arla, an AFS implementation

[Magnus Ahltorp]

> > This is perfectly normal. I do a lot of magic with the inodes' device
> > numbers. I set the device numbers of the inode to the device numbers
> > of the cached inode to be able to do bmap directly.
> 
> According to POSIX, a programmer is allowed to assume that if stat
> returns device and/or inode numbers that are different then one may
> assume that they originate from different files. If device and inode
> numbers are changing this will confuse code seriously, the program
> will be fooled to deduce that the file must have been replaced under
> its feet.

I made a design choice here. I can support that the device numbers are
in some respect "correct", but the cost of this is ugly code that is
unmaintainable. If anyone is willing to change the code and/or the
linux kernel without uglifying the code, fine with me.

/Magnus
map at stacken.kth.se