[LTP] [PATCH] aio-stress: fix opened fd leak

[Jan Stancek]

On Thu, Jan 16, 2025 at 4:50 PM Cyril Hrubis <chrubis@suse.cz> wrote:
>
> Hi!
> > In cases where worker completes specified number of `iterations` before
> > oper completes all `total_ios`, such oper remains on active_list.
> > This is a problem because cleanup only walks over finished list,
> > and closes fds only for items on this list. So it's possible for
> > test to exhaust ulimit for open files and fail:
> >
> > $ ./aio-stress -a1 -I100 -o2 -r4 -f1 -d1^
> > aio-stress.c:1347: TPASS: Test passed
> > aio-stress.c:1285: TINFO: starting with random write
> > aio-stress.c:1296: TINFO: file size 1024MB, record size 0KB, depth 1, I/O per iteration 1
> > aio-stress.c:1298: TINFO: max io_submit 1, buffer alignment set to 4KB
> > aio-stress.c:1300: TINFO: threads 1 files 1 contexts 1 context offset 2MB verification off
> > aio-stress.c:1314: TBROK: open(file0.bin,1052738,0600) failed: EMFILE (24)
>
> Looks like the test has many options and some combinations may set
> things in a way that does not make much sense.

I think the options I saw in failures were defaults, the "-a1" above I
added to reproduce it faster.

>
> > Clean/free also items from active_list. Also don't let `status` from cleanup
> > be set to zero once it becomes non-zero.
> >
> > Signed-off-by: Jan Stancek <jstancek@redhat.com>
> > ---
> >  testcases/kernel/io/ltp-aiodio/aio-stress.c | 7 ++++++-
> >  1 file changed, 6 insertions(+), 1 deletion(-)
> >
> > diff --git a/testcases/kernel/io/ltp-aiodio/aio-stress.c b/testcases/kernel/io/ltp-aiodio/aio-stress.c
> > index 419fc2cd25a4..e84308d1d0e8 100644
> > --- a/testcases/kernel/io/ltp-aiodio/aio-stress.c
> > +++ b/testcases/kernel/io/ltp-aiodio/aio-stress.c
> > @@ -1122,7 +1122,12 @@ restart:
> >       while (t->finished_opers) {
> >               oper = t->finished_opers;
> >               oper_list_del(oper, &t->finished_opers);
> > -             status = finish_oper(t, oper);
> > +             status = finish_oper(t, oper) ? : status;
> > +     }
> > +     while (t->active_opers) {
> > +             oper = t->active_opers;
> > +             oper_list_del(oper, &t->active_opers);
> > +             status = finish_oper(t, oper) ? : status;
> >       }
>
>
> I guess that this works as a minimal fix.
>
> Reviewed-by: Cyril Hrubis <chrubis@suse.cz>
>
> --
> Cyril Hrubis
> chrubis@suse.cz
>