[OpenAFS] gcc reading beyond end of file on AFS

[emoy@apple.com]

I downloaded osi_vnodeops.c, which had a few other changes besides the 
memset (corrected number of arguments to osi_FlushPages()).  I did have 
to change #include <afsincludes.h> to #include <afs/afsincludes.h>, but 
maybe it was moved in the latest sources as well.

Anyways, that does seem to fix the problem.

Thanks,

Edward Moy
Apple Computer, Inc.
emoy@apple.com

(This message is from me as a reader of this list, and not a statement
from Apple.)

On Saturday, January 25, 2003, at 10:42  AM, Derrick J Brashear wrote:

> On Sat, 25 Jan 2003, RD Schaffer wrote:
>
>> Hi there,
>>
>>    I cannot explain technically what is going on. But I understand 
>> that
>> this is a bug in open afs, which I found a little while ago. If you
>> need a working version of afs to allow builds, you can use the arla
>> client. (I believe they already had this problem but fixed it.) I do
>> not know the schedule for a bug fix for openafs, but am interested to
>> know as well.
>
> Get RCS version 1.4.2.5 of src/afs/DARWIN/osi_vnodeops.c and drop it in
> your tree, then rebuild the client. Alternately, apply this patch:
> *** openafs/src/afs/DARWIN/osi_vnodeops.c       2002/10/16 03:58:17 
> 1.10
> --- openafs/src/afs/DARWIN/osi_vnodeops.c       2003/01/07 23:03:35 
> 1.11
> ***************
> *** 467,472 ****
> --- 467,477 ----
>       }
>       afs_BozonUnlock(&tvc->pvnLock, tvc);
>       AFS_GUNLOCK();
> +
> +     /* Zero out rest of last page if there wasn't enough data in the 
> file */
> +     if (code == 0 && auio.uio_resid > 0)
> +       memset(aiov.iov_base, 0, auio.uio_resid);
> +
>       kernel_upl_unmap(kernel_map, pl);
>       if (!nocommit) {
>         if (code)