Re: coreutils-8.14, "rm -r" fails with EBADF

[Jim Meyering]

Joachim Schmitz wrote:

>> From: Paul Eggert [mailto:address@hidden
>> Sent: Thursday, December 15, 2011 7:07 PM
>> To: Joachim Schmitz
>> Cc: Jim Meyering; address@hidden; address@hidden
>> Subject: Re: coreutils-8.14, "rm -r" fails with EBADF
>>
>> On 12/15/11 08:28, Jim Meyering wrote:
>> > If you can debug it further to narrow down what/how it is failing,
>> > eventually we'll find the cause.
>>
>> One way to do that might be to put a breakpoint on all the following
> functions,
>> and to let us know what their arguments are when called, and what they
>> return.  You also might try to compile coreutils with debugging turned
> off, so
>> that functions aren't inlined.
>
> I guess you meant turn of optimization rather than debugging?
>
>> diropen
>> fts_build
>> fts_safe_changedir
>> restore_initial_cwd
>> fchdir
>> cwd_advance_fd
>
> OK here we go:
> address@hidden:/home/jojo/Floss/coreutils-8.14/src $ run -debug -inspect=on
> ./rm -r /tmp/foo
...
> Stepping into fts_build shows that the dup(dir_fd) in line 1385 fails, with
> errno being set to EBADF
>
> * 1375              if (continue_readdir)
> (eInspect 1,840):
> * 1381              else if (nlinks || type == BREAD) {
> (eInspect 1,840):
> * 1382                      if (ISSET(FTS_CWDFD))
> (eInspect 1,840):
> * 1384                          dir_fd = dup (dir_fd);
> (eInspect 1,840):p dir_fd
> $3 = 3
> (eInspect 1,840):next
> * 1385                          if (0 <= dir_fd)
> (eInspect 1,840):p dir_fd
> $4 = -1
...
> So... it seems we can't dup() an fd for a directory here!

Right.
Your config.h file suggests you're not using a gnulib dup replacement,
so I suspect a bug in your system's dup implementation.

I thought gnulib had pretty good dup* test coverage,
but see that no dup* test uses a directory file descriptor.