samba - Jul 2025 - SMB3 Unix Extensions - creating special files

Thanks for spotting this. I can confirm that I see different behaviour 
with different kernels:

6.13.0 - mkfifo and ln-s work as expected.
6.14.0 - mkfifo works, ln-s gives 'operation not supported'.
 >=6.15.7 - both give 'operation not supported'.

Which implies possibly more than one regression?

Thanks,

Matthew


On 31/07/2025 18:37, Paulo Alcantara wrote:
> [You don't often get email from pc at manguebit.org. Learn why this is
important at https://aka.ms/LearnAboutSenderIdentification ]
> 
> Ralph Boehme <slow at samba.org> writes:
> 
>> ...adding linux-cifs and Steve to the loop....
>>
>> Looks to be a client issue: the client is checking for existence of the
>> targets, the server returns ENOENT and then that's it. There no
attempt
>> to create either a symlink nor the fifo as reparse points.
>>
>> @Steve: any idea of what could be going wrong? Iirc this is supposed to
>> be working in the client.
> 
> With Linux v6.16 and samba master (f1a828016921):
> 
> root at fed:~# mount.cifs //192.168.124.1/test /mnt/1 -o
username=testuser,password=foo-123,unix
> root at fed:~# mount -t cifs
> //192.168.124.1/test on /mnt/1 type cifs
(rw,relatime,vers=3.1.1,cache=strict,upcall_target=app,username=testuser,uid=0,noforceuid,gid=0,noforcegid,addr=192.168.124.1,file_mode=0755,dir_mode=0755,soft,posix,posixpaths,serverino,reparse=nfs,nativesocket,symlink=unix,rsize=4194304,wsize=4194304,bsize=1048576,retrans=1,echo_interval=60,actimeo=1,closetimeo=1)
> root at fed:~# (cd /mnt/1; rm -rf *; mknod chr c 2 1; mknod blk b 3 4;
mknod fifo p; ln -s f0 l0; python -c "import socket as s; sock =
s.socket(s.AF_UNIX); sock.bind('sock')"; ls -lh)
> ln: failed to create symbolic link 'l0': Operation not supported
> total 0
> brwxrwxrwx 1 root fsgqa 3, 4 Jul 31 14:31 blk
> crwxrwxrwx 1 root fsgqa 2, 1 Jul 31 14:31 chr
> prwxrwxrwx 1 root fsgqa    0 Jul 31 14:31 fifo
> -rwxrwxrwx 1 root fsgqa    0 Jul 31 14:31 sock
> 
> I see a regression when attempting to create symlinks and sockets.  Note
> the 'nativesocket' and 'symlink=unix' options, which are
definitely
> wrong for SMB3.1.1 POSIX mounts.  It should have 'symlink=native'
and
> 'nonativesocket'.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: OpenPGP_signature.asc
Type: application/pgp-signature
Size: 203 bytes
Desc: OpenPGP digital signature
URL:
<http://lists.samba.org/pipermail/samba/attachments/20250731/cfa3a928/OpenPGP_signature.sig>

Matthew Richardson <m.richardson at ed.ac.uk> writes:
> Thanks for spotting this. I can confirm that I see different behaviour 
> with different kernels:
>
> 6.13.0 - mkfifo and ln-s work as expected.
> 6.14.0 - mkfifo works, ln-s gives 'operation not supported'.
>  >=6.15.7 - both give 'operation not supported'.
>
> Which implies possibly more than one regression?
Yes.  It used to work on older kernels because the client used to create
special files with NFS reparse points by default, which is required for
SMB3.1.1 POSIX mounts.  6c06be908ca1 ("cifs: Check if server supports
reparse points before using them") then added a check for
FILE_SUPPORTS_REPARSE_POINTS, which breaks against samba because it
isn't set.  IOW, we should skip this check for SMB3.1.1 POSIX mounts.