Re: Hurd NFS translator

Applied, thanks!

Michael Kelly, le mer. 09 juil. 2025 20:38:58 +0100, a ecrit:
> On 21/06/2025 18:18, Samuel Thibault wrote:
> > Better validate progressively.
> I have attached the first in a series of patches to enable NFSv3 support in
> the NFS translator. These will each include a small number of changes to
> simplify validation. There is no command line argument to activate the NFSv3
> code at present and the patches will not change that until there is
> sufficient functionality in place to complete a build of both gnumach and
> hurd using NFSv3. These are my test cases for ordinary files.
> 
> This first patch is extracted from my local code that is capable of building
> the above on NFSv3. I have tested this patch alone by building gnumach on an
> NFSv2 server and then booting with that image without obvious problem.
> 
> To link gnumach on NFS it is necessary to use my patched binutils ld. I
> spent quite a long time working on 2 bugs that I found which were exposed by
> it not being possible to mmap an NFS mounted file on Hurd. I have a possible
> fix for 1 bug and an understanding of the other, which allows me to
> workaround it, but which might need quite a bit of development to fix
> properly. Is it recommended to liaise with the Debian or the upstream
> maintainer in this case?
> 
> Please indicate whether this patch is acceptable and if approved I will
> begin generating further patches. In total there is around 500 lines of
> contextual diff yet to be presented. I also have work in progress to add
> TCP/IP transport option which would give a performance benefit on NFSv3.
> 
> Patch summary:
> 
> 1) All 64 bit integers defined in the NFSv3 RPC appear to be uint64_t so I
> have modified the xdr_encode_64bit and xdr_decode_64bit interface
> accordingly. These functions are only used in NFSv3 code.
> 
> 2) bug in xdr_encode_64bit: 'bit shift' and 'bitwise and' occur in the wrong
> sequence
> 
> 3) bugs in xdr_decode_fattr: fsid and ino are both 64 bit values in NFSv3
> 
> 4) bugs in process_wcc_stat: sizeof(int) wrongly used with integer pointer
> arithmetic
> 
> 5) bugs in netfs_attempt_write: offset is 64 bit and missing count field