Re: Kernel bugzilla 198569: fanotify_mark() and FAN_Q_OVERFLOW

[Amir Goldstein <amir73il@gmail.com>]

On Thu, Apr 23, 2020 at 3:42 PM Michael Kerrisk (man-pages)
<mtk.manpages@gmail.com> wrote:
>
> Hello Amir, Jan,
>
> On Thu, 23 Apr 2020 at 14:19, Amir Goldstein <amir73il@gmail.com> wrote:
> >
> > On Thu, Apr 23, 2020 at 2:23 PM Jan Kara <jack@suse.cz> wrote:
> > >
> > > Hi Michael!
> > >
> > > On Thu 23-04-20 12:36:26, Michael Kerrisk (man-pages) wrote:
> > > > Would you be able to take a look at
> > > > https://bugzilla.kernel.org/show_bug.cgi?id=198569
> > > >
> > > > It relates to some text you added to the fanotify_mark(2) manual page:
> > > >
> > > >        FAN_Q_OVERFLOW
> > > >               Create an event when an overflow of the event queue occurs.
> > > >               The size of the event queue is limited to 16384 entries  if
> > > >               FAN_UNLIMITED_QUEUE is not set in fanotify_init(2).
> > > >
> > > > This was in the following commit
> > > >
> > > > [[
> > > > commit 5d730f864a6603b090cd1078668cede05d02b8c4
> > > > Author: Heinrich Schuchardt <xypron.glpk@gmx.de>
> > > > Date:   Tue Nov 8 23:13:38 2016 +0100
> > > >
> > > >     fanotify_mark.2: Mention FAN_Q_OVERFLOW
> > > >
> > > >     To receive overflow events it is necessary to set this bit
> > > >     in fanotify_mark().
> > > > ]]
> > > >
> > > > As far as I can see, FAN_Q_OVERFLOW (test program, reading the kernel
> > > > source) is only an output flag. But on the other hand, I know you are
> > > > generally careful, so I wonder if something changed (though, at a
> > > > quick glance, I could not see evidence that it has).
> > >
> > > Yeah, the manpage is wrong AFAICT. FAN_Q_OVERFLOW is not accepted in the
> > > input mask. It is only output event flag.
> > >
> >
> > Right.
> >
> > Note that fanotify.7 documents FAN_Q_OVERFLOW as part of event output mask
> > so perhaps the FAN_Q_OVERFLOW entry for fanotify_mark.2 input mask should
> > just be removed?
>
> Yes, I'll just remove that piece from fanotify_mark.2.
> >
> > Similarly, the input only flag FAN_EVENT_ON_CHILD is documented in
> > fanotify_mark.2, but not in fanotify.7.
> >
> > FAN_ONDIR has been input only until v5.1 and since then it can also be
> > in output mask for group with FAN_REPORT_FID.
> > This is documented in fanotify_mark.2, but not in fanotify.7.
>
> I'm overlooking something: where is this detail documented in
> fanotify_mark.2? (I mean, I see mention of FAN_ONDIR, but no mention
> of a change in v5.1.)

fanotify_mark.2:
The FAN_ONDIR flag is reported in an event mask only if the fanotify_fd
file descriptor has been initialized with the flag FAN_REPORT_FID.

fanotify_init.2:
FAN_REPORT_FID (since Linux 5.1)

>
> And would you be willing to write a small patch for fanotify.7 :-) ?
>

I guess I was asking for it ;-)
I have some more work cut out for me on this man page for v5.7 or later.
Adding yet another fanotify_init flag, so I'll make this change part of my
next series.

Thanks,
Amir.