bug#9620: dd: bogus behavior when interrupted

[Paul Eggert]

This happened with coreutils 8.13 on Fedora 14 x86-64
(coreutils compiled with GCC 4.6.1).  I interrupted
'dd' with control-C, but it didn't respond right away;
instead, it churned away and created the entire output file,
issuing a bogus diagnostic about the input file.  Here's
the transcript:

$ dd if=/dev/zero of=$HOME/junk/zero bs=1024 count=1000000
^C1000000+0 records in
1000000+0 records out
1024000000 bytes (1.0 GB) copied, 20.1583 s, 50.8 MB/s
dd: closing input file `/dev/zero': Bad file descriptor
$ ls -l zero
-rw-r--r-- 1 eggert eggert 1024000000 Sep 27 12:18 zero

The problem with the diagnostic is intermittent.  It usually
does not happen.  Usually, there's simply an unconscionably long
wait between the time I type ^C and the time that dd exits, e.g.:

$ dd if=/dev/zero of=zero bs=1024 count=1000000
^C487034+0 records in
487034+0 records out
498722816 bytes (499 MB) copied, 11.6897 s, 42.7 MB/s

(here I waited about 10 seconds between the time I typed
^C and the time that dd exited).

The filesystem is ext4 atop md (RAID-1).

The same problem occurs with /bin/dd (coreutils 8.5) so
if it is a coreutils bug it's not a new one.

Don't have time to debug this right now but thought I'd
get a bug report into the system.  Quite possibly it is not
a coreutils bug at all, but a kernel bug, but in that case
where do I report it? to a Fedora mailing list?