> Hmm, as long as we don't have fds[1] that could die gracefully
> and applications that could handle this...
>
> ... I would call this _impossible_
>
> The thing is, that a fd is valid from sucessful open to
> sucessful close. So everything this fd depends on, is open too.
>
> But we could introduce a new signal (bad, since not POSIX), or
> use a new semantic for an existing one (even worse) to say "Hey,
> this fd is dead now! Any use from now on will cause errors."
I think we need no new signals. All what we need is to revoke
all file handles connected to the device having been unplugged.
> Another simplier problem is, if the ressource is _temporarly_
> unavailable[2].
The question is whether we really want to handle such situation
without userland noticing the exchange or whether the right view
is "the old device ceased to exist, a new one appeared with the
same name, all programs need to re-open it."
(Please note that in the most common cases, i.e. swapping of disks
in a disk array or replacing faulty network cards, the exchange can
be fully transparent even without introducing any additional mechanisms
and without userland programs noticing it.)
Have a nice fortnight
-- Martin `MJ' Mares <mj@ucw.cz> http://atrey.karlin.mff.cuni.cz/~mj/ Faculty of Math and Physics, Charles University, Prague, Czech Rep., Earth "This message transmited on 100% recycled electrons."- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.rutgers.edu Please read the FAQ at http://www.tux.org/lkml/