Bug 85460 – glibc test suite broken, all needs fixing

• Product: Gentoo Linux
• Status: RESOLVED DUPLICATE

[Description ferret 2005-03-15 22:50:50 UTC]

Reproducing:
1. mount -t reiserfs /dev/blah /tmp -o noatime
2. FEATURES=makecheck emerge glibc

The test is definitely in the latest glibcs (sys-libs/glibc-2.3.4.20050125-r1
and above), not sure if it's in earlier ones.

This is probably true for more systems than you'd think, since it's the default
for the handbook.

glibc does a test to see if atimes are enabled (in libio/tst-atime.c).  It first
tries to check if noatime is set as a mount option, which is clever since the
test is guaranteed to fail when noatime is on.  What isn't clever is that it
uses statvfs(2) to do this, despite the fact that the man page mentions that any
of the given values can be undefined.

I have a program that can print this struct, available here:
http://linux.ferret.vze.com/statvfs-tester.c
For a fs with noatime enabled, sv.f_flag should bitwise OR with 1024 (i.e. the
tenth bit from the right should be 1)

Suggested fix: no idea.  It isn't a bug in reiserfs, really, since the man page
says populating the struct isn't necessary, but it would be nice for reiserfs to
do this.  So glibc's test suite is slightly broken, perhaps it needs to display
a warning and pass the test if sv.f_flag == 0, or maybe it needs to find out by
some other method if the filesystem is mounted with noatime option.

kernel version is 2.6.10, reiserfs version used is 3.6

[Comment 1 ferret 2005-03-16 00:55:28 UTC]

*** Bug 85459 has been marked as a duplicate of this bug. ***

[Comment 2 SpanKY 2005-03-16 15:36:36 UTC]

known issue ... older glibc's had an ugly ugly hack in them to that tried to re-mount w/out noatime, but that was quickly punted

[Comment 3 ferret 2005-03-17 13:19:05 UTC]

I don't see any great ways of resolving this without pretty much disabling the test entirely.  There just is no way of determining whether a filesystem has no mount options enabled, or has just not bothered to inform statvfs.  I've mentioned this to some reiserfs guys who are going to look into making the statvfs work properly.  other than that I'm going to resolve this as LATER (I think that's the right resolution to use?) since there are a lot of tests AFTER this one that fail too.  Maybe some day when I'm bored I'll fix the WHOLE test suite.

[Comment 4 Kevin F. Quinn (RETIRED) 2005-10-01 11:55:14 UTC]

re-opening to mark as dup of bug #86515, bringing all the "glibc fails atime
test" bugs together

[Comment 5 Kevin F. Quinn (RETIRED) 2005-10-01 11:55:51 UTC]

marking dup as noted above :)

*** This bug has been marked as a duplicate of 86515 ***