Applied patch from Gernot Payer:
================================
Hi all,
in recent runs I experienced unexpected failures with these testcases and
found out that subtestcases in these programs have some bugs.
mincore01
---------
One test case tests, whether with parameter length < 0 EINVAL is returned.
However this parameter has type size_t, which is unsigned.
The kernel only returns EINVAL if start + length < start (overflow). Otherwise
this syscall will return ENOMEM (unmapped virtual memory in
[start,start+length) ) or might even succeed.
Because of this uncertainty, I dropped this subtestcase.
shmget02
--------
One sub test case tests what happens, if a nonexisting key is used without
flag IPC_CREAT. In this testcase it is assumed that -1 is never used as a
key. However -1 is a perfectly legal key, so it might exist.
My patch tests for a nonexisting key, which can be used in this test case.
capset02
--------
Subtestcase 4 assumes that capset cannot be used at all by a normal process,
because capability CAP_SETPCAP is missing. However with current 2.6.x kernels
a root process can set all capabilities it already owns to itself (i.e.
everything except SET_PCAP).
My patch now changes two things:
- a child process is created which does nothing else than sleeping
- subtestcase 4 uses capset on the child process
mfg
Gernot
3 files changed