it all green everywhere.