get_unused_fd()

[Magnus Ahltorp]

> 	Umm... Say it, I'ld like to see reasons for that and they'ld
> better be very serious. Anyway, it's up to Linus to make any decisions on
> that stuff. Exporting something makes it much more permanent. If symbol is
> available only from the kernel - well, one has to consider kernel alone
> when/if he'll want to change it. If it is exported... any change may break
> 3rd-party modules. Any additional exported symbol to == potential boat
> anchor.

The policy seems to be "if we have a module in the _mainstream_ kernel
that uses a symbol, let's export it", and not "if there is a need for
a symbol, let's export it".

> 	Wait a bit. If you have dentry and want to fit a struct file atop
> of it - you don't need either get_unused_fd() or get_empty_filp(). There
> is open_dentry() (fs/exec.c; exported). It looks like cleaner solution.
> OTOH you didn't describe your problem, so...

I looked at that function, and it solves the _current_ problem I have,
yes. But what is the real problem of modules allocating space in the
file descriptor table? And, how am I to know that it is boat anchorish
to use a plain simple space allocation function, and it is not
anchorish to use a function that only exists to serve different
binfmts?

Would open_dentry() be exported if binfmts didn't exist as modules? I
doubt so. There isn't even a comment above open_dentry that says "use
this function if...", and that _really_ leads to boat anchorish
behaviour. Documentation, not restriction, is needed.

/Magnus
map at stacken.kth.se