How to fix |mkfifo()| failure if |pathname| is on NFS ? / was: Re: [EXTERNAL] Re: mkfifo: cannot set permissions of 'x.fifo': Not a directory

[Corinna Vinschen]

On Aug 23 01:05, Roland Mainz via Cygwin wrote:
> Note that Cygwin does not interpret the file |myfifo.fifo| as FIFO,
> instead it comes back as a symlink "myfifo.fifo -> ':\0:c4:1000'".
> 
> AFAIK there are (at least) these two options to fix the problems:
> 1. Check whether the filesystem for the fifos path is NFS
> (cgywin.dll's |fs.fs_is_nfs()|), and if it is a symlink check if it
> starts with ':\0:c4:' (assuming "c4" is the prefix for inodes created
> with |mkfifo()|). If this condition is |true|, then cygwin |stat()|,
> |open()| etc. should treat this inode as FIFO.

The downside is that it is not possible to diffentiate between Cygwin
FIFOs and real FIFOs created from the remote side in `ls -l'
output.  Note that Cygwin returns the NFS stat info verbatim, so
a real FIFO is returned as a real FIFO:

  linux$ mkfifo bar
  cygwin$ ls -l bar
  prw-r--r-- 1 corinna vinschen  0 Aug 25 13:58 bar

The idea was always to use NFS as exchange medium, but not as
installation medium for the entire distro or to keep Cygwin home
dirs on NFS.  There were times where NFS was pretty unstable.
I used NFS for quite some time to build Cygwin packages, but at one
point I got trouble (performance problems with multiple concurrent
processes accessing an NFS share, build errors out of the blue),
so I switched to Samba shares, albeit grudgingly.  I'm not yet
sure if the problems are fixed.  At least a recent OpenSSH build
ran through without problems...

Anyway.  How would you like to make sure that a Cygwin application
can differ between real FIFOs and Cygwin FIFOs?

> 2. Check whether the filesystem for the fifos path is NFS
> (cgywin.dll's |fs.fs_is_nfs()|), and then just refuse |mkfifo()| with
> |ENOSYS| (not implemented)

I like the idea.

> Better algorithm for [1] might be to check whether the inode is a
> symlink, and then do a |fs.fs_is_nfs()| on the symlinks |pathname|,
> assuming this is more performant.

Even better would be if we could just create and use real FIFOs
on NFS(*).  But while NtQueryEaFile can be used to fetch real
NFS file info, and while NtCreateFile can be used to create real
synmlinks via NFS, I don't see an interface resembling mknod/mkfifo.

(*) Not using them in the sense that Cygwin can actually communicate
with the real remote FIFO, but in the sense that the fake is good enough
so that our fhandler_fifo class can actually utilize the file just like
FIFOs on NTFS.


Corinna