Re: kern/38374: fdfile leak

[Greg Oster <oster%cs.usask.ca@localhost>]

The following reply was made to PR kern/38374; it has been noted by GNATS.

From: Greg Oster <oster%cs.usask.ca@localhost>
To: gnats-bugs%NetBSD.org@localhost
Cc: yamt%NetBSD.org@localhost
Subject: Re: kern/38374: fdfile leak 
Date: Wed, 09 Apr 2008 11:22:44 -0600

 YAMAMOTO Takashi writes:
 > The following reply was made to PR kern/38374; it has been noted by GNATS.
 > 
 > From: yamt%mwd.biglobe.ne.jp@localhost (YAMAMOTO Takashi)
 > To: gnats-bugs%NetBSD.org@localhost
 > Cc: ad%NetBSD.org@localhost, gnats-admin%netbsd.org@localhost, 
 > netbsd-bugs%netbsd.org@localhost,
 >         ad%netbsd.org@localhost
 > Subject: Re: kern/38374: fdfile leak
 > Date: Wed,  9 Apr 2008 07:50:33 +0900 (JST)
 > 
 >  > The following reply was made to PR kern/38374; it has been noted by GNATS.
 >  > 
 >  > From: Thomas Klausner <wiz%NetBSD.org@localhost>
 >  > To: gnats-bugs%NetBSD.org@localhost
 >  > Cc: yamt%NetBSD.org@localhost
 >  > Subject: Re: kern/38374: fdfile leak
 >  > Date: Tue, 8 Apr 2008 23:24:32 +0200
 >  > 
 >  >  Greg suggests I add the following information to the PR.
 >  >  
 >  >  With the bulk build interrupted (CTRL-Z) and only a few userland
 >  >  programs running (terminals, X server, screen, amarok),
 >  >  # vmstat -m | grep fdfile && fstat | wc
 >  >  says:
 >  >  vmstat: Kmem statistics are not being gathered by the kernel.
 >  >  fdfile        64     5665    0     4637    22     5    17    17     0 inf
 >     0
 >  >       573    5041   42966
 >  >  
 >  >  4637 + 573 = 5210, which is a few hundred off from 5665.
 >  >  Not sure if the measurement is correct though, but if it is, there
 >  >  might be another leak somewhere or the fix is incomplete.
 >  >   Thomas
 >  
 >  - it's normal that the number of allocated fdfile is not same as
 >    the number of open files.
 >  
 >  - you should subtract the number of cached objects.
 >    ("vmstat -C|grep fdfile|awk '{print $3*$4}'" + some more.)
 
 I think yamt's patch should be checked in... Could someone please do 
 so?  Thanks.
 
 (I can check it in if folks don't have time...)
 
 Later...
 
 Greg Oster