bug#26363: [PATCH] tail: 'tail -F dir/file' reverts to polling mode if 'dir' is removed

[Pádraig Brady]

On 04/04/17 09:47, Sebastian Kisela wrote:
> * src/tail.c (tail_forever_inotify):  Add the IN_DELETE_SELF flag when
> creating watch for the parent directory.  After the parent directory
> is removed, an event is caught and then we switch from inotify to
> polling mode.  Till now, inotify has always frozen because it waited for
> an event from a watched dir, which has been already deleted and was not
> added again.
> * tests/tail-2/inotify-dir-recreate.sh: Add a test case.
> * tests/local.mk: Reference the new test.
> * NEWS: Mention the bug fix.
> 
> Reported at https://bugzilla.redhat.com/1283760

Excellent. That looks like a fine fall back.

Not having looked at the C code,
have you considered if another dir/files are already setup with inotify watches,
and if files are deleted but still open?
Other comments below...

> @@ -1628,6 +1629,24 @@ tail_forever_inotify (int wd, struct File_spec *f,
> size_t n_files,
>        ev = void_ev;
>        evbuf_off += sizeof (*ev) + ev->len;
> 
> +      /* If a directory is deleted, IN_DELETE_SELF is emmited
> +         with ev->name of length 0.
> +         We need to catch it, otherwise it would wait forever,
> +         as wd for directory becomes inactive. Revert to polling now.   */
> +      if ((ev->mask & IN_DELETE_SELF) && !ev->len)
> +        {
> +          for (i = 0; i < n_files; i++)
> +            {
> +              if (ev->wd == f[i].parent_wd)
> +                {
> +                  hash_free (wd_to_name);
> +                  error (0, 0,
> +                      _("directory containing watched file was removed"));
> +                  return true;
> +                }
> +            }
> +        }
> +
>        if (ev->len) /* event on ev->name in watched directory.  */
>          {
>            size_t j;

> +
> +# Check for existence of the important message
> +check_tail_output_ ()
> +{
> +  grep "$1" out &>/dev/null ||  fail=1
> +}
> +
> +# Prepare the file to be watched
> +mkdir dir && touch dir/file
> +sleep 1

Why this sleep?

> +timeout 60 tail --follow=name --retry dir/file &>out  & pid=$!
> +sleep 1

We can't depend on absolute sleeps like this. See:
http://www.pixelbeat.org/docs/coreutils-testing.html#backoff
or the existing tail-2/ tests

cheers,
Pádraig