On Die, 24 Apr 2001, Daniel Phillips wrote:

> > I have a small problem when using the directory notifications. when the
> > directory itself that I watch in my program is deleted, I don't receive an
> > event, although I'd expect a delete event for the "." entry (IMHO). This
> > makes it much easier to detect this relatively rare case, otherwise one
> > would need to add twice as many watchpoints (for every directory its parent
> > directory as well), which is no good solution in my opinion.
> With your patch you are sending the delete event twice, once to the directory 
> being deleted and again to the containing directory (if you have registered 
> for events on it).  That doesn't sound right to me.  

Right. I don't see the problem for that though, instead I consider it the 
actually correct behaviour. 

a) a program that registered on the parent directory wants the event because
  its directory is modified

b) a program that registered on "this" directory wants the event to 
recognize that "its" directory got modified: it got deleted. technically,
the modified timestamp of the directory is changed and its contents 
has changed ( the "." link disappeared). 

Its very simple to distinguish between these two cases (just a stat on the 
file name. if it doesn't succeed, it was case b), otherwise a)). Otherwise, 
you have to register twice as many event handlers. 

> What's really needed here is an "end-of-events" event.

Yes, for compatibility one could add another event type to prevent that an 
existing application can't handle the changed behaviour (which should be 
rare IMHO). I did decide against this because the code is relatively new and 
therefore the legacy-code problem is not that big (IMHO). 

> You are also sending the event with the i_zombie semaphore held, which is 
> probably not a good idea.

Yes, this way you can detect by a stat() call on if the directory itself got 
deleted or not. 

I'm not sure if this is "safe" or not. Is it ?

Would a patch that adds another event type be preferred ? should be no big 
problem to implement if its safe to emit that event on a "zombie" inode. 
Should I come up with a patch for this variant ?


Dirk
-
To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in
the body of a message to [EMAIL PROTECTED]

Reply via email to