Re: CVS commit: src/sys

[Andrew Doran <ad%netbsd.org@localhost>]

On Sat, Apr 11, 2009 at 11:47:34AM -0400, Christos Zoulas wrote:

> Module Name:  src
> Committed By: christos
> Date:         Sat Apr 11 15:47:34 UTC 2009
> 
> Modified Files:
>       src/sys/dev/dmover: dmover_io.c
>       src/sys/dev/putter: putter.c
>       src/sys/kern: kern_drvctl.c sys_mqueue.c
>       src/sys/net: bpf.c bpfdesc.h if_tap.c
>       src/sys/opencrypto: cryptodev.c
>       src/sys/sys: mqueue.h
> 
> Log Message:
> Fix PR/37878 and PR/37550: Provide stat(2) for all devices and don't use
> fbadop_stat.

The locking is all screwed up or missing. stat() can be expected to report
bad times for these files. I wrote the below to you in private and forgot to
copy the list.

Subject: Re: fix for 37878 and 37550                                            
From: Andrew Doran <ad%netbsd.org@localhost>                                    
          
Date: Fri, 10 Apr 2009 23:55:33 +0000                                           
To: Christos Zoulas <christos%zoulas.com@localhost>                             
          
                                                                                
On Fri, Apr 10, 2009 at 06:00:46PM -0400, Christos Zoulas wrote:                
                                                                                
> I've implemented stat functions for all the devices that called
> fbadop_stat   
> and tried to fill the fields as best as I could. I am concerned about the     
> cost of getnanotime() and the extra fields in the softc struct. Is it
> worth   
> it? We could also simply fix this by implementing fnullop_stat, which is      
> effectively what soo_stat() ends up being for non-unix sockets.               
>                                                                               
> Comments?                                                                     
                                                                                
None on the general approach, but, in the patch the stat fops are missing       
locking. Likewise for the crypto updates, >uintptr_t writes are unlikely to     
be atomic.