[LTP] [PATCH v2 10/13] getdtablesize01: Handle ENFILE errno

[Cyril Hrubis]

Hi!
> > When process has exhausted the maximal number of file descriptors it can
> > open we are supposted to fail with EMFILE, when that happened to the
> > whole system before we hit the per process limit we get ENFILE. So I was
> > suggesting that we may as well check that we were able to open expected
> > number of file descriptors and that the last open() failed with EMFILE.
> 
> But isn't that we check already ? I mean, the test already gets the limit
> on number of open files for this process and tries to open one less fd.

As far as I can tell it tries to open as much fds as is the limit, the
loop is kind of silly since it starts at 1 and the condition is <= but
that should call open the open number of limit times.

> So I don't think we should check that error. If we get that error, something
> else is wrong in the accounting and should be treated as an error.

But the loop never finishes and the open() always fails with EMFILE
because process has at least three fds opened when it starts (the
stdin/stdout/stderr are inherited from the parent). Hence the loop is
always broken on fd == -1.

I guess that we may as well change the loop to infinite one, which would
be less confusing in this case.

-- 
Cyril Hrubis
chrubis@suse.cz