Re: kern/46955: process deadlock (tstile) runing amanda sendsize - rename bug in tmpfs ???

[Wolfgang Stukenbrock <wolfgang.stukenbrock%nagler-company.com@localhost>]

Hi,

it looks like that I've a working version for 5.1_STABLE now, but the 
integration I've done is (very) ugly ...

- I took the whole tmpfs code from current.
- Need to modify two other makefiles (rump and ...) because on file in 
tmpfs is gone and another (tmpfs_rename.c) came up.
- need to "import" genfs_rename.c, because most of the implementation is 
now there. (Done by a copy to the tmpfs-folder and an include statement 
 in the tmpfs source file.)
- borrow some other new routines from genfs, because they are needed 
now. (simply copied into the tmpfs sources in order to avoid changes in 
other parts)
- add a second param to VOP_UNLOCK() - added 0 as it was in the old code
- need to add the INTERLOCK flag to the call of vget() - call semantic 
has changed in -current.
- revert KERN_NAME_MAX change by adding a define to NAME_MAX instead of 
the assert "MAXNAMELEN == NAME_MAX"
- borrow some new routines from kern_auth.c by copying into the tmpfs 
sources ...

And here is the main problem with my "fast-and-dirty" integration:
It looks like nearly everything changed in kern_auth.c and some new 
listener lists have been added that are not present in 5.1.x. I think 
they cannot be simply used because the other code does not know then. To 
add this knowledge everything kauth-related must be changed too in order 
to work correctly - to much work ...
So I've took the main auth-methods from the old implemantation useing 
the "gereric" auth routine as done in 5.1.x , but I additionaly needed 
to allow root (uid == 0) explicitly to allow chown calls to get the 
install-root filesystem of the install-kernels up correctly.
(Does this work in -current ???? Is there a listener that returns ALLOW 
if root is trying to chown something? Or has the createion of the device 
nodes in the install-kernels has been changed?)
The auth-part is still not realy tested, but it seems to work as before 
till now.

If you are interested in this work, let me know, I will send it.

But the auth-integration should be reworked again!
Any hints how to revert the new-stuff into the old 5.1.x semantics 
correctly?
The "problem" is the auth-vnode part and all related definitions.


W. Stukenbrock


David Holland wrote:

> The following reply was made to PR kern/46955; it has been noted by GNATS.
>
> From: David Holland <dholland-bugs%netbsd.org@localhost>
> To: gnats-bugs%NetBSD.org@localhost
> Cc: 
> Subject: Re: kern/46955: process deadlock (tstile) runing amanda sendsize -
>  rename bug in tmpfs ???
> Date: Sat, 15 Sep 2012 16:57:53 +0000
>
>  On Fri, Sep 14, 2012 at 08:40:04AM +0000, Wolfgang Stukenbrock wrote:
>   >  I've forgot to look into the Bug-database prior sending this one - sorry.
>   >  
>   >  In -current there seems to be a patch for the rename problem that I've 
>   >  still not integrated in my 5.1.2. (PR 36681).
>   >  
>   >  I'm gooing to do this as next step.
>  
>  That is probably a good idea, but the available patches only fix ffs.
>  
>   >  dholland stated that this fix will solve the rename problem in tmpfs.
>   >  So I think it will solve our problem here too.
>  
>  I don't think there's a netbsd-5 version of the tmpfs rename patches,
>  but I could have missed it going by.
>  
>  (Unfortunately, each filesystem's rename needs to be fixed separately,
>  and they have historically been *all* wrong.)
>  
>  -- 
>  David A. Holland
>  dholland%netbsd.org@localhost
>  


--


Dr. Nagler & Company GmbH
Hauptstraße 9
92253 Schnaittenbach

Tel. +49 9622/71 97-42
Fax +49 9622/71 97-50

Wolfgang.Stukenbrock%nagler-company.com@localhost
http://www.nagler-company.com


Hauptsitz: Schnaittenbach
Handelregister: Amberg HRB
Gerichtsstand: Amberg
Steuernummer: 201/118/51825
USt.-ID-Nummer: DE 273143997
Geschäftsführer: Dr. Martin Nagler, Prof. Dr. Dr. Karl-Kuno Kunze