tst-ttyname failure

[Adhemerval Zanella]

On 17/11/2017 19:28, Florian Weimer wrote:
> I need the two attached patches in order to get tst-ttyname to succeed when running on Fedora as a non-root user.
> 
> info:Â  entering chroot 1
> info:Â Â Â  testcase: basic smoketest
> info:Â Â Â Â Â  ttyname: PASS {name="/dev/pts/5", errno=0}
> info:Â Â Â Â Â  ttyname_r: PASS {name="/dev/pts/5", ret=0, errno=0}
> error: ../sysdeps/unix/sysv/linux/tst-ttyname.c:122: write (setroups, "deny"): Operation not permitted
> info:Â  entering chroot 2
> error: ../sysdeps/unix/sysv/linux/tst-ttyname.c:122: write (setroups, "deny"): Operation not permitted
> error: 2 test failures
> 
> I get the same failure when running the test on kernel-3.10.0-768.el7.x86_64 (there I tested as root).
> 
> This is on top of master with commit ce003e5d4cd94c5380699b0dadeaaf825813afbe (support_become_root: Enable file creation in user namespaces).  The test failure was present before I pushed that change.
> 
> Would you please double-check that they do not break the test on whatever distributions you tested on and do not interfere with the test objective?
> 
> Thanks,
> Florian

I am still seeing failures on Ubuntu 16.04 (4.4.0-71-generic) on master:

info:  entering chroot 1
info:    testcase: basic smoketest
info:      ttyname: PASS {name="/dev/pts/38", errno=0}
info:      ttyname_r: PASS {name="/dev/pts/38", ret=0, errno=0}
error: ../sysdeps/unix/sysv/linux/tst-ttyname.c:285: mount ("devpts", "dev/pts", "devpts", MS_NOSUID|MS_NOEXEC, "newinstance,ptmxmode=0666,mode=620") == 0: Invalid argument
info:  entering chroot 2
info:    testcase: basic smoketest
info:      ttyname: PASS {name="/dev/pts/38", errno=0}
info:      ttyname_r: PASS {name="/dev/pts/38", ret=0, errno=0}
error: ../sysdeps/unix/sysv/linux/tst-ttyname.c:357: mount ("devpts", "dev/pts", "devpts", MS_NOSUID|MS_NOEXEC, "newinstance,ptmxmode=0666,mode=620") == 0: Invalid argument

I noted, at least on my environment, it requires running as a more privileged user
to avoid the above failure. Is it the expected behaviour? Should we move it to
xcheck and add the a possible requirement [1]?

[1] https://sourceware.org/glibc/wiki/Testing/Testsuite#Details_about_make_xcheck_specific_tests