On Tue, Mar 31, 2026 at 02:10:09PM -0600, Shuah Khan wrote:
> On 3/9/26 21:14, Yi Lai wrote:
> > The clone3_cap_checkpoint_restore.c selftest defined its own 'struct
> > libcap' to manually cast and manipulate the internal memory layout of
> > the opaque 'cap_t' type.
> >
> > Starting with libcap v2.60 (commit aca076443591 "Make cap_t operations
> > thread safe"), a '__u8 mutex' was added to the beginning of the internal
> > 'struct _cap_struct'. This structural shift breaks the selftest's custom
> > struct layout assumptions, causing capability corruption and leading to
> > failures during test execution.
> >
> > Fix this by removing the custom struct definition and use the standard
> > cap_set_flag() API.
>
> What king of failures are you seeing? Can you send the failures from
> the run?
>
I can reproduce the failure on CentOS Stream 10 with libcap 2.69:
TAP version 13
1..1
# Starting 1 tests from 1 test cases.
# RUN global.clone3_cap_checkpoint_restore ...
# clone3() syscall supported
# clone3_cap_checkpoint_restore.c:151:clone3_cap_checkpoint_restore:Child has
PID 10647
cap_set_proc: Operation not permitted
# clone3_cap_checkpoint_restore.c:160:clone3_cap_checkpoint_restore:Expected
set_capability() (-1) == 0 (0)
# clone3_cap_checkpoint_restore.c:161:clone3_cap_checkpoint_restore:Could not
set CAP_CHECKPOINT_RESTORE
# clone3_cap_checkpoint_restore: Test terminated by assertion
# FAIL global.clone3_cap_checkpoint_restore
not ok 1 global.clone3_cap_checkpoint_restore
# FAILED: 0 / 1 tests passed.
# Totals: pass:0 fail:1 xfail:0 xpass:0 skip:0 error:0
After applying the proposed fix patch that uses standard libcap APIs,
the test passes in my test environment. It also passes on CentOS Stream
9 with libcap 2.48 (<2.60).
Regards,
Yi Lai
> >
> > Fixes: 1d27a0be16d6 ("selftests: add clone3() CAP_CHECKPOINT_RESTORE test")
> > Signed-off-by: Yi Lai <[email protected]>
> > ---
>
> thanks,
> -- Shuah
